From 33addc3f94bcd28a92a02c3c96b39c2efb083bb7 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 08:46:22 +0000 Subject: [PATCH 01/19] release 0.3.22 section 22: the morning (main's word, the founder's rule, the fifth cut 5b673577, the move staged, the fleet's reading, the Arc answer) Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index ea0107ca5..91a693d7b 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -198,3 +198,13 @@ Main's word on route (A) or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 B **The knob:** the hash lane's 74585c91 (main's order of 7 October: the Ember clock ladder continues below 45 percent in 100 MHz steps to a 20 percent floor; the search stops at the knee, the first row more than the tolerance under the cap point's rate, or on a faulted row, the fingerprint check; the best MH per watt within tolerance is the point; lock_result and the card's lock_* fields for the UI; tests known-failed first on a fake helper) sat on the mirror's master, not on the release line; cherry-picked onto release-0.3.24 as e181f497 with ember.rs resolved as the union (the knob's floor and fine step beside the efficient-point ceiling of 7 October: EFFICIENT_W, DEFAULT_CAP_PCT, power_ceiling), the plan-count test updated to the knob's ladder on the 5090 (1 + 6 + 7; b6e2845f); app gate GREEN on build-1 (294 + 35 + 8), pre-push 60; the DMG re-cut under the lock on b6e2845f; the UI lane's drawing of the lock fields asked onto that tip. The rule: the knob never sets a lock below the knee without the user's own choice. **The measured Ember line (the Counter lane, read on PC 1 overnight):** the installed app's stock power-limit climb on the RTX 5080 lands at 60.3 MH/s at 123 W (0.489 MH/W, clock_cap 2,936; run-ca3-pc1-ember-5080-20261007 at 07:05 BST, the app's own tune complete before the script's cast fault), while the clock-lock grid on the same card gives 71.1 MH/s at 103.7 W at the 1,000 MHz lock (0.686 MH/W) and 71.2 at 146.6 W at 1,100 on class v4 (0.486) (run-ca3-pc1-v4-eff-5080-20261007-d at 02:54 BST), so the core-clock lock is worth about 40 percent more per watt and 18 percent more rate than the climb alone on the 5080; on the 5090 the knob's reference rows are class v4 at 1,200 MHz (133.8 MH/s at 305 W, 0.439) and class v3 at 1,300 (134.6 at 223 W, 0.603), the knee at 1,300 on both, measured four times (docs/bench-log.md, the 7 to 8 October entry). The 9070 XT tune row follows. Nothing else changes in the cut: the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. + +## 22. The morning: main's word, the fifth cut, the move staged (09:3x to 09:5x BST, 8 October) + +**The night's silence** was the harness: main's command failed at 23:36 BST and it got no turn until 09:3x; the build-server lane and the fleet lane sat on backgrounded chains from 22:54 and 22:58 BST with nothing lost on the boxes (both idle); all three answered at 09:31 to 09:34. **Main's word (09:3x BST):** route (A) by the shipper (the node lane re-cuts the floor from the minute 10:45 BST; the shipper applies the pack-gate line with the fleet's tooling, places the signed move file at at_epoch 0 the moment the pin is green, names the minute, restarts build-1's three nodes and publishes the Mac entry at the minute; a pin not green by 10:35 slides the minute to the pin plus 25; the ceiling 12:50); the build-server lane builds the seed, Windows and hive pairs on the morning pin under lease class release and hands them over; the Windows app chain is the shipper's after the move (0.3.23 take 3 skipped as moot); "PC 2 clear" for the hash lane's Arc job; the two lanes respawned (the new fleet lane takes the readings after the move, not the move). Then (09:5x): the project lead's word, get it live now: the minute is the last FETCHED plus ten, 10:45 the ceiling not the target; everything but the crossing done by 11:00 BST, the Windows chain in parallel. **Standing authority from main:** if the reachable boxes have not all FETCHED by 10:35 the shipper says "slide" itself; when the margin to a floor's publish ceiling falls under 15 minutes, re-cut without asking. **The founder's rule ("this cannot happen again"):** every ask to main carries a default action and a deadline, silence at the deadline means the default, never a stand-down; the Counter lane mirrors every clock the shipper holds today from the shipper's runbook (scratch r0324/RUNBOOK-0324-move.md, twelve steps with commands, hosts, keys and read-backs). + +**The fifth cut, the pin: release-0.3.24-node = 5b673577** (dfbd1e10 with program_class_v5_activation_daa 68,400, epoch 19, nothing else; cut at 09:33:18 BST from build-1's seed at DAA 56,329, the chain at 1.0 DAA/s all night; the publish DAA at 10:45 about 60,630, the floor about 12:54 BST, the three heights under it; the floor holds for a publish up to DAA 61,200, about 10:54 BST); every gate green at 09:39:15 BST (build at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary with digest cc9026909eddbadb and the mixed-version refusal against a 2720d8d2 node both ways; the testnet canary on b2e856ed); the node-lane pair igneumd a3b1a2c9 / igneum-miner cfa9f5ca under /srv/artefacts/0324-5b673577/node-lane, igneum-pow-v5 8 paths; the fast-time SUMMARY on it due about 09:55. The app side: release-0.3.24 = 0c47b59a (the knob's display knob-24 2c4dc617 merged, UI 88; the Windows pin to 5b673577 at ebbdbc32); the Mac node pair (igneumd aee81264, igneum-miner 129275ef) and the DMG 7e6e3eb3 built under the lock, the Mac entry re-staged in both folders. + +**The move staged (09:42 to 09:44 BST):** the tarball fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B) served from build-1; the move file m5b67-1 (commit 5b673577, want_version igneumd/2.1.0-5b673577, want_digest cc9026909eddbadb, both pair slots on the tarball, at_epoch 0) signed with the fleet key and placed at 09:42:14 BST, the served file's signature verified; the pack-gate line (cfa9f5ca into PAIR_MINER_SHA16 in env-last and box-dn3.sh's default) applied on 32 of 32 reachable boxes at 09:43:34 with each gate read back (unreachable dn3-relay and p2-4090-1b behind dead Vast proxies; dn3-pool-a destroyed by the fleet's hourly waste pass at 09:02 BST, "exited" on Vast, so the open pool is dn3-pool-b alone until a second member is rented); build-1's three nodes (the seed, node1-dn3 down since the night, the observer: a plain process, not a unit) restart by the shipper's script at the minute, dry-run clean. The fleet's morning reading: the eight boxes it read as "bc5945fe" had NO Devnet 3 node (its reader hashed its own bash); dn3-g2's node died 23:49 BST, dn3-g1's 01:46, the others overnight, no shutdown lines (the death evidence being read); all eight restarted on 2720d8d2 before the minute so they fetch and move with the rest. The eleven prover boxes read "pair ok" at the tip all night; hub-1 at DAA 56,281 with 2 peers at 09:34. + +**The Arc B580 (09:36 to 09:5x BST):** the class v5 kit worker (27faa253) failed its self-test on PC 2's Arc on both the v5 pack and the v4 control (96 of 96 lanes bad; the cache and dataset FNVs right); the Intel lane answered from the app's own log: the installed 0.3.21 worker's self-test PASSED on the Arc with the devnet pack (96 of 96 at 21:23 BST, then 54 blocks accepted with the CPU re-check at 10.58 MH/s), so the /miners row stands as a measurement; the kit worker regressed because class-v5 1095eaa8 (and master) lack proto-opencl/intel_rotr.h (26e135a3), which release-0.3.23 and 0.3.24 carry; the Intel lane lands it on the mirror's master; the Intel kit holds out of 0.3.24 (main's rule), the crossing time 09:36 BST on the page row. From 9b057c02ed164d174b44b0f3c67805ea87fc79d4 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 08:49:41 +0000 Subject: [PATCH 02/19] release 0.3.22: THE MINUTE 10:05 BST set in m5b67-1, the pairs and the hive on the pin, what moves at the minute Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index 91a693d7b..1721d2982 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -208,3 +208,5 @@ Main's word on route (A) or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 B **The move staged (09:42 to 09:44 BST):** the tarball fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B) served from build-1; the move file m5b67-1 (commit 5b673577, want_version igneumd/2.1.0-5b673577, want_digest cc9026909eddbadb, both pair slots on the tarball, at_epoch 0) signed with the fleet key and placed at 09:42:14 BST, the served file's signature verified; the pack-gate line (cfa9f5ca into PAIR_MINER_SHA16 in env-last and box-dn3.sh's default) applied on 32 of 32 reachable boxes at 09:43:34 with each gate read back (unreachable dn3-relay and p2-4090-1b behind dead Vast proxies; dn3-pool-a destroyed by the fleet's hourly waste pass at 09:02 BST, "exited" on Vast, so the open pool is dn3-pool-b alone until a second member is rented); build-1's three nodes (the seed, node1-dn3 down since the night, the observer: a plain process, not a unit) restart by the shipper's script at the minute, dry-run clean. The fleet's morning reading: the eight boxes it read as "bc5945fe" had NO Devnet 3 node (its reader hashed its own bash); dn3-g2's node died 23:49 BST, dn3-g1's 01:46, the others overnight, no shutdown lines (the death evidence being read); all eight restarted on 2720d8d2 before the minute so they fetch and move with the rest. The eleven prover boxes read "pair ok" at the tip all night; hub-1 at DAA 56,281 with 2 peers at 09:34. **The Arc B580 (09:36 to 09:5x BST):** the class v5 kit worker (27faa253) failed its self-test on PC 2's Arc on both the v5 pack and the v4 control (96 of 96 lanes bad; the cache and dataset FNVs right); the Intel lane answered from the app's own log: the installed 0.3.21 worker's self-test PASSED on the Arc with the devnet pack (96 of 96 at 21:23 BST, then 54 blocks accepted with the CPU re-check at 10.58 MH/s), so the /miners row stands as a measurement; the kit worker regressed because class-v5 1095eaa8 (and master) lack proto-opencl/intel_rotr.h (26e135a3), which release-0.3.23 and 0.3.24 carry; the Intel lane lands it on the mirror's master; the Intel kit holds out of 0.3.24 (main's rule), the crossing time 09:36 BST on the page row. + +**THE MINUTE: 10:05:00 BST on 8 October (09:05:00Z),** set in the signed move file m5b67-1 at 09:48:12 BST and read back served (commit 5b673577, want_digest cc9026909eddbadb, at_utc 2026-10-08T09:05:00Z, the signature good), 49 minutes inside the 10:54 ceiling, after the fast-time SUMMARY PASS on 5b673577 at 09:47:45 BST (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8, the stale node refused, 0 rejections, the restart across the boundary resynced in 12.1 s, four sinks equal) and FETCHED 35 of 39 at 09:46 (dn3-relay and p2-4090-1b behind dead Vast proxies, p2-3090-1 refusing ssh since last night, dn3-agg48 the new 48 GB aggregator prover renting: the first three fall off onto ba75bf6f and rejoin by the pull, the fourth fetches with its bring-up; named in the file's note); dn3-g1 and g2 back on 2720d8d2 at 09:43 and 09:44 with their miners on and no panic or OOM on either, the other six dead nodes restarting in the same pass. **The pairs on the pin (the build-server lane, 09:43 to 09:47 BST):** seed igneumd 3a204fd9 / igneum-miner 464dca07 (glibc 2.34, the string twice), win igneumd.exe 0b144d7d / igneum-miner.exe 0cc68d9e, the hive package igneum-hive-0.3.24-5b673577.tar.gz 025bf01f (29,876,121 B, the three kit zips, smoked in ubuntu:20.04), all under /srv/artefacts/0324-5b673577/ beside the node lane's hands pair a3b1a2c9 / cfa9f5ca; the engine strings read igneum-pow/src/ (the worktree's path) where the node lane's read igneum-pow-v5/src/, the same code at 1c420786, rule 6a's pattern taking both. At the minute: the pullers restart node and miner from the pair; build-1's seed, node1-dn3 and observer by the shipper's script (dry-run clean; the observer a plain process, restarted with its own command line); the Mac entry (DMG 7e6e3eb3: app 0c47b59a with the knob and its display, the Mac node pair aee81264/129275ef, interface 1.0.2, the floor file kept) and the hive 025bf01f into both token folders with the public Mac and HiveOS aliases, one deploy; "PC 2 go" for the Windows chain (installer-kit-0c47b59a.zip b03fba28 cut, the app cross running, the PC 1 host job publishing). The 0.3.25 line opened by the update-return lane: amd-clock-25 a002732a (the AMD core-clock knob through ADLX manual tuning in the telemetry tool; the exe must be rebuilt for the kit), its default at 13:00 BST. From 8b0a5a8a403a0c63a443ed0a0d6f8a14865ca570 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 09:11:06 +0000 Subject: [PATCH 03/19] release rules 11 extended (never a kill pattern the shell carries; a kept datadir's locks); release 0.3.22 section 23: the move executed, 0.3.24 Mac and HiveOS LIVE 10:08:35 BST, the Windows chain on PC 2 Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 4 ++++ docs/plans/release-rules.md | 2 +- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index 1721d2982..c5c28cc7d 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -210,3 +210,7 @@ Main's word on route (A) or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 B **The Arc B580 (09:36 to 09:5x BST):** the class v5 kit worker (27faa253) failed its self-test on PC 2's Arc on both the v5 pack and the v4 control (96 of 96 lanes bad; the cache and dataset FNVs right); the Intel lane answered from the app's own log: the installed 0.3.21 worker's self-test PASSED on the Arc with the devnet pack (96 of 96 at 21:23 BST, then 54 blocks accepted with the CPU re-check at 10.58 MH/s), so the /miners row stands as a measurement; the kit worker regressed because class-v5 1095eaa8 (and master) lack proto-opencl/intel_rotr.h (26e135a3), which release-0.3.23 and 0.3.24 carry; the Intel lane lands it on the mirror's master; the Intel kit holds out of 0.3.24 (main's rule), the crossing time 09:36 BST on the page row. **THE MINUTE: 10:05:00 BST on 8 October (09:05:00Z),** set in the signed move file m5b67-1 at 09:48:12 BST and read back served (commit 5b673577, want_digest cc9026909eddbadb, at_utc 2026-10-08T09:05:00Z, the signature good), 49 minutes inside the 10:54 ceiling, after the fast-time SUMMARY PASS on 5b673577 at 09:47:45 BST (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8, the stale node refused, 0 rejections, the restart across the boundary resynced in 12.1 s, four sinks equal) and FETCHED 35 of 39 at 09:46 (dn3-relay and p2-4090-1b behind dead Vast proxies, p2-3090-1 refusing ssh since last night, dn3-agg48 the new 48 GB aggregator prover renting: the first three fall off onto ba75bf6f and rejoin by the pull, the fourth fetches with its bring-up; named in the file's note); dn3-g1 and g2 back on 2720d8d2 at 09:43 and 09:44 with their miners on and no panic or OOM on either, the other six dead nodes restarting in the same pass. **The pairs on the pin (the build-server lane, 09:43 to 09:47 BST):** seed igneumd 3a204fd9 / igneum-miner 464dca07 (glibc 2.34, the string twice), win igneumd.exe 0b144d7d / igneum-miner.exe 0cc68d9e, the hive package igneum-hive-0.3.24-5b673577.tar.gz 025bf01f (29,876,121 B, the three kit zips, smoked in ubuntu:20.04), all under /srv/artefacts/0324-5b673577/ beside the node lane's hands pair a3b1a2c9 / cfa9f5ca; the engine strings read igneum-pow/src/ (the worktree's path) where the node lane's read igneum-pow-v5/src/, the same code at 1c420786, rule 6a's pattern taking both. At the minute: the pullers restart node and miner from the pair; build-1's seed, node1-dn3 and observer by the shipper's script (dry-run clean; the observer a plain process, restarted with its own command line); the Mac entry (DMG 7e6e3eb3: app 0c47b59a with the knob and its display, the Mac node pair aee81264/129275ef, interface 1.0.2, the floor file kept) and the hive 025bf01f into both token folders with the public Mac and HiveOS aliases, one deploy; "PC 2 go" for the Windows chain (installer-kit-0c47b59a.zip b03fba28 cut, the app cross running, the PC 1 host job publishing). The 0.3.25 line opened by the update-return lane: amd-clock-25 a002732a (the AMD core-clock knob through ADLX manual tuning in the telemetry tool; the exe must be rebuilt for the kit), its default at 13:00 BST. + +## 23. The move executed: 0.3.24 Mac and HiveOS LIVE (10:05 to 10:09 BST, 8 October) + +**THE MINUTE 10:05:00 BST** fired on the signed file m5b67-1 (FETCHED 36 of 39 at 10:00: dn3-agg48 the 48 GB aggregator prover still renting, p2-3090-1 refusing ssh since the night, p2-4090-1b behind the dead proxy ssh2; all eight nodes dead overnight back on 2720d8d2 and fetched by 09:53, dn3-r03 and r04 given a saved environment by hand). **build-1's three on the pin:** node1-dn3 (26671, JSON 28670) and the observer (26651, JSON 28650) up at 10:08 on "igneumd/2.1.0-5b673577", digest cc902690, object version 6, the N15 line ("the 15611 records up to the tip 15611 are continuous"); the seed (26631, JSON 27632) at 10:09 after a first start at 10:08 panicked on the old process's RocksDB lock (meta/LOCK "Resource temporarily unavailable"), up on cc902690 with 3 outgoing connections. The shipper's fault, two parts: the three-node script's --go branch did not execute at 10:05 (the dry-run text printed and it exited: the kill pattern `pgrep -f appdir=...` matched the ssh shell carrying the script, the fleet's fault class of 22:34 the night before), so the move ran by hand at 10:07 with the victims found by process name and cmdline; and the seed restarted before the old process had released its datadir (rule 11 extended). **The 0.3.24 Mac entry LIVE at 10:08:35 BST** (the copy at 10:05:01, one Vercel deploy), both token folders, channel devnet-3: Igneum-Miner-0.3.24-5b673577.dmg 7e6e3eb3 (45,652,380 B; app 0c47b59a: the core-clock knob and its display, the Power Helper unattended, the dashboard caps, the pool daemon's reconnect, the testnet re-cut; the Mac node pair aee81264/129275ef from 5b673577), interface 1.0.2, the floor file kept for the apps below 0.3.22, read back from both folders and the public Mac alias; **the HiveOS package igneum-hive-0.3.24.tar.gz 025bf01f** (the 5b673577 hive pair, the two sub-version 3 kits and the v5 kit e6c088bb) in both folders and on the public alias at the same deploy; dl/public/igneum-downloads.json names 0.3.24 for Mac and HiveOS. The notes carry the knob, the floor from the minute and "update before the move minute or the node stops following Devnet 3". Owed from the fleet: the APPLIED count per box, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07 (the installer job from the a4c5a855 kit, sha ad7a50e8, and the payload igneum-windows-app-0.3.24-a4c5a855-msvc.zip 7f12cbe3 with the PC 1 host 0e241c94 built at 09:50 in 7 s, the 5b673577 Windows pair 0b144d7d/0cc68d9e, the 0c47b59a app exes ce7f82a6/99edc6aa/d035368e; host.sha256 at a4c5a855), the rule 14 smoke as the gate, then the Windows entry into both folders with the public alias and the card ("Mac, HiveOS and Windows now"); the hash lane's Arc re-read on the rotate-fold kit (packs-ca3-v5-20261008T085619Z.zip 65b47211) and the update-return lane's PC 1 re-probe and PC 2 S4U proof after the smoke. The crossing at DAA 68,400 about 12:54 BST (the node lane's watch on build-1's seed). diff --git a/docs/plans/release-rules.md b/docs/plans/release-rules.md index 9f9282b2d..75ec9ad16 100644 --- a/docs/plans/release-rules.md +++ b/docs/plans/release-rules.md @@ -15,7 +15,7 @@ Every cut of the Igneum Miner app and its node runs under these. The dated plan 8. **glibc classes:** HiveOS 2.31 (`--ship hive`, smoke in ubuntu:20.04 on the box), seeds and generic 2.35 (`--ship seed`), fleet 24.04 boxes native 2.39. 9. **The Mac builds only the macOS binaries and the DMG,** one at a time under the build lock; every other build, suite and the Windows cross-build runs on the box or a PC; the app gate is `build-remote.sh -- test --release` from the crate dir. 10. **A pin is green on its own suites and gates.** Lines taken on one binary carry to another only when the code is byte-identical, stated in the tip. No known-red pins: a stale test takes a test-only commit on top. -11. **Kill by pid, never by name,** on the shared Mac; a merge worktree never checks out master. +11. **Kill by pid, never by name,** on the shared Mac; a merge worktree never checks out master. Extended 8 October 2026 (twice in one night: the fleet's rescue script at 22:34 BST and the shipper's three-node move at 10:07): a kill or a pgrep by a pattern that the calling shell's own command line carries matches that shell and kills the script before its work; find the victims by process name (pgrep -x igneumd) and the appdir in /proc//cmdline, never by `pgrep -f `; and a node restarted onto a kept datadir waits for the old process to release its RocksDB locks (the seed's first start panicked on meta/LOCK at 10:08 BST; the retry at 10:09 was clean). 12. **The Discord card only when every platform is live.** Live manifest changes beyond the binaries (a moved consensus floor) go out only on the founder's explicit word, staged beside the release with their digest and a one-line diff. 13. **Ship on green:** no calendar waits; when the gates are green, publish and state the clock time (UK). Checkpoints are for slips, not for waiting. 15. **The version bump is the release branch's first commit (7 October 2026, after the 0.3.23 miss).** When a release-0.3.N branch opens, its first commit moves the six version places (app/igneum-app/Cargo.toml and Cargo.lock, app/windows/version.h, app/igneum-app/resources/igneum-app.rc's four fields, packaging/mac/app/Info.plist, the installer's AppVersion), never left to the cut: release-0.3.23 opened at 4cdcab31 and carried 0.3.22 in every place until 21:26 BST, so the app exes and the window host crossed from its first closed tip read 0.3.22 and were void. Gate check: on a push to release-0.3.N the pre-push gate (tools/ci/release-version-check.sh, self-test on tonight's shape first) reads all six places against the branch name and goes red on any mismatch; the .rc was the second layer of the same miss (the box cross failed in build.rs at 21:30 BST). From f97ed89621a61a9cc16c8112b064d721e14f84fa Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 09:57:36 +0000 Subject: [PATCH 04/19] Bench table: the model sweep's rows 16 to 19 (RTX A6000 53.64 MH/s at 239.1 W in the datacentre table, v5 level with v4 on an uncapped sm_86; GTX 1080 Ti 16.26 at 124.3 W with the one-in-three cache self-test flag until a second host decides; RTX 4060 19.09 with no watts, the host has no power sensor; RTX 2060 14.13 at 120.6 W, the floor of the list; rented and untuned, 8 October 2026) Co-Authored-By: Claude Fable 5.1 --- site/miner-bench.json | 88 +++++++++++++++++++++++++++++++++++++++++++ site/miners.html | 10 ++--- 2 files changed, 93 insertions(+), 5 deletions(-) diff --git a/site/miner-bench.json b/site/miner-bench.json index a2b002777..83b41a087 100644 --- a/site/miner-bench.json +++ b/site/miner-bench.json @@ -1109,6 +1109,94 @@ "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA RTX A6000 (48 GB)", + "group": "datacentre", + "mh_s": 53.64, + "watts": 239.1, + "mh_per_w": 0.224, + "driver_os": "NVIDIA driver 595.84, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 16 (a rented card, result cb2-rtx-a6000-48-gb-9600-result.json; 09:46Z)", + "note": "rented single-card host, 0.15 h; power.draw mean 239.1 W (max 251.0, limit 300 W), SM 1,913 MHz mean, memory 7,601 MHz, 65 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (53.82 MH/s, fingerprint matched), so on an uncapped sm_86 card v5 is level with v4 and the A4000's v5 reading narrows to its power cap; read ceiling 53.8 MH/s, the hash at 1.00 of it. Repeats the 7 October row (53.66 then, 53.64 now) at 43 W less on this host; a 3090-class rate at a 3090-class draw", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented; Pascal has no clock lock on any host)", + "card": "NVIDIA GTX 1080 Ti (11 GB)", + "group": "buy", + "mh_s": 16.26, + "watts": 124.3, + "mh_per_w": 0.131, + "driver_os": "NVIDIA driver 580.173.02, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 17 (a rented card, result cb2-gtx-1080-ti-11-gb-0377-result.json; 09:49Z)", + "note": "FLAG: one of three class v4 builds on this host failed its cache self-test (the 256 MiB cache's FNV read wrong on that build while the other two and the class v5 kit build were clean: the host's VRAM or a Pascal-only build race); a second host decides, and until then this row carries the flag. The rate is the mean of the two clean runs (16.261, 16.260). Rented single-card host, 0.10 h; power.draw mean 124.3 W (max 125.5, limit 280 W; no power average on Pascal), SM 1,936 MHz mean, memory 5,005 MHz, 59 C; self-test PASS on the clean builds (96 of 96 vector lanes); the class v5 kit holds the rate (16.27 MH/s, fingerprint matched); read ceiling 16.3 MH/s, the hash at 1.00 of it. The first Pascal row: a 2017 card that runs the class at its own ceiling", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card; Pascal has no clock lock; no measured tune point)" + } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA RTX 4060 (8 GB)", + "group": "buy", + "mh_s": 19.09, + "watts": null, + "mh_per_w": null, + "driver_os": "NVIDIA driver 595.91.07, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 18 (a rented card, result cb2-rtx-4060-8-gb-0672-result.json; 09:53Z)", + "note": "watts not read: this host exposes no power sensor (every power field absent on all 244 samples, as the 7 October 4060 host); a third host is being rented for the watts, and the card's draw is about 60 to 70 W by its 115 W limit and the 4060 Ti's 79 W, an estimate until then. Rented single-card host, 0.12 h; SM 2,732 MHz mean, memory 8,251 MHz, 57 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (19.11 MH/s, fingerprint matched); read ceiling 30.2 MH/s, the hash at 0.63 of it (the 128-bit bus: the independent-read line binds, as on the 2070 Super). Repeats the 7 October row (19.09 then, 19.10 now)", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA RTX 2060 (6 GB)", + "group": "buy", + "mh_s": 14.13, + "watts": 120.6, + "mh_per_w": 0.117, + "driver_os": "NVIDIA driver 580.159.03, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 19 (a rented card, result cb2-rtx-2060-0673-result.json; 09:55Z)", + "note": "rented single-card host, 0.15 h; power.draw mean 120.6 W (max 124.6, limit 160 W; no power average on Turing), SM 1,948 MHz mean, memory 6,801 MHz, 69 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (14.14 MH/s, fingerprint matched); read ceiling 16.1 MH/s, the hash at 0.88 of it. The floor of the list and the smallest card on it: the class v4 set fits with 4.6 GB spare; the same Turing per-watt floor as the 2070 Super and the 2080 Ti, so every Turing card earns the same per watt and only the absolute rate differs", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } } ] } diff --git a/site/miners.html b/site/miners.html index be1d30221..8abcfc7f6 100644 --- a/site/miners.html +++ b/site/miners.html @@ -213,7 +213,7 @@ table{min-width:560px}
Igneum/GPU bench table
-
33 cards you can buy, 11 datacentre, 0 fleet tuning models
+
36 cards you can buy, 12 datacentre, 0 fleet tuning models

GPU bench table

Measured hash rates per card on the Igneum lottery hash, with the generator version, the miner version, the date and the log entry behind each number.

@@ -265,10 +265,10 @@ table.bench2 tr.detail .d{display:block;margin:0 0 4px}table.bench2 tr.detail b{

Best desktop card: NVIDIA RTX 5090, 136.1 MH/s (measured, 2026-10-06). Best stock MH per wall watt: NVIDIA RTX 5070 Ti, 0.577 at 135.9 W, untuned (measured by the fleet, 2026-10-08). Best tuned: NVIDIA RTX 5090, 0.563 at 226.8 W (measured, 2026-10-06).

One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. MH per wall watt uses board or wall power; a row whose watts are the chip's (Apple silicon: GPU plus DRAM from IOReport) says so and is not ranked on that column. Integrated GPUs are not listed.

Cards you can buy

-
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.50.577135.9 WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.40.539145.6 WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-07620.248249.5 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-07590.221267.3 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.90.201293.5 WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-07520.506102.8 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.80.468110.8 WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.20.289142.6 WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.70.317128.6 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-07390.219178.3 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-08390.22177.1 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.10.261142.3 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.10.256129.5 WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.70.316100.4 WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.30.41575.4 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.30.291107.3 WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.20.312100.2 WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.90.241111.6 WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-08260.116224.6 WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.10.25977.5 WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.10.25379.2 WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.10.115166 WStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock
-

33 cards. Details opens a card's Hive flight-sheet values, its full class v4 cost, miner, source and note.

-
Datacentre (11 cards, rented for the measurement; about three times the rented dollars per hash of a desktop card) -
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.40.487855.6 WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-073130.723432.9 WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.70.645385.6 WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.30.605411.7 WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-071550.517299.6 WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.40.52266.3 WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.10.515268.6 WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.50.452288.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.40.234240.7 WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.40.279112.8 WStock
+
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.50.577135.9 WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.40.539145.6 WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-07620.248249.5 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-07590.221267.3 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.90.201293.5 WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-07520.506102.8 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.80.468110.8 WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.20.289142.6 WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.70.317128.6 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-07390.219178.3 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-08390.22177.1 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.10.261142.3 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.10.256129.5 WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.70.316100.4 WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.30.41575.4 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.30.291107.3 WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.20.312100.2 WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.90.241111.6 WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-08260.116224.6 WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.10.25977.5 WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.10.25379.2 WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.10.115166 WStock
NVIDIA RTX 4060 (8 GB)class v4 not measured · 2026-10-0819.1not measurednot readStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
NVIDIA GTX 1080 Ti (11 GB)class v4 not measured · 2026-10-0816.30.131124.3 WStock
NVIDIA RTX 2060 (6 GB)class v4 not measured · 2026-10-0814.10.117120.6 WStock
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock
+

36 cards. Details opens a card's Hive flight-sheet values, its full class v4 cost, miner, source and note.

+
Datacentre (12 cards, rented for the measurement; about three times the rented dollars per hash of a desktop card) +
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.40.487855.6 WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-073130.723432.9 WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.70.645385.6 WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.30.605411.7 WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-071550.517299.6 WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.40.52266.3 WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.10.515268.6 WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.50.452288.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.40.234240.7 WStock
NVIDIA RTX A6000 (48 GB)class v4 not measured · 2026-10-0853.60.224239.1 WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.40.279112.8 WStock
Earlier classes (6 rows: the genesis program, the hourly program, class v3 before the shadow; not comparable with the table above)

The bench numbers of 3 and 4 October 2026. A higher MH/s here is a different hash, not a faster card.

From fe940213a63baf4b3c358d215578a36d6bccec0d Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:01:24 +0000 Subject: [PATCH 05/19] Counter ASIC 3.0 status: the 0.3.25 pairs and app tip (78dfe65e on the ceiling cut e0644958, its fast-time PASS at 10:54), the tiers' class-flip UI half (d949e274), the four builder lanes and the class v6 declaration with its four research lanes spawned and the outline on the mirror, the Arc re-read and v5lock jobs live, the held-push cause (the branch's own hook predating the per-process fix) Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index a803d9124..51088a13b 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -465,7 +465,7 @@ Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of | 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | | 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | -The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. THE 0.3.25 PAIRS ON build-1 (the build-server lane, /srv/artefacts/0325-7bd2940f/): the seed pair at 10:34:44 BST (igneumd c7fc542b, igneum-miner 4494ecc4, glibc 2.34), the Windows pair at 10:36:13 (igneumd.exe 5d1dea23, igneum-miner.exe eee7bdfa), the hive package igneum-hive-0.3.25-7bd2940f.tar.gz at 10:37:49 (sha d977797f..., the three kit zips, smoked in ubuntu:20.04); the kit re-cut from 92f004f1 (sha 676240f6, 424,540 B) staged in both folders, the PC 1 host from it bc8d4f79 (in host.sha256 at the shipper's 24680e1d), the 0.3.25 Windows payload from 24680e1d cutting. Every pair of the cut exists by 10:38; the pin's gate set and the crossing are the only waits. FOUR NEW LANES ON THE FOUNDER'S ORDER (11:00 BST, "build all this today to close this gap"), mirrored by the coordinator as the shipper's clocks are: the explorer (a5ef1d5801084005b; explorer.igneum.network by 16:00), the canonical DEX and the Sepolia certificate verifier (a74a8267813d6ea34; the AMM by 14:00, the swap UI by 17:00, the verifier by 20:00), the builder pages, faucet and grants (adb29da59baf27898; /build and /grants by 15:00, the faucet by 16:00), three reference apps that only work on a proven chain (a2060899d2a27d31c; /light by 16:00, /receipt by 18:00, the Sepolia oracle demo by 21:00); the build-server lane stands up rpc.devnet.igneum.network by 12:00; they do not touch the 0.3.25 cut, the crossing or the fleet, sharing the boxes' lease pools (class measure) and the master-only deploy; a lane silent past 25 minutes gets the word from the coordinator and then main. THE FAST-TIME GATE ON THE 0.3.25 PAIR: SUMMARY PASS (cross-0325-39f127a1) at 09:54:40Z (10:54 BST) on the pair 39f127a1 (the node code and object byte for byte e0644958's; igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 09:42:25Z to 09:54:40Z, every check green (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8 at rung 1 at 9,985 bps, the stale node refused, the restart step resynced in 8 s, four sinks equal, 0 PoW rejections); the ceiling's two new fields absent from the 60x file so the ceiling stayed at never there (the node lane's note); to the shipper the same minute; the pin line names e0644958 and its gates. THE FOUNDER'S WORD AT 11:0x BST ("can we add in any more layers? class rotating? things that would render an ASIC useless as soon as it dropped"): the class v6 design opens today as a rotating family, the research lane and the hash lane under the coordinator, the design doc docs/design/class-v6-rotating-family.md by 18:00 BST with the chip-model rows beside each layer (what it does to k and capex for a fixed-function chip and to the per-joule edge for a GPU-like chip; what it costs every GPU tier, Apple included): (1) per-era draws of the class parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement), drawn from chain state like the program; (2) the state-derived dataset's size tracking chain-state growth with a floor, so fixed-memory silicon ages out; (3) scheduled family epochs by height (every 180 days by default) with no release; (4) the (c''') acceptance floor and the F8-form uniformity test generalised to each era's parameter draw, redraw on failure, so layers 1 and 3 need no per-era cryptanalysis. Per layer: the gate it needs (the family analysed as a family: the attack board's shape over the testnet period), the known-failed test, an honest line on what a fully general chip still gets. No consensus code this week; the document, the numbers and the gate plan. Per tier for the founder tonight: what each layer does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE ARC RE-READ IN ITS CHAIN (the hash lane, 10:57 UK): no clear came from the shipper, so the default ran at 10:50: the rotate-fold kit's fetch (sha 65b47211) published to PC 2 at 10:51:41, the run (run-ca3-pc2-v5-intel-bench-20261008, the v5 lane's script c0d398a1) in the publish chain behind another lane's publish-jobs.sh sign --deploy from the build-server worktree (the publisher serialises); the fingerprint line by 11:15 if the publisher frees inside ten minutes, else the blocking process named by 11:10. Queued on PC 1 behind the same publisher: run-ca3-pc1-v5lock-5090-20261008 (class v5 against v4 at unlocked, 1,300 and 1,200 MHz, the v5 kit's CUDA packs), its rows by 12:30; the AMD grid after it from about 11:25. The class-key work on ember-tiers-25 started; the v6 cost rows by 16:00 taken. THE TIERS' CLASS-FLIP STATE, THE UI HALF (the UI lane, 10:57 BST): tiers-class-25 at d949e274 on the mirror, off release-0.3.25's tip 24680e1d (the shipper having merged tiers-25 d3d0704a into release-0.3.25 at 111dae69), the crate gate GREEN 303 + 35 + 8 on build-1, the full pre-push GREEN 60, the UI tests 74 green known-failed first (the v4 tiers stayed on the buttons after the flip on d3d0704a); after the flip the table reads "re-measuring for class v5" on every button with the start minute or "queued (within ten minutes of the crossing)", the v4 watts never current, the strip's sentence naming the crossing; the knee note under the table when knee_loss_pct is over 2 percent; the captures tiers-flip-dark.png and -light.png; the fields tiers_class, program_class, tiers_remeasure_at, knee_loss_pct (the shape sent to the hash lane at 10:4x; the engine sha by 12:30); the default: the display rides 0.3.25 inert if the engine half is late and lights up on 0.3.26. THE FOUNDER'S WORD AT 11:1x BST: class v6 is DECLARED with the four layers as its spine (per-era parameter draws, the dataset tracking chain state, scheduled family epochs by height, the acceptance floor generalised to parameters), and deep past-and-future research opens now under the coordinator with serious resources ("see if anything can be optimised, added or invented"; reading public research is in-house, nothing paid or asked of anyone outside): four research lanes today, (A) history (every ASIC-resistant proof-of-work and how it fell or held: Ethash and the E3 and Linzhi chips, ProgPoW's review, RandomX and its chip analyses, Cuckoo, Equihash and the Z9, Argon2 and Scrypt and the Litecoin chips, KawPow, Autolykos, Octopus, kHeavyHash's chips; the exact mechanism each chip used and what the design missed, each mapped to Igneum's layers with "does v6 close it" as a sentence and a number), (B) the hardware future five years out (PIM and processing-near-memory, HBM3e and HBM4, LPDDR6, 3D DRAM, CXL memory pools, wafer-scale, chiplets, FPGA with HBM; for each the chip-model k band against a state-sized dataset and dependent random reads, and the one layer that would blunt it), (C) invention (layers beyond the four, each a paragraph, a known-failed test and a chip-model row: data-dependent program graphs, latency-bound dependent reads tied to the shard proof, randomised memory topology per era, VRAM-size ratchets, proof-carrying hashes sampled by the pool, time-locked parameter commitments, and what the lane invents; rejecting what costs GPUs more than chips), (D) the family gate (how a parameter family is cryptanalysed as a family: sampling bounds, coverage, the F8-form and (c''') tests over the parameter space, the attack board's shape over the testnet period, so layers 1, 3 and 4 can be automatic with a proof of what was tested). Resources: all four boxes under lease class measure, PC 1 by job for card rows, the rented fleet for one-shot measurements inside the ceiling. Deliverables: a first synthesis in docs/design/class-v6-rotating-family.md by 20:00 BST (the four layers priced, every finding from A to D with its number, a ranked list of what v6 adds beyond the four, the honest line on what a fully general chip still gets), the full report by 09:00 tomorrow, one line to main per lane as each lands; per tier at 20:00: what v6 does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE 0.3.25 APP TIP AND PIN CANDIDATE (the shipper, 10:58 BST): the app tip 9b93e649 (push gate GREEN; the crate unchanged from e0d4425f; the node-source pin to e0644958 and host.sha256 bc8d4f79); the pin candidate the node lane's ceiling cut e0644958 (digest 1b37cb9d, every gate green 10:53, the fast-time SUMMARY PASS 10:54, the floor at DAA 82,800 about 16:53 BST, a publish up to 14:53 without a second cut); the tiers' class-key halves: the UI lane's tiers-class-25 d949e274 green and inert alone, merged with the hash lane's engine sha the moment it lands (12:30), gated as a pair on the release tip, riding only if green by the 13:00 pin; the 0.3.24 Windows take 2 failed at a new place (Inno stopped the app and copied nothing); the update-return lane owns the fix on release-0.3.25 by 12:30, the default the 0.3.25 Windows entry waiting for a clean take 3 while Mac and HiveOS move at the minute. THE FOUR CLASS V6 RESEARCH LANES SPAWNED (the coordinator, 11:0x BST, each with its worktree, its box resources under lease class measure, its clocks and the rules): lane A history (a603a938582c43ab5; the first cut docs/analysis/class-v6/history.md by 15:00), lane B the hardware future (a4f73e2a6f2d1b757; hardware-future.md by 16:00), lane C invention (a5dfe95ee8c47cd0f; invention.md by 17:00), lane D the family gate (a07a99a3788566af2; family-gate.md by 17:00); each feeds the research lane's synthesis docs/design/class-v6-rotating-family.md by 20:00 (its outline by 13:00; the hash lane's per-tier rows by 16:00); the full reports by 09:00 tomorrow; the coordinator's lane mirror carries their clocks. A HELD PUSH AND ITS CAUSE (11:00 BST): the hash lane's push of ca3-v4-amend was refused at 10:58 by the gh-account hook reading the founder's gh (his personal login active again; nothing switched by any lane); the cause is the branch's own hook, which predates the per-process fix (34b0884d): the hook runs the branch's tools/ci, so every branch older than 09:58 must merge the mirror's master before its next push, under which the check reads Igneum's own gh directory and skips under the suspension marker; the rule to every lane. Live: the Arc re-read on PC 2 (published 10:59:47) and the v5lock job on PC 1 (published 10:53, about 12 minutes). THE CLASS V6 OUTLINE ON THE MIRROR (the research lane, docs/design/class-v6-rotating-family.md on counter-asic-4, the commit after fb61ed4b, pushed 10:5x UTC, two hours ahead of 13:00): section 0 the founder's table (per layer, what it does to a fixed-function chip and to a GPU-like chip on its release day, and the 5090, 5070 Ti and M5 Max columns, measured where the night's rows exist, the 5070 Ti scaled until the hash lane's row); the honest frame on top: the four layers render a FIXED-FUNCTION chip useless on the first era its wired value leaves (one tape-out lives one era) and move nothing for the stored-dataset chip with a programmable core except the core's size and the N5 project it forces; that chip keeps 3.6x at zero premium and 2.1x at k = 1 on a 5090 at its knee. The layer table (sections 1 and 2) names the bands each draw takes and the measured rows that set them: the mixer in {4, 8, 16} (x16 open), the op-mix weights within B = 4 with shuffle and mulhi capped (shfl 55.8 pJ per op), the read width in {1, 4} words (w64 excluded by the 5 October rows), the block shape 64 to 256 (never 1,024), N left to the ladder's signal (an unconditional draw retires the Apple tier at 200,000). Open numbers asked of the hash lane with defaults at 16:00: the 5070 Ti row (the rented 5070 scaled), the x16 mixer's verifier and build (the chip model's estimate), two re-weighted shadow packs for the op-mix band (the microbench arithmetic). Layers 2 to 4 and the gate plan are skeletons with their sources named, filling by 18:00 with the four research lanes' cuts, the synthesis by 20:00. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From dea360cc654641a23f0f7924e6b35d888969b7d6 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:02:02 +0000 Subject: [PATCH 06/19] release record section 24: the first lock on cc902690 (checkpoint 1931, 10:06:32 BST), the move's count and the three puller faults closed, 0.3.24 Windows take 2 (Inno stopped the app and copied nothing; the /IGNOTA=1 rule), 0.3.25 opened (tip 78dfe65e, the ceiling cut e0644958, digest 1b37cb9d, the DMG 501ba293 staged, the clock), the tiers at the class flip, the signed apps pipeline (Igneum Labs LLC) Co-Authored-By: Claude Fable 5.1 --- docs/plans/release-0.3.22.md | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-) diff --git a/docs/plans/release-0.3.22.md b/docs/plans/release-0.3.22.md index c5c28cc7d..be4aefc3e 100644 --- a/docs/plans/release-0.3.22.md +++ b/docs/plans/release-0.3.22.md @@ -201,7 +201,7 @@ Main's word on route (A) or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 B ## 22. The morning: main's word, the fifth cut, the move staged (09:3x to 09:5x BST, 8 October) -**The night's silence** was the harness: main's command failed at 23:36 BST and it got no turn until 09:3x; the build-server lane and the fleet lane sat on backgrounded chains from 22:54 and 22:58 BST with nothing lost on the boxes (both idle); all three answered at 09:31 to 09:34. **Main's word (09:3x BST):** route (A) by the shipper (the node lane re-cuts the floor from the minute 10:45 BST; the shipper applies the pack-gate line with the fleet's tooling, places the signed move file at at_epoch 0 the moment the pin is green, names the minute, restarts build-1's three nodes and publishes the Mac entry at the minute; a pin not green by 10:35 slides the minute to the pin plus 25; the ceiling 12:50); the build-server lane builds the seed, Windows and hive pairs on the morning pin under lease class release and hands them over; the Windows app chain is the shipper's after the move (0.3.23 take 3 skipped as moot); "PC 2 clear" for the hash lane's Arc job; the two lanes respawned (the new fleet lane takes the readings after the move, not the move). Then (09:5x): the project lead's word, get it live now: the minute is the last FETCHED plus ten, 10:45 the ceiling not the target; everything but the crossing done by 11:00 BST, the Windows chain in parallel. **Standing authority from main:** if the reachable boxes have not all FETCHED by 10:35 the shipper says "slide" itself; when the margin to a floor's publish ceiling falls under 15 minutes, re-cut without asking. **The founder's rule ("this cannot happen again"):** every ask to main carries a default action and a deadline, silence at the deadline means the default, never a stand-down; the Counter lane mirrors every clock the shipper holds today from the shipper's runbook (scratch r0324/RUNBOOK-0324-move.md, twelve steps with commands, hosts, keys and read-backs). +**The night's silence** was the harness: main's command failed at 23:36 BST and it got no turn until 09:3x; the build-server lane and the fleet lane sat on backgrounded chains from 22:54 and 22:58 BST with nothing lost on the boxes (both idle); all three answered at 09:31 to 09:34. **Main's word (09:3x BST):** route (A) by the shipper (the node lane re-cuts the floor from the minute 10:45 BST; the shipper applies the pack-gate line with the fleet's tooling, places the signed move file at at_epoch 0 the moment the pin is green, names the minute, restarts build-1's three nodes and publishes the Mac entry at the minute; a pin not green by 10:35 slides the minute to the pin plus 25; the ceiling 12:50); the build-server lane builds the seed, Windows and hive pairs on the morning pin under lease class release and hands them over; the Windows app chain is the shipper's after the move (0.3.23 take 3 skipped as moot); "PC 2 clear" for the hash lane's Arc job; the two lanes respawned (the new fleet lane takes the readings after the move, not the move). Then (09:5x): the founder's word, get it live now: the minute is the last FETCHED plus ten, 10:45 the ceiling not the target; everything but the crossing done by 11:00 BST, the Windows chain in parallel. **Standing authority from main:** if the reachable boxes have not all FETCHED by 10:35 the shipper says "slide" itself; when the margin to a floor's publish ceiling falls under 15 minutes, re-cut without asking. **The founder's rule ("this cannot happen again"):** every ask to main carries a default action and a deadline, silence at the deadline means the default, never a stand-down; the Counter lane mirrors every clock the shipper holds today from the shipper's runbook (scratch r0324/RUNBOOK-0324-move.md, twelve steps with commands, hosts, keys and read-backs). **The fifth cut, the pin: release-0.3.24-node = 5b673577** (dfbd1e10 with program_class_v5_activation_daa 68,400, epoch 19, nothing else; cut at 09:33:18 BST from build-1's seed at DAA 56,329, the chain at 1.0 DAA/s all night; the publish DAA at 10:45 about 60,630, the floor about 12:54 BST, the three heights under it; the floor holds for a publish up to DAA 61,200, about 10:54 BST); every gate green at 09:39:15 BST (build at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary with digest cc9026909eddbadb and the mixed-version refusal against a 2720d8d2 node both ways; the testnet canary on b2e856ed); the node-lane pair igneumd a3b1a2c9 / igneum-miner cfa9f5ca under /srv/artefacts/0324-5b673577/node-lane, igneum-pow-v5 8 paths; the fast-time SUMMARY on it due about 09:55. The app side: release-0.3.24 = 0c47b59a (the knob's display knob-24 2c4dc617 merged, UI 88; the Windows pin to 5b673577 at ebbdbc32); the Mac node pair (igneumd aee81264, igneum-miner 129275ef) and the DMG 7e6e3eb3 built under the lock, the Mac entry re-staged in both folders. @@ -214,3 +214,15 @@ Main's word on route (A) or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 B ## 23. The move executed: 0.3.24 Mac and HiveOS LIVE (10:05 to 10:09 BST, 8 October) **THE MINUTE 10:05:00 BST** fired on the signed file m5b67-1 (FETCHED 36 of 39 at 10:00: dn3-agg48 the 48 GB aggregator prover still renting, p2-3090-1 refusing ssh since the night, p2-4090-1b behind the dead proxy ssh2; all eight nodes dead overnight back on 2720d8d2 and fetched by 09:53, dn3-r03 and r04 given a saved environment by hand). **build-1's three on the pin:** node1-dn3 (26671, JSON 28670) and the observer (26651, JSON 28650) up at 10:08 on "igneumd/2.1.0-5b673577", digest cc902690, object version 6, the N15 line ("the 15611 records up to the tip 15611 are continuous"); the seed (26631, JSON 27632) at 10:09 after a first start at 10:08 panicked on the old process's RocksDB lock (meta/LOCK "Resource temporarily unavailable"), up on cc902690 with 3 outgoing connections. The shipper's fault, two parts: the three-node script's --go branch did not execute at 10:05 (the dry-run text printed and it exited: the kill pattern `pgrep -f appdir=...` matched the ssh shell carrying the script, the fleet's fault class of 22:34 the night before), so the move ran by hand at 10:07 with the victims found by process name and cmdline; and the seed restarted before the old process had released its datadir (rule 11 extended). **The 0.3.24 Mac entry LIVE at 10:08:35 BST** (the copy at 10:05:01, one Vercel deploy), both token folders, channel devnet-3: Igneum-Miner-0.3.24-5b673577.dmg 7e6e3eb3 (45,652,380 B; app 0c47b59a: the core-clock knob and its display, the Power Helper unattended, the dashboard caps, the pool daemon's reconnect, the testnet re-cut; the Mac node pair aee81264/129275ef from 5b673577), interface 1.0.2, the floor file kept for the apps below 0.3.22, read back from both folders and the public Mac alias; **the HiveOS package igneum-hive-0.3.24.tar.gz 025bf01f** (the 5b673577 hive pair, the two sub-version 3 kits and the v5 kit e6c088bb) in both folders and on the public alias at the same deploy; dl/public/igneum-downloads.json names 0.3.24 for Mac and HiveOS. The notes carry the knob, the floor from the minute and "update before the move minute or the node stops following Devnet 3". Owed from the fleet: the APPLIED count per box, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07 (the installer job from the a4c5a855 kit, sha ad7a50e8, and the payload igneum-windows-app-0.3.24-a4c5a855-msvc.zip 7f12cbe3 with the PC 1 host 0e241c94 built at 09:50 in 7 s, the 5b673577 Windows pair 0b144d7d/0cc68d9e, the 0c47b59a app exes ce7f82a6/99edc6aa/d035368e; host.sha256 at a4c5a855), the rule 14 smoke as the gate, then the Windows entry into both folders with the public alias and the card ("Mac, HiveOS and Windows now"); the hash lane's Arc re-read on the rotate-fold kit (packs-ca3-v5-20261008T085619Z.zip 65b47211) and the update-return lane's PC 1 re-probe and PC 2 S4U proof after the smoke. The crossing at DAA 68,400 about 12:54 BST (the node lane's watch on build-1's seed). + +## 24. After the move; 0.3.25 opened the same morning (10:06 to 11:05 BST, 8 October) + +**The first lock on cc902690:** checkpoint 1931, blue score 57,930, at 10:06:32 BST on dn3-g1, 92 seconds after the minute. The move's count: 25 boxes APPLIED by the puller, 11 moved by hand (the three puller faults: the miner sha missing from box-dn3.sh's gate list on 18 miners, the MINE=0 boxes with no env on 11 nodes, the unquoted env-last that took 9 nodes down), p2-3090-1 and p2-4090-1b unreachable, dn3-agg48 joining with its bring-up. The three faults are closed on every box before the next move (the fleet lane, 10:41 BST): the gate list is a file the puller appends to at fetch time, every box holds move/env-last, the puller writes a quoted copy of the environment before sourcing it; a dry move file read FETCHED 38 of 39 at 10:39 BST. + +**0.3.24 Windows did not ship.** Take 2 on PC 2 (installer 2b8774b1 from kit 45836da4, the TDateTime line gone) ran silent without /IGNOTA=1: the stop step closed the 0.3.21 app, then Inno copied no file (every exe still 0.3.21, install-running.flag PRESENT from 09:24:17Z, so the abort fell between PrepareToInstall and the copy), and nothing relaunched the app (the installer's own relaunch is postinstall skipifsilent, or OtaRelaunch on /IGNOTA=1), which left PC 2 without its job runner for 16 minutes until the relay's igneum-agent started the app. Two rules from it: every install job carries /VERYSILENT /NORESTART /IGNOTA=1 and ends on the new engine's --version; the Inno log's reason is the update-return lane's fix on release-0.3.25, known-failed first. The 0.3.24 Windows entry is skipped; the Windows line lands with 0.3.25; the Devnet 3 card publishes on the 0.3.25 Windows entry. + +**0.3.25 opened at 10:2x BST on the founder's word "push 0.3.25 everywhere asap".** release-0.3.25 from release-0.3.24's final tip 45836da4, the version bump first (rule 15): amd-clock-25 e2962b89 (the AMD core-clock ladder; the ADLX telemetry exe 1d8e055d a kit input), tiers-25 d3d0704a (the three-tier Ember Tune, both halves), the Intel rotate-fold header already in the tree through tiers-25 (9088293a), the node-source pin and host.sha256 bc8d4f79 (PC 1, MSVC, 10:37 BST; the one-line form was not read by host-gate.py, so the sha on its own line, the note a comment line above it). Crate gate on build-1 GREEN at e0d4425f (303+35+8, 10:29 BST) and unchanged since; the tip 9b93e649 at 10:58 BST. The node line moved twice under it: 7bd2940f (c6629572 plus unpaid proof records re-announced every 120 s, nothing consensus), then the ceiling cut **e0644958** (proving_fee_ceiling_activation_daa 82,800 in the Devnet 3 object, four gwei per pgas; every gate green 10:53 BST, fast-time SUMMARY PASS 10:54 on the same object; digest 1b37cb9da3911c2e; the 0.3.24 nodes and it refuse each other both ways, so the move is the one-minute form; DAA 82,800 about 16:53 BST; a publish up to DAA 75,600 = 14:53 BST needs no second cut). The Mac pair and DMG on e0644958: Igneum-Miner-0.3.25-e0644958.dmg 501ba293 (45,780,063 B; node pair 556926b1/d2dfe966), staged in both token folders at 11:00 BST; the pair tarball fleet/e0644958-node-lane.tgz fed2f4d6 served. The clock: the pin named at 13:00 BST after the 68,400 crossing reads clean (about 12:54), the fleet publishes m0644-1 on that line, the minute = last FETCHED + 10 (13:40 provisional, never past 14:53), Mac and HiveOS at the minute, Windows within the hour behind on a clean take 3, the card after. + +**The tiers at the class flip (main, 10:4x BST):** a power-capped A4000 reads 17 percent under class v4 at an equal fingerprint under class v5 (the A100 1.3, uncapped consumer cards level), so the engine half gains a class key (a class change invalidates the stored tiers and re-runs the search within ten minutes) and the UI half shows "re-measuring for class v5" and never applies v4 tiers as current after the flip (tiers-class-25 d949e274, green, inert alone). Both ride 0.3.25 only if green together on the release tip by the 13:00 pin, else the tiers ship as gated at d3d0704a and this record states that the v4 tiers may be off by the measured percentage until the re-tune. The knee: the hash lane's 5090 rows at the 1,300 lock under v5 against v4 by 12:30; over 2 percent loss and the knee is re-found under v5 with the percentage, else the v4 knee stands. + +**Signed apps (main's order, built 09:40 to 10:43 BST, signing-pipeline-25 at 075fb413):** packaging/mac/sign-notarize.sh inside build-dmg.sh (Developer ID over the ad hoc signature, hardened runtime, timestamp, notarytool --wait, stapler, spctl read-back; config dir ~/.config/igneum/apple-notary) and packaging/windows/sign.ps1 inside build-installer.ps1 (signtool through a cloud CSP or SSL.com's CodeSignTool; the payload exes before ISCC, Setup and the uninstaller through Inno's SignTool line; config ~/.config/igneum/win-codesign). Every artefact ends in "signed: " or "unsigned: " in the log and in .signing, never silent; a present but broken config stops the build; the publisher lives once in packaging/sign/publisher.txt, **Igneum Labs LLC** (the signing entity, Wyoming, filed 8 October 2026; the app's display name stays Igneum Miner). Known-failed self-tests on the pre-push gate (9 Mac cases; the 7 Windows cases run as a PC 2 job). Both config paths absent at the commit; the branch lands on master on PC 2's "7 of 7" line, else 15:00 BST. From 443ac757fec24baec2071ba022a1ad53ce2a6277 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:05:07 +0000 Subject: [PATCH 07/19] Bench table: the model sweep's last rows (RTX 4070 Ti 31.24 MH/s at 95.4 W; L40S 56.41 at 220.7 W in the datacentre table; RTX 4080 Super 42.60 at 134.9 W); the GTX 1080 Ti row's flag cleared by its second host (five of five self-tests pass, the failure was the first host's VRAM); the page carries the sweep's two findings as one line (class v4 cap-proof; class v5 level on uncapped cards and 17 to 18 percent under on two power-capped hosts, so a v5 rate needs the power state beside it) Co-Authored-By: Claude Fable 5.1 --- site/build.mjs | 3 ++ site/miner-bench.json | 70 +++++++++++++++++++++++++++++++++++++++++-- site/miners.html | 11 +++---- 3 files changed, 77 insertions(+), 7 deletions(-) diff --git a/site/build.mjs b/site/build.mjs index ccd3f4571..f1dcd61db 100644 --- a/site/build.mjs +++ b/site/build.mjs @@ -477,6 +477,8 @@ for (const [file, active] of PAGES) { const bestStock = ranked(cur.filter(r => tunedState(r)[0] === 'stock'))[0]; const bestTuned = ranked(cur.filter(r => tunedState(r)[0] === 'tuned'))[0]; const name = (r) => esc(r.card.replace(/ \(.*$/, '')); + // the rented model sweep of 8 October 2026 (23 rows on 22 models, every class v5 fingerprint matched): its two page-level findings + const sweepLine = `

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found class v4's rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held class v4's rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it; the table's rates are class v4.

`; const bestLine = best ? `

Best desktop card: ${name(best)}, ${esc(fmt(best.mh_s))} MH/s (measured, ${esc(best.date)}).` + (bestStock ? ` Best stock MH per wall watt: ${name(bestStock)}, ${esc(fmt3(bestStock.mh_per_w))} at ${esc(fmt(bestStock.watts))} W, untuned (${esc(bestStock.by)}, ${esc(bestStock.date)}).` : '') + (bestTuned ? ` Best tuned: ${name(bestTuned)}, ${esc(fmt3(bestTuned.mh_per_w))} at ${esc(fmt(bestTuned.watts))} W (measured, ${esc(bestTuned.date)}).` : '') + '

' : ''; @@ -553,6 +555,7 @@ table.bench2 tr.detail .d{display:block;margin:0 0 4px}table.bench2 tr.detail b{ sortStyle, '

The table

', bestLine, + sweepLine, '

One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. MH per wall watt uses board or wall power; a row whose watts are the chip\'s (Apple silicon: GPU plus DRAM from IOReport) says so and is not ranked on that column. Integrated GPUs are not listed.

', '

Cards you can buy

', table, diff --git a/site/miner-bench.json b/site/miner-bench.json index 83b41a087..3673fdc95 100644 --- a/site/miner-bench.json +++ b/site/miner-bench.json @@ -1145,8 +1145,8 @@ "watts": 124.3, "mh_per_w": 0.131, "driver_os": "NVIDIA driver 580.173.02, Ubuntu 24.04 (CUDA 12.8.1 image)", - "source": "model sweep 2026-10-08, row 17 (a rented card, result cb2-gtx-1080-ti-11-gb-0377-result.json; 09:49Z)", - "note": "FLAG: one of three class v4 builds on this host failed its cache self-test (the 256 MiB cache's FNV read wrong on that build while the other two and the class v5 kit build were clean: the host's VRAM or a Pascal-only build race); a second host decides, and until then this row carries the flag. The rate is the mean of the two clean runs (16.261, 16.260). Rented single-card host, 0.10 h; power.draw mean 124.3 W (max 125.5, limit 280 W; no power average on Pascal), SM 1,936 MHz mean, memory 5,005 MHz, 59 C; self-test PASS on the clean builds (96 of 96 vector lanes); the class v5 kit holds the rate (16.27 MH/s, fingerprint matched); read ceiling 16.3 MH/s, the hash at 1.00 of it. The first Pascal row: a 2017 card that runs the class at its own ceiling", + "source": "model sweep 2026-10-08, rows 17 and 22 (two rented hosts, results cb2-gtx-1080-ti-11-gb-0377-result.json and cb2-gtx-1080-ti-11-gb-1452-result.json; 09:49Z and 10:02Z)", + "note": "the mean of the first host's two clean runs (16.26 MH/s at 124.3 W, the card uncapped at a 280 W limit); a second host read 16.29 MH/s at 129.4 W under a 130 W host cap with five of five self-tests passing, so the first host's one-in-three cache self-test failure was that host's VRAM and not Pascal, and the flag is cleared. Power.draw on the first host max 125.5 W (no power average on Pascal), SM 1,936 MHz mean, memory 5,005 MHz, 59 C; the class v5 kit held the rate on the uncapped host (16.27 MH/s) and read 13.27 MH/s on the 130 W-capped one (fingerprint matched on both); read ceiling 16.3 MH/s, the hash at 1.00 of it. The first Pascal row: a 2017 card that runs the class at its own ceiling", "hive": { "core_mhz": null, "mem_mhz": null, @@ -1197,6 +1197,72 @@ "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA RTX 4070 Ti (12 GB)", + "group": "buy", + "mh_s": 31.24, + "watts": 95.4, + "mh_per_w": 0.327, + "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 20 (a rented card, result cb2-rtx-4070-ti-12-gb-0695-result.json; 09:57Z)", + "note": "rented single-card host, 0.18 h; power.draw mean 95.4 W (max 96.7, limit 285 W), SM 2,825 MHz mean, memory 10,251 MHz, 49 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.28 MH/s, fingerprint matched); read ceiling 31.7 MH/s, the hash at 0.99 of it. Repeats the 7 October rate to 0.07 percent on another provider at 12 W less; the 192-bit Ada pair (4070 Ti, 4070 Super) land on one number, 31.2 MH/s under 100 W", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA L40S (48 GB)", + "group": "datacentre", + "mh_s": 56.41, + "watts": 220.7, + "mh_per_w": 0.256, + "driver_os": "NVIDIA driver 570.133.20, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 21 (a rented card, result cb2-l40s-48-gb-0890-result.json; 09:57Z)", + "note": "rented single-card host, 0.17 h; power.draw mean 220.7 W (max 227.2, limit 350 W), SM 2,520 MHz, memory 9,001 MHz, 45 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (56.51 MH/s, fingerprint matched); read ceiling 57.8 MH/s, the hash at 0.98 of it. Repeats the 7 October row (56.36 then, 56.41 now) at 20 W less on this host; a 384-bit GDDR6 card that lands where the 3090 Ti and the A6000 do at a datacentre price. Driver 570.133, the oldest of the sweep, ran the CUDA 12.8 worker clean: 570 is a safe floor", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } + }, + { + "generator": "v2", + "miner": "bench pack, igneum-worker-cuda, class v4 program (3 x 300 dispatches of 2^22)", + "date": "2026-10-08", + "by": "measured by the fleet", + "v4_cost": "not measured (class v4 program only)", + "tuned": "stock, bench only (rented, no clock control on the host)", + "card": "NVIDIA RTX 4080 Super (16 GB)", + "group": "buy", + "mh_s": 42.6, + "watts": 134.9, + "mh_per_w": 0.316, + "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04 (CUDA 12.8.1 image)", + "source": "model sweep 2026-10-08, row 23 (a rented card, result cb2-rtx-4080-super-16-g-2453-result.json; 10:03Z)", + "note": "rented single-card host, 0.12 h; power.draw mean 134.9 W (max 138.4, limit 320 W), SM 2,740 MHz mean, memory 11,254 MHz, 51 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (42.68 MH/s, fingerprint matched); read ceiling 43.1 MH/s, the hash at 0.99 of it. The 23 Gbps GDDR6X gives it 4.6 percent over the 4080 for 6 W more at the same MH per watt: the Ada 256-bit tier is 41 to 43 MH/s whichever card", + "hive": { + "core_mhz": null, + "mem_mhz": null, + "pl_w": null, + "label": "stock (rented card, no clock control on the host; no measured tune point)" + } } ] } diff --git a/site/miners.html b/site/miners.html index 8abcfc7f6..02d373359 100644 --- a/site/miners.html +++ b/site/miners.html @@ -213,7 +213,7 @@ table{min-width:560px}
Igneum/GPU bench table
-
36 cards you can buy, 12 datacentre, 0 fleet tuning models
+
38 cards you can buy, 13 datacentre, 0 fleet tuning models

GPU bench table

Measured hash rates per card on the Igneum lottery hash, with the generator version, the miner version, the date and the log entry behind each number.

@@ -263,12 +263,13 @@ table.bench2 tr.detail .d{display:block;margin:0 0 4px}table.bench2 tr.detail b{

The table

Best desktop card: NVIDIA RTX 5090, 136.1 MH/s (measured, 2026-10-06). Best stock MH per wall watt: NVIDIA RTX 5070 Ti, 0.577 at 135.9 W, untuned (measured by the fleet, 2026-10-08). Best tuned: NVIDIA RTX 5090, 0.563 at 226.8 W (measured, 2026-10-06).

+

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found class v4's rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held class v4's rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it; the table's rates are class v4.

One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. MH per wall watt uses board or wall power; a row whose watts are the chip's (Apple silicon: GPU plus DRAM from IOReport) says so and is not ranked on that column. Integrated GPUs are not listed.

Cards you can buy

-
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.50.577135.9 WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.40.539145.6 WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-07620.248249.5 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-07590.221267.3 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.90.201293.5 WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-07520.506102.8 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.80.468110.8 WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.20.289142.6 WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.70.317128.6 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-07390.219178.3 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-08390.22177.1 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.10.261142.3 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.10.256129.5 WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.70.316100.4 WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.30.41575.4 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.30.291107.3 WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.20.312100.2 WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.90.241111.6 WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-08260.116224.6 WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.10.25977.5 WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.10.25379.2 WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.10.115166 WStock
NVIDIA RTX 4060 (8 GB)class v4 not measured · 2026-10-0819.1not measurednot readStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
NVIDIA GTX 1080 Ti (11 GB)class v4 not measured · 2026-10-0816.30.131124.3 WStock
NVIDIA RTX 2060 (6 GB)class v4 not measured · 2026-10-0814.10.117120.6 WStock
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock
-

36 cards. Details opens a card's Hive flight-sheet values, its full class v4 cost, miner, source and note.

-
Datacentre (12 cards, rented for the measurement; about three times the rented dollars per hash of a desktop card) -
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.40.487855.6 WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-073130.723432.9 WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.70.645385.6 WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.30.605411.7 WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-071550.517299.6 WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.40.52266.3 WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.10.515268.6 WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.50.452288.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.40.234240.7 WStock
NVIDIA RTX A6000 (48 GB)class v4 not measured · 2026-10-0853.60.224239.1 WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.40.279112.8 WStock
+
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.50.577135.9 WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.40.539145.6 WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-07620.248249.5 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-07590.221267.3 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.90.201293.5 WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-07520.506102.8 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.80.468110.8 WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4080 Super (16 GB)class v4 not measured · 2026-10-0842.60.316134.9 WStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.20.289142.6 WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.70.317128.6 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-07390.219178.3 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-08390.22177.1 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.10.261142.3 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.10.256129.5 WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.70.316100.4 WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.30.41575.4 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.30.291107.3 WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.20.312100.2 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0831.20.32795.4 WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.90.241111.6 WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-08260.116224.6 WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.10.25977.5 WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.10.25379.2 WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.10.115166 WStock
NVIDIA RTX 4060 (8 GB)class v4 not measured · 2026-10-0819.1not measurednot readStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
NVIDIA GTX 1080 Ti (11 GB)class v4 not measured · 2026-10-0816.30.131124.3 WStock
NVIDIA RTX 2060 (6 GB)class v4 not measured · 2026-10-0814.10.117120.6 WStock
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock
+

38 cards. Details opens a card's Hive flight-sheet values, its full class v4 cost, miner, source and note.

+
Datacentre (13 cards, rented for the measurement; about three times the rented dollars per hash of a desktop card) +
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.40.487855.6 WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-073130.723432.9 WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.70.645385.6 WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.30.605411.7 WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-071550.517299.6 WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.40.52266.3 WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.10.515268.6 WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.50.452288.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0856.40.256220.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.40.234240.7 WStock
NVIDIA RTX A6000 (48 GB)class v4 not measured · 2026-10-0853.60.224239.1 WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.40.279112.8 WStock
Earlier classes (6 rows: the genesis program, the hourly program, class v3 before the shadow; not comparable with the table above)

The bench numbers of 3 and 4 October 2026. A higher MH/s here is a different hash, not a faster card.

From 93a24c5901e52d5bd0b39d60a7b4eb9f094a46ff Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:07:44 +0000 Subject: [PATCH 08/19] DEX: the AMM in the Uniswap v2 shape for Devnet 3 (factory, pairs, router, wrapped IGN, two faucet test tokens), its Foundry tests and the deploy script (8 October 2026) Co-Authored-By: Claude Fable 5.1 --- contracts/dex/foundry.toml | 20 +++ contracts/dex/script/Deploy.s.sol | 56 +++++++ contracts/dex/src/ERC20.sol | 59 +++++++ contracts/dex/src/IgneumPair.sol | 174 ++++++++++++++++++++ contracts/dex/src/IgneumRouter.sol | 253 +++++++++++++++++++++++++++++ contracts/dex/src/TestToken.sol | 29 ++++ contracts/dex/src/WIGN.sol | 27 +++ contracts/dex/test/Dex.t.sol | 137 ++++++++++++++++ contracts/dex/test/Vm.sol | 22 +++ 9 files changed, 777 insertions(+) create mode 100644 contracts/dex/foundry.toml create mode 100644 contracts/dex/script/Deploy.s.sol create mode 100644 contracts/dex/src/ERC20.sol create mode 100644 contracts/dex/src/IgneumPair.sol create mode 100644 contracts/dex/src/IgneumRouter.sol create mode 100644 contracts/dex/src/TestToken.sol create mode 100644 contracts/dex/src/WIGN.sol create mode 100644 contracts/dex/test/Dex.t.sol create mode 100644 contracts/dex/test/Vm.sol diff --git a/contracts/dex/foundry.toml b/contracts/dex/foundry.toml new file mode 100644 index 000000000..efe433d0b --- /dev/null +++ b/contracts/dex/foundry.toml @@ -0,0 +1,20 @@ +[profile.default] +src = "src" +test = "test" +script = "script" +out = "out" +libs = [] +solc_version = "0.8.28" +# Devnet 3's executor runs the EVM through revm; Paris keeps the bytecode off PUSH0 and transient storage so it runs +# on any post-Merge configuration of it (8 October 2026). +evm_version = "paris" +optimizer = true +optimizer_runs = 200 +fs_permissions = [{ access = "read-write", path = "./deploy-out.json" }] +# no remote dependencies: the tests and the script declare the cheatcode interface they use (contracts/dex/test/Vm.sol) +auto_detect_remappings = false +offline = true + +[rpc_endpoints] +# the Devnet 3 EVM reaches the build box through a tunnel on this port (never the Mac); see docs/contracts/devnet-3.json +devnet3 = "http://127.0.0.1:36790" diff --git a/contracts/dex/script/Deploy.s.sol b/contracts/dex/script/Deploy.s.sol new file mode 100644 index 000000000..a3d733380 --- /dev/null +++ b/contracts/dex/script/Deploy.s.sol @@ -0,0 +1,56 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {Vm, VM_ADDRESS} from "../test/Vm.sol"; +import {WIGN} from "../src/WIGN.sol"; +import {TestToken} from "../src/TestToken.sol"; +import {IgneumFactory} from "../src/IgneumPair.sol"; +import {IgneumRouter} from "../src/IgneumRouter.sol"; + +/// Deploys the AMM on Devnet 3 from the key in DEX_DEPLOYER_KEY (read from the environment, never printed), seeds +/// three pools from the faucet and 200 IGN, and writes the addresses to deploy-out.json for docs/contracts. +/// +/// forge script script/Deploy.s.sol:Deploy --rpc-url devnet3 --broadcast --sig "run()" +/// +/// Devnet 3, test tokens, no value. +contract Deploy { + Vm constant vm = Vm(VM_ADDRESS); + + function run() external { + uint256 key = vm.envUint("DEX_DEPLOYER_KEY"); + address deployer = vm.addr(key); + vm.startBroadcast(key); + + WIGN wign = new WIGN(); + TestToken tta = new TestToken("Test Token A", "TTA"); + TestToken ttb = new TestToken("Test Token B", "TTB"); + IgneumFactory factory = new IgneumFactory(); + IgneumRouter router = new IgneumRouter(address(factory), address(wign)); + + tta.drip(); + ttb.drip(); + tta.approve(address(router), type(uint256).max); + ttb.approve(address(router), type(uint256).max); + uint256 deadline = block.timestamp + 1 hours; + router.addLiquidityIGN{value: 100 ether}(address(tta), 500 ether, 0, 0, deployer, deadline); + router.addLiquidityIGN{value: 100 ether}(address(ttb), 500 ether, 0, 0, deployer, deadline); + router.addLiquidity(address(tta), address(ttb), 500 ether, 500 ether, 0, 0, deployer, deadline); + + vm.stopBroadcast(); + + string memory json = string.concat( + "{\n", + " \"WIGN\": \"", vm.toString(address(wign)), "\",\n", + " \"TTA\": \"", vm.toString(address(tta)), "\",\n", + " \"TTB\": \"", vm.toString(address(ttb)), "\",\n", + " \"IgneumFactory\": \"", vm.toString(address(factory)), "\",\n", + " \"IgneumRouter\": \"", vm.toString(address(router)), "\",\n", + " \"pair_TTA_WIGN\": \"", vm.toString(factory.getPair(address(tta), address(wign))), "\",\n", + " \"pair_TTB_WIGN\": \"", vm.toString(factory.getPair(address(ttb), address(wign))), "\",\n", + " \"pair_TTA_TTB\": \"", vm.toString(factory.getPair(address(tta), address(ttb))), "\",\n", + " \"deployer\": \"", vm.toString(deployer), "\"\n", + "}\n" + ); + vm.writeFile("deploy-out.json", json); + } +} diff --git a/contracts/dex/src/ERC20.sol b/contracts/dex/src/ERC20.sol new file mode 100644 index 000000000..668ec63d3 --- /dev/null +++ b/contracts/dex/src/ERC20.sol @@ -0,0 +1,59 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +/// A plain ERC-20 with EIP-2612-free approvals, shared by the pair's liquidity token, the wrapped coin and the test +/// tokens. Devnet 3, test tokens, no value. +contract ERC20 { + string public name; + string public symbol; + uint8 public constant decimals = 18; + uint256 public totalSupply; + mapping(address => uint256) public balanceOf; + mapping(address => mapping(address => uint256)) public allowance; + + event Transfer(address indexed from, address indexed to, uint256 value); + event Approval(address indexed owner, address indexed spender, uint256 value); + + constructor(string memory name_, string memory symbol_) { + name = name_; + symbol = symbol_; + } + + function _mint(address to, uint256 value) internal { + totalSupply += value; + balanceOf[to] += value; + emit Transfer(address(0), to, value); + } + + function _burn(address from, uint256 value) internal { + balanceOf[from] -= value; + totalSupply -= value; + emit Transfer(from, address(0), value); + } + + function _transfer(address from, address to, uint256 value) internal { + balanceOf[from] -= value; + balanceOf[to] += value; + emit Transfer(from, to, value); + } + + function approve(address spender, uint256 value) external returns (bool) { + allowance[msg.sender][spender] = value; + emit Approval(msg.sender, spender, value); + return true; + } + + function transfer(address to, uint256 value) external returns (bool) { + _transfer(msg.sender, to, value); + return true; + } + + function transferFrom(address from, address to, uint256 value) external returns (bool) { + uint256 allowed = allowance[from][msg.sender]; + if (allowed != type(uint256).max) { + allowance[from][msg.sender] = allowed - value; + } + _transfer(from, to, value); + return true; + } +} diff --git a/contracts/dex/src/IgneumPair.sol b/contracts/dex/src/IgneumPair.sol new file mode 100644 index 000000000..2a3dc2a08 --- /dev/null +++ b/contracts/dex/src/IgneumPair.sol @@ -0,0 +1,174 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {ERC20} from "./ERC20.sol"; + +interface IERC20Minimal { + function balanceOf(address) external view returns (uint256); + function transfer(address, uint256) external returns (bool); +} + +/// A constant-product pool in the Uniswap v2 shape: reserves of two tokens, a 0.3 percent fee kept in the pool, +/// liquidity tokens for the depositors, MINIMUM_LIQUIDITY locked at the first mint. No protocol fee, no price +/// accumulators, no flash callback. Devnet 3, test tokens, no value. +contract IgneumPair is ERC20("Igneum LP", "IGN-LP") { + uint256 public constant MINIMUM_LIQUIDITY = 10 ** 3; + + address public immutable factory; + address public token0; + address public token1; + + uint112 private reserve0; + uint112 private reserve1; + uint32 private blockTimestampLast; + + uint256 private unlocked = 1; + + event Mint(address indexed sender, uint256 amount0, uint256 amount1); + event Burn(address indexed sender, uint256 amount0, uint256 amount1, address indexed to); + event Swap(address indexed sender, uint256 amount0In, uint256 amount1In, uint256 amount0Out, uint256 amount1Out, address indexed to); + event Sync(uint112 reserve0, uint112 reserve1); + + modifier lock() { + require(unlocked == 1, "Pair: locked"); + unlocked = 0; + _; + unlocked = 1; + } + + constructor() { + factory = msg.sender; + } + + function initialize(address token0_, address token1_) external { + require(msg.sender == factory, "Pair: forbidden"); + token0 = token0_; + token1 = token1_; + } + + function getReserves() public view returns (uint112, uint112, uint32) { + return (reserve0, reserve1, blockTimestampLast); + } + + function _safeTransfer(address token, address to, uint256 value) private { + (bool ok, bytes memory data) = token.call(abi.encodeWithSelector(IERC20Minimal.transfer.selector, to, value)); + require(ok && (data.length == 0 || abi.decode(data, (bool))), "Pair: transfer failed"); + } + + function _update(uint256 balance0, uint256 balance1) private { + require(balance0 <= type(uint112).max && balance1 <= type(uint112).max, "Pair: overflow"); + reserve0 = uint112(balance0); + reserve1 = uint112(balance1); + blockTimestampLast = uint32(block.timestamp); + emit Sync(reserve0, reserve1); + } + + function _sqrt(uint256 y) private pure returns (uint256 z) { + if (y > 3) { + z = y; + uint256 x = y / 2 + 1; + while (x < z) { + z = x; + x = (y / x + x) / 2; + } + } else if (y != 0) { + z = 1; + } + } + + function _min(uint256 x, uint256 y) private pure returns (uint256) { + return x < y ? x : y; + } + + /// Mints liquidity for the tokens sent to the pair since the last sync. Called by the router. + function mint(address to) external lock returns (uint256 liquidity) { + (uint112 r0, uint112 r1,) = getReserves(); + uint256 balance0 = IERC20Minimal(token0).balanceOf(address(this)); + uint256 balance1 = IERC20Minimal(token1).balanceOf(address(this)); + uint256 amount0 = balance0 - r0; + uint256 amount1 = balance1 - r1; + if (totalSupply == 0) { + liquidity = _sqrt(amount0 * amount1) - MINIMUM_LIQUIDITY; + _mint(address(0xdead), MINIMUM_LIQUIDITY); + } else { + liquidity = _min(amount0 * totalSupply / r0, amount1 * totalSupply / r1); + } + require(liquidity > 0, "Pair: insufficient liquidity minted"); + _mint(to, liquidity); + _update(balance0, balance1); + emit Mint(msg.sender, amount0, amount1); + } + + /// Burns the liquidity tokens sent to the pair and pays both tokens out pro rata. Called by the router. + function burn(address to) external lock returns (uint256 amount0, uint256 amount1) { + uint256 balance0 = IERC20Minimal(token0).balanceOf(address(this)); + uint256 balance1 = IERC20Minimal(token1).balanceOf(address(this)); + uint256 liquidity = balanceOf[address(this)]; + amount0 = liquidity * balance0 / totalSupply; + amount1 = liquidity * balance1 / totalSupply; + require(amount0 > 0 && amount1 > 0, "Pair: insufficient liquidity burned"); + _burn(address(this), liquidity); + _safeTransfer(token0, to, amount0); + _safeTransfer(token1, to, amount1); + _update(IERC20Minimal(token0).balanceOf(address(this)), IERC20Minimal(token1).balanceOf(address(this))); + emit Burn(msg.sender, amount0, amount1, to); + } + + /// Pays out up to the amounts asked and checks the fee-adjusted product did not fall. The input must already + /// sit in the pair (the router sends it first). + function swap(uint256 amount0Out, uint256 amount1Out, address to) external lock { + require(amount0Out > 0 || amount1Out > 0, "Pair: insufficient output amount"); + (uint112 r0, uint112 r1,) = getReserves(); + require(amount0Out < r0 && amount1Out < r1, "Pair: insufficient liquidity"); + require(to != token0 && to != token1, "Pair: invalid to"); + if (amount0Out > 0) _safeTransfer(token0, to, amount0Out); + if (amount1Out > 0) _safeTransfer(token1, to, amount1Out); + uint256 balance0 = IERC20Minimal(token0).balanceOf(address(this)); + uint256 balance1 = IERC20Minimal(token1).balanceOf(address(this)); + uint256 amount0In = balance0 > r0 - amount0Out ? balance0 - (r0 - amount0Out) : 0; + uint256 amount1In = balance1 > r1 - amount1Out ? balance1 - (r1 - amount1Out) : 0; + require(amount0In > 0 || amount1In > 0, "Pair: insufficient input amount"); + uint256 adjusted0 = balance0 * 1000 - amount0In * 3; + uint256 adjusted1 = balance1 * 1000 - amount1In * 3; + require(adjusted0 * adjusted1 >= uint256(r0) * uint256(r1) * 1000 ** 2, "Pair: K"); + _update(balance0, balance1); + emit Swap(msg.sender, amount0In, amount1In, amount0Out, amount1Out, to); + } + + /// Sends any balance above the reserves to `to`. + function skim(address to) external lock { + _safeTransfer(token0, to, IERC20Minimal(token0).balanceOf(address(this)) - reserve0); + _safeTransfer(token1, to, IERC20Minimal(token1).balanceOf(address(this)) - reserve1); + } + + /// Sets the reserves to the balances. + function sync() external lock { + _update(IERC20Minimal(token0).balanceOf(address(this)), IERC20Minimal(token1).balanceOf(address(this))); + } +} + +/// Creates one pair per unordered token pair and remembers it. +contract IgneumFactory { + mapping(address => mapping(address => address)) public getPair; + address[] public allPairs; + + event PairCreated(address indexed token0, address indexed token1, address pair, uint256 count); + + function allPairsLength() external view returns (uint256) { + return allPairs.length; + } + + function createPair(address tokenA, address tokenB) external returns (address pair) { + require(tokenA != tokenB, "Factory: identical addresses"); + (address token0, address token1) = tokenA < tokenB ? (tokenA, tokenB) : (tokenB, tokenA); + require(token0 != address(0), "Factory: zero address"); + require(getPair[token0][token1] == address(0), "Factory: pair exists"); + bytes32 salt = keccak256(abi.encodePacked(token0, token1)); + pair = address(new IgneumPair{salt: salt}()); + IgneumPair(pair).initialize(token0, token1); + getPair[token0][token1] = pair; + getPair[token1][token0] = pair; + allPairs.push(pair); + emit PairCreated(token0, token1, pair, allPairs.length); + } +} diff --git a/contracts/dex/src/IgneumRouter.sol b/contracts/dex/src/IgneumRouter.sol new file mode 100644 index 000000000..11e320a4e --- /dev/null +++ b/contracts/dex/src/IgneumRouter.sol @@ -0,0 +1,253 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {IgneumFactory, IgneumPair} from "./IgneumPair.sol"; + +interface IERC20Router { + function balanceOf(address) external view returns (uint256); + function transfer(address, uint256) external returns (bool); + function transferFrom(address, address, uint256) external returns (bool); +} + +interface IWIGN is IERC20Router { + function deposit() external payable; + function withdraw(uint256) external; +} + +/// The router in the Uniswap v2 shape: adds and removes liquidity, swaps along a path of pairs, quotes. Pairs are +/// looked up on the factory, never derived from an init-code hash. Devnet 3, test tokens, no value. +contract IgneumRouter { + IgneumFactory public immutable factory; + address public immutable WIGN; + + modifier ensure(uint256 deadline) { + require(deadline >= block.timestamp, "Router: expired"); + _; + } + + constructor(address factory_, address wign_) { + factory = IgneumFactory(factory_); + WIGN = wign_; + } + + receive() external payable { + require(msg.sender == WIGN, "Router: only WIGN"); + } + + // ---- pure maths ---- + + function sortTokens(address tokenA, address tokenB) public pure returns (address token0, address token1) { + require(tokenA != tokenB, "Router: identical addresses"); + (token0, token1) = tokenA < tokenB ? (tokenA, tokenB) : (tokenB, tokenA); + require(token0 != address(0), "Router: zero address"); + } + + function quote(uint256 amountA, uint256 reserveA, uint256 reserveB) public pure returns (uint256 amountB) { + require(amountA > 0, "Router: insufficient amount"); + require(reserveA > 0 && reserveB > 0, "Router: insufficient liquidity"); + amountB = amountA * reserveB / reserveA; + } + + function getAmountOut(uint256 amountIn, uint256 reserveIn, uint256 reserveOut) public pure returns (uint256 amountOut) { + require(amountIn > 0, "Router: insufficient input amount"); + require(reserveIn > 0 && reserveOut > 0, "Router: insufficient liquidity"); + uint256 amountInWithFee = amountIn * 997; + amountOut = amountInWithFee * reserveOut / (reserveIn * 1000 + amountInWithFee); + } + + function getAmountIn(uint256 amountOut, uint256 reserveIn, uint256 reserveOut) public pure returns (uint256 amountIn) { + require(amountOut > 0, "Router: insufficient output amount"); + require(reserveIn > 0 && reserveOut > 0, "Router: insufficient liquidity"); + amountIn = (reserveIn * amountOut * 1000) / ((reserveOut - amountOut) * 997) + 1; + } + + // ---- views ---- + + function pairFor(address tokenA, address tokenB) public view returns (address pair) { + pair = factory.getPair(tokenA, tokenB); + require(pair != address(0), "Router: no pair"); + } + + function getReserves(address tokenA, address tokenB) public view returns (uint256 reserveA, uint256 reserveB) { + (address token0,) = sortTokens(tokenA, tokenB); + (uint112 r0, uint112 r1,) = IgneumPair(pairFor(tokenA, tokenB)).getReserves(); + (reserveA, reserveB) = tokenA == token0 ? (r0, r1) : (r1, r0); + } + + function getAmountsOut(uint256 amountIn, address[] memory path) public view returns (uint256[] memory amounts) { + require(path.length >= 2, "Router: invalid path"); + amounts = new uint256[](path.length); + amounts[0] = amountIn; + for (uint256 i; i < path.length - 1; i++) { + (uint256 reserveIn, uint256 reserveOut) = getReserves(path[i], path[i + 1]); + amounts[i + 1] = getAmountOut(amounts[i], reserveIn, reserveOut); + } + } + + function getAmountsIn(uint256 amountOut, address[] memory path) public view returns (uint256[] memory amounts) { + require(path.length >= 2, "Router: invalid path"); + amounts = new uint256[](path.length); + amounts[amounts.length - 1] = amountOut; + for (uint256 i = path.length - 1; i > 0; i--) { + (uint256 reserveIn, uint256 reserveOut) = getReserves(path[i - 1], path[i]); + amounts[i - 1] = getAmountIn(amounts[i], reserveIn, reserveOut); + } + } + + // ---- liquidity ---- + + function _addLiquidity(address tokenA, address tokenB, uint256 amountADesired, uint256 amountBDesired, uint256 amountAMin, uint256 amountBMin) + private + returns (uint256 amountA, uint256 amountB) + { + if (factory.getPair(tokenA, tokenB) == address(0)) { + factory.createPair(tokenA, tokenB); + } + (uint256 reserveA, uint256 reserveB) = getReserves(tokenA, tokenB); + if (reserveA == 0 && reserveB == 0) { + (amountA, amountB) = (amountADesired, amountBDesired); + } else { + uint256 amountBOptimal = quote(amountADesired, reserveA, reserveB); + if (amountBOptimal <= amountBDesired) { + require(amountBOptimal >= amountBMin, "Router: insufficient B amount"); + (amountA, amountB) = (amountADesired, amountBOptimal); + } else { + uint256 amountAOptimal = quote(amountBDesired, reserveB, reserveA); + require(amountAOptimal <= amountADesired && amountAOptimal >= amountAMin, "Router: insufficient A amount"); + (amountA, amountB) = (amountAOptimal, amountBDesired); + } + } + } + + function addLiquidity( + address tokenA, + address tokenB, + uint256 amountADesired, + uint256 amountBDesired, + uint256 amountAMin, + uint256 amountBMin, + address to, + uint256 deadline + ) external ensure(deadline) returns (uint256 amountA, uint256 amountB, uint256 liquidity) { + (amountA, amountB) = _addLiquidity(tokenA, tokenB, amountADesired, amountBDesired, amountAMin, amountBMin); + address pair = pairFor(tokenA, tokenB); + _pull(tokenA, msg.sender, pair, amountA); + _pull(tokenB, msg.sender, pair, amountB); + liquidity = IgneumPair(pair).mint(to); + } + + function addLiquidityIGN(address token, uint256 amountTokenDesired, uint256 amountTokenMin, uint256 amountIGNMin, address to, uint256 deadline) + external + payable + ensure(deadline) + returns (uint256 amountToken, uint256 amountIGN, uint256 liquidity) + { + (amountToken, amountIGN) = _addLiquidity(token, WIGN, amountTokenDesired, msg.value, amountTokenMin, amountIGNMin); + address pair = pairFor(token, WIGN); + _pull(token, msg.sender, pair, amountToken); + IWIGN(WIGN).deposit{value: amountIGN}(); + require(IWIGN(WIGN).transfer(pair, amountIGN), "Router: WIGN transfer failed"); + liquidity = IgneumPair(pair).mint(to); + if (msg.value > amountIGN) _sendIGN(msg.sender, msg.value - amountIGN); + } + + function removeLiquidity(address tokenA, address tokenB, uint256 liquidity, uint256 amountAMin, uint256 amountBMin, address to, uint256 deadline) + public + ensure(deadline) + returns (uint256 amountA, uint256 amountB) + { + address pair = pairFor(tokenA, tokenB); + _pull(pair, msg.sender, pair, liquidity); + (uint256 amount0, uint256 amount1) = IgneumPair(pair).burn(to); + (address token0,) = sortTokens(tokenA, tokenB); + (amountA, amountB) = tokenA == token0 ? (amount0, amount1) : (amount1, amount0); + require(amountA >= amountAMin, "Router: insufficient A amount"); + require(amountB >= amountBMin, "Router: insufficient B amount"); + } + + function removeLiquidityIGN(address token, uint256 liquidity, uint256 amountTokenMin, uint256 amountIGNMin, address to, uint256 deadline) + external + ensure(deadline) + returns (uint256 amountToken, uint256 amountIGN) + { + (amountToken, amountIGN) = removeLiquidity(token, WIGN, liquidity, amountTokenMin, amountIGNMin, address(this), deadline); + require(IERC20Router(token).transfer(to, amountToken), "Router: token transfer failed"); + IWIGN(WIGN).withdraw(amountIGN); + _sendIGN(to, amountIGN); + } + + // ---- swaps ---- + + function _swap(uint256[] memory amounts, address[] memory path, address to_) private { + for (uint256 i; i < path.length - 1; i++) { + (address input, address output) = (path[i], path[i + 1]); + (address token0,) = sortTokens(input, output); + uint256 amountOut = amounts[i + 1]; + (uint256 amount0Out, uint256 amount1Out) = input == token0 ? (uint256(0), amountOut) : (amountOut, uint256(0)); + address to = i < path.length - 2 ? pairFor(output, path[i + 2]) : to_; + IgneumPair(pairFor(input, output)).swap(amount0Out, amount1Out, to); + } + } + + function swapExactTokensForTokens(uint256 amountIn, uint256 amountOutMin, address[] calldata path, address to, uint256 deadline) + external + ensure(deadline) + returns (uint256[] memory amounts) + { + amounts = getAmountsOut(amountIn, path); + require(amounts[amounts.length - 1] >= amountOutMin, "Router: insufficient output amount"); + _pull(path[0], msg.sender, pairFor(path[0], path[1]), amounts[0]); + _swap(amounts, path, to); + } + + function swapTokensForExactTokens(uint256 amountOut, uint256 amountInMax, address[] calldata path, address to, uint256 deadline) + external + ensure(deadline) + returns (uint256[] memory amounts) + { + amounts = getAmountsIn(amountOut, path); + require(amounts[0] <= amountInMax, "Router: excessive input amount"); + _pull(path[0], msg.sender, pairFor(path[0], path[1]), amounts[0]); + _swap(amounts, path, to); + } + + function swapExactIGNForTokens(uint256 amountOutMin, address[] calldata path, address to, uint256 deadline) + external + payable + ensure(deadline) + returns (uint256[] memory amounts) + { + require(path[0] == WIGN, "Router: invalid path"); + amounts = getAmountsOut(msg.value, path); + require(amounts[amounts.length - 1] >= amountOutMin, "Router: insufficient output amount"); + IWIGN(WIGN).deposit{value: amounts[0]}(); + require(IWIGN(WIGN).transfer(pairFor(path[0], path[1]), amounts[0]), "Router: WIGN transfer failed"); + _swap(amounts, path, to); + } + + function swapExactTokensForIGN(uint256 amountIn, uint256 amountOutMin, address[] calldata path, address to, uint256 deadline) + external + ensure(deadline) + returns (uint256[] memory amounts) + { + require(path[path.length - 1] == WIGN, "Router: invalid path"); + amounts = getAmountsOut(amountIn, path); + require(amounts[amounts.length - 1] >= amountOutMin, "Router: insufficient output amount"); + _pull(path[0], msg.sender, pairFor(path[0], path[1]), amounts[0]); + _swap(amounts, path, address(this)); + IWIGN(WIGN).withdraw(amounts[amounts.length - 1]); + _sendIGN(to, amounts[amounts.length - 1]); + } + + // ---- transfers ---- + + function _pull(address token, address from, address to, uint256 value) private { + (bool ok, bytes memory data) = token.call(abi.encodeWithSelector(IERC20Router.transferFrom.selector, from, to, value)); + require(ok && (data.length == 0 || abi.decode(data, (bool))), "Router: transferFrom failed"); + } + + function _sendIGN(address to, uint256 value) private { + (bool ok,) = to.call{value: value}(""); + require(ok, "Router: IGN send failed"); + } +} diff --git a/contracts/dex/src/TestToken.sol b/contracts/dex/src/TestToken.sol new file mode 100644 index 000000000..55e65103d --- /dev/null +++ b/contracts/dex/src/TestToken.sol @@ -0,0 +1,29 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {ERC20} from "./ERC20.sol"; + +/// A test token with its own faucet: anyone takes DRIP tokens once an hour. Devnet 3, test tokens, no value. +contract TestToken is ERC20 { + uint256 public constant DRIP = 1_000 ether; + uint256 public constant DRIP_INTERVAL = 1 hours; + mapping(address => uint256) public lastDrip; + + event Drip(address indexed to, uint256 value); + + constructor(string memory name_, string memory symbol_) ERC20(name_, symbol_) {} + + /// Mints DRIP tokens to the caller; refused inside DRIP_INTERVAL of the caller's last drip. + function drip() external { + require(block.timestamp >= lastDrip[msg.sender] + DRIP_INTERVAL, "TestToken: one drip an hour"); + lastDrip[msg.sender] = block.timestamp; + _mint(msg.sender, DRIP); + emit Drip(msg.sender, DRIP); + } + + /// Seconds until the caller may drip again (0 when it may). + function dripWait(address who) external view returns (uint256) { + uint256 next = lastDrip[who] + DRIP_INTERVAL; + return block.timestamp >= next ? 0 : next - block.timestamp; + } +} diff --git a/contracts/dex/src/WIGN.sol b/contracts/dex/src/WIGN.sol new file mode 100644 index 000000000..bcae887d0 --- /dev/null +++ b/contracts/dex/src/WIGN.sol @@ -0,0 +1,27 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {ERC20} from "./ERC20.sol"; + +/// Wrapped IGN, the WETH9 shape: one WIGN per IGN held by this contract, minted on deposit and burned on withdraw. +/// Devnet 3, test tokens, no value. +contract WIGN is ERC20("Wrapped IGN", "WIGN") { + event Deposit(address indexed to, uint256 value); + event Withdrawal(address indexed from, uint256 value); + + receive() external payable { + deposit(); + } + + function deposit() public payable { + _mint(msg.sender, msg.value); + emit Deposit(msg.sender, msg.value); + } + + function withdraw(uint256 value) external { + _burn(msg.sender, value); + emit Withdrawal(msg.sender, value); + (bool ok,) = msg.sender.call{value: value}(""); + require(ok, "WIGN: send failed"); + } +} diff --git a/contracts/dex/test/Dex.t.sol b/contracts/dex/test/Dex.t.sol new file mode 100644 index 000000000..9db6fff30 --- /dev/null +++ b/contracts/dex/test/Dex.t.sol @@ -0,0 +1,137 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {Vm, VM_ADDRESS} from "./Vm.sol"; +import {WIGN} from "../src/WIGN.sol"; +import {TestToken} from "../src/TestToken.sol"; +import {IgneumFactory, IgneumPair} from "../src/IgneumPair.sol"; +import {IgneumRouter} from "../src/IgneumRouter.sol"; + +/// The AMM end to end on Foundry's EVM: pair creation, liquidity in and out, token and IGN swaps, the faucet's +/// hour, the k check. Every number is a test number: Devnet 3, test tokens, no value. +contract DexTest { + Vm constant vm = Vm(VM_ADDRESS); + + WIGN wign; + TestToken tta; + TestToken ttb; + IgneumFactory factory; + IgneumRouter router; + address alice = address(0xA11CE); + address bob = address(0xB0B); + + receive() external payable {} + + function setUp() public { + wign = new WIGN(); + tta = new TestToken("Test Token A", "TTA"); + ttb = new TestToken("Test Token B", "TTB"); + factory = new IgneumFactory(); + router = new IgneumRouter(address(factory), address(wign)); + vm.deal(alice, 1_000 ether); + vm.deal(bob, 1_000 ether); + } + + function _drip(address who) internal { + vm.startPrank(who); + tta.drip(); + ttb.drip(); + tta.approve(address(router), type(uint256).max); + ttb.approve(address(router), type(uint256).max); + vm.stopPrank(); + } + + function _seed() internal returns (address pairAB, address pairAW) { + _drip(alice); + vm.startPrank(alice); + router.addLiquidity(address(tta), address(ttb), 500 ether, 500 ether, 0, 0, alice, block.timestamp + 60); + router.addLiquidityIGN{value: 100 ether}(address(tta), 500 ether, 0, 0, alice, block.timestamp + 60); + vm.stopPrank(); + pairAB = factory.getPair(address(tta), address(ttb)); + pairAW = factory.getPair(address(tta), address(wign)); + } + + function test_faucet_drips_once_an_hour() public { + vm.prank(bob); + tta.drip(); + require(tta.balanceOf(bob) == 1_000 ether, "drip amount"); + vm.prank(bob); + vm.expectRevert(bytes("TestToken: one drip an hour")); + tta.drip(); + vm.warp(block.timestamp + 1 hours); + vm.prank(bob); + tta.drip(); + require(tta.balanceOf(bob) == 2_000 ether, "second drip"); + } + + function test_add_liquidity_creates_pairs_and_mints() public { + (address pairAB, address pairAW) = _seed(); + require(pairAB != address(0) && pairAW != address(0) && pairAB != pairAW, "pairs"); + require(factory.allPairsLength() == 2, "two pairs"); + (uint256 rA, uint256 rB) = router.getReserves(address(tta), address(ttb)); + require(rA == 500 ether && rB == 500 ether, "AB reserves"); + (uint256 rT, uint256 rW) = router.getReserves(address(tta), address(wign)); + require(rT == 500 ether && rW == 100 ether, "AW reserves"); + require(IgneumPair(pairAB).balanceOf(alice) == 500 ether - 1000, "LP minus the locked minimum"); + require(IgneumPair(pairAB).balanceOf(address(0xdead)) == 1000, "locked minimum"); + require(wign.balanceOf(pairAW) == 100 ether, "WIGN held by the pair"); + } + + function test_swap_exact_tokens_for_tokens_keeps_k() public { + (address pairAB,) = _seed(); + _drip(bob); + address[] memory path = new address[](2); + path[0] = address(tta); + path[1] = address(ttb); + uint256[] memory quoted = router.getAmountsOut(10 ether, path); + // 10 in at 0.3 percent on 500/500: 9.871580343970612988 out + require(quoted[1] == 9871580343970612988, "quote"); + (uint112 r0b, uint112 r1b,) = IgneumPair(pairAB).getReserves(); + vm.prank(bob); + uint256[] memory amounts = router.swapExactTokensForTokens(10 ether, quoted[1], path, bob, block.timestamp + 60); + require(amounts[1] == quoted[1], "swap matches the quote"); + require(ttb.balanceOf(bob) == 1_000 ether + quoted[1], "bob received"); + (uint112 r0a, uint112 r1a,) = IgneumPair(pairAB).getReserves(); + require(uint256(r0a) * uint256(r1a) >= uint256(r0b) * uint256(r1b), "k did not fall"); + } + + function test_swap_ign_both_ways() public { + _seed(); + address[] memory path = new address[](2); + path[0] = address(wign); + path[1] = address(tta); + uint256 before = tta.balanceOf(bob); + vm.prank(bob); + uint256[] memory amounts = router.swapExactIGNForTokens{value: 1 ether}(0, path, bob, block.timestamp + 60); + require(tta.balanceOf(bob) == before + amounts[1] && amounts[1] > 4.9 ether && amounts[1] < 5 ether, "IGN to TTA"); + path[0] = address(tta); + path[1] = address(wign); + uint256 ignBefore = bob.balance; + vm.startPrank(bob); + tta.approve(address(router), type(uint256).max); + uint256[] memory back = router.swapExactTokensForIGN(amounts[1], 0, path, bob, block.timestamp + 60); + vm.stopPrank(); + require(bob.balance == ignBefore + back[1] && back[1] < 1 ether && back[1] > 0.99 ether, "TTA to IGN"); + } + + function test_remove_liquidity_returns_both_tokens() public { + (address pairAB,) = _seed(); + uint256 lp = IgneumPair(pairAB).balanceOf(alice); + vm.startPrank(alice); + IgneumPair(pairAB).approve(address(router), lp); + (uint256 a, uint256 b) = router.removeLiquidity(address(tta), address(ttb), lp, 0, 0, alice, block.timestamp + 60); + vm.stopPrank(); + require(a == 500 ether - 1000 && b == 500 ether - 1000, "pro rata minus the locked share"); + require(IgneumPair(pairAB).balanceOf(alice) == 0, "LP burned"); + } + + function test_expired_deadline_is_refused() public { + _seed(); + address[] memory path = new address[](2); + path[0] = address(tta); + path[1] = address(ttb); + vm.prank(alice); + vm.expectRevert(bytes("Router: expired")); + router.swapExactTokensForTokens(1 ether, 0, path, alice, block.timestamp - 1); + } +} diff --git a/contracts/dex/test/Vm.sol b/contracts/dex/test/Vm.sol new file mode 100644 index 000000000..930eb7b97 --- /dev/null +++ b/contracts/dex/test/Vm.sol @@ -0,0 +1,22 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +/// The Foundry cheatcodes this project uses, declared here so the tree needs no remote dependency. +interface Vm { + function startBroadcast(uint256 privateKey) external; + function stopBroadcast() external; + function envUint(string calldata name) external view returns (uint256); + function envOr(string calldata name, string calldata defaultValue) external view returns (string memory); + function toString(address value) external pure returns (string memory); + function toString(uint256 value) external pure returns (string memory); + function writeFile(string calldata path, string calldata data) external; + function deal(address who, uint256 newBalance) external; + function prank(address msgSender) external; + function startPrank(address msgSender) external; + function stopPrank() external; + function warp(uint256 newTimestamp) external; + function expectRevert(bytes calldata revertData) external; + function addr(uint256 privateKey) external pure returns (address); +} + +address constant VM_ADDRESS = address(uint160(uint256(keccak256("hevm cheat code")))); From b86e4845b591b5ca45c779ce120c19844d5149a2 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:14:48 +0000 Subject: [PATCH 09/19] DEX: the AMM live on Devnet 3 (factory 0x09FF42db, router 0x9a6fA842, WIGN 0x47447783, TTA, TTB, three seeded pools; docs/contracts/devnet-3.json), the faucet's first drip, the Seed script with the node's own gas estimates (Devnet 3 charges proving gas inside execution gas) Co-Authored-By: Claude Fable 5.1 --- contracts/dex/foundry.toml | 3 +-- contracts/dex/script/Deploy.s.sol | 28 ++++++++++++----------- contracts/dex/script/Seed.s.sol | 37 +++++++++++++++++++++++++++++++ contracts/dex/src/TestToken.sol | 7 ++++-- contracts/dex/test/Dex.t.sol | 4 ++-- contracts/dex/test/Vm.sol | 1 + docs/contracts/devnet-3.json | 24 ++++++++++++++++++++ 7 files changed, 85 insertions(+), 19 deletions(-) create mode 100644 contracts/dex/script/Seed.s.sol create mode 100644 docs/contracts/devnet-3.json diff --git a/contracts/dex/foundry.toml b/contracts/dex/foundry.toml index efe433d0b..947428e34 100644 --- a/contracts/dex/foundry.toml +++ b/contracts/dex/foundry.toml @@ -11,9 +11,8 @@ evm_version = "paris" optimizer = true optimizer_runs = 200 fs_permissions = [{ access = "read-write", path = "./deploy-out.json" }] -# no remote dependencies: the tests and the script declare the cheatcode interface they use (contracts/dex/test/Vm.sol) +# no remote dependencies: the tests and the script declare the cheatcode interface they use (test/Vm.sol); solc 0.8.28 is fetched once by forge auto_detect_remappings = false -offline = true [rpc_endpoints] # the Devnet 3 EVM reaches the build box through a tunnel on this port (never the Mac); see docs/contracts/devnet-3.json diff --git a/contracts/dex/script/Deploy.s.sol b/contracts/dex/script/Deploy.s.sol index a3d733380..6c1b1e508 100644 --- a/contracts/dex/script/Deploy.s.sol +++ b/contracts/dex/script/Deploy.s.sol @@ -38,19 +38,21 @@ contract Deploy { vm.stopBroadcast(); - string memory json = string.concat( - "{\n", - " \"WIGN\": \"", vm.toString(address(wign)), "\",\n", - " \"TTA\": \"", vm.toString(address(tta)), "\",\n", - " \"TTB\": \"", vm.toString(address(ttb)), "\",\n", - " \"IgneumFactory\": \"", vm.toString(address(factory)), "\",\n", - " \"IgneumRouter\": \"", vm.toString(address(router)), "\",\n", - " \"pair_TTA_WIGN\": \"", vm.toString(factory.getPair(address(tta), address(wign))), "\",\n", - " \"pair_TTB_WIGN\": \"", vm.toString(factory.getPair(address(ttb), address(wign))), "\",\n", - " \"pair_TTA_TTB\": \"", vm.toString(factory.getPair(address(tta), address(ttb))), "\",\n", - " \"deployer\": \"", vm.toString(deployer), "\"\n", - "}\n" - ); + string memory json = "{\n"; + json = _line(json, "WIGN", address(wign), ","); + json = _line(json, "TTA", address(tta), ","); + json = _line(json, "TTB", address(ttb), ","); + json = _line(json, "IgneumFactory", address(factory), ","); + json = _line(json, "IgneumRouter", address(router), ","); + json = _line(json, "pair_TTA_WIGN", factory.getPair(address(tta), address(wign)), ","); + json = _line(json, "pair_TTB_WIGN", factory.getPair(address(ttb), address(wign)), ","); + json = _line(json, "pair_TTA_TTB", factory.getPair(address(tta), address(ttb)), ","); + json = _line(json, "deployer", deployer, ""); + json = string.concat(json, "}\n"); vm.writeFile("deploy-out.json", json); } + + function _line(string memory json, string memory key, address value, string memory comma) private pure returns (string memory) { + return string.concat(json, " \"", key, "\": \"", vm.toString(value), "\"", comma, "\n"); + } } diff --git a/contracts/dex/script/Seed.s.sol b/contracts/dex/script/Seed.s.sol new file mode 100644 index 000000000..3befa3a4d --- /dev/null +++ b/contracts/dex/script/Seed.s.sol @@ -0,0 +1,37 @@ +// SPDX-License-Identifier: MIT +pragma solidity ^0.8.24; + +import {Vm, VM_ADDRESS} from "../test/Vm.sol"; +import {TestToken} from "../src/TestToken.sol"; +import {IgneumRouter} from "../src/IgneumRouter.sol"; + +/// Seeds the three pools of an already deployed AMM (the addresses from the environment): a drip of each test +/// token, approvals, 100 IGN beside 500 TTA, 100 IGN beside 500 TTB, 500 TTA beside 500 TTB. Run with +/// --skip-simulation so every gas limit comes from the node's own eth_estimateGas: Devnet 3 charges the proving +/// dimension inside the execution gas, so Foundry's local estimate runs out (8 October 2026, drip() at 133,603 gas: +/// out of gas; the node's estimate 685,513). +/// +/// DEX_TTA=0x.. DEX_TTB=0x.. DEX_ROUTER=0x.. forge script script/Seed.s.sol:Seed --rpc-url devnet3 --broadcast --slow --skip-simulation --sig "run()" +/// +/// Devnet 3, test tokens, no value. +contract Seed { + Vm constant vm = Vm(VM_ADDRESS); + + function run() external { + uint256 key = vm.envUint("DEX_DEPLOYER_KEY"); + address deployer = vm.addr(key); + TestToken tta = TestToken(vm.envAddress("DEX_TTA")); + TestToken ttb = TestToken(vm.envAddress("DEX_TTB")); + IgneumRouter router = IgneumRouter(payable(vm.envAddress("DEX_ROUTER"))); + vm.startBroadcast(key); + if (tta.dripWait(deployer) == 0 && tta.balanceOf(deployer) < 1_000 ether) tta.drip(); + if (ttb.dripWait(deployer) == 0 && ttb.balanceOf(deployer) < 1_000 ether) ttb.drip(); + tta.approve(address(router), type(uint256).max); + ttb.approve(address(router), type(uint256).max); + uint256 deadline = block.timestamp + 1 hours; + router.addLiquidityIGN{value: 100 ether}(address(tta), 500 ether, 0, 0, deployer, deadline); + router.addLiquidityIGN{value: 100 ether}(address(ttb), 500 ether, 0, 0, deployer, deadline); + router.addLiquidity(address(tta), address(ttb), 500 ether, 500 ether, 0, 0, deployer, deadline); + vm.stopBroadcast(); + } +} diff --git a/contracts/dex/src/TestToken.sol b/contracts/dex/src/TestToken.sol index 55e65103d..c58c8f3d8 100644 --- a/contracts/dex/src/TestToken.sol +++ b/contracts/dex/src/TestToken.sol @@ -15,7 +15,8 @@ contract TestToken is ERC20 { /// Mints DRIP tokens to the caller; refused inside DRIP_INTERVAL of the caller's last drip. function drip() external { - require(block.timestamp >= lastDrip[msg.sender] + DRIP_INTERVAL, "TestToken: one drip an hour"); + uint256 last = lastDrip[msg.sender]; + require(last == 0 || block.timestamp >= last + DRIP_INTERVAL, "TestToken: one drip an hour"); lastDrip[msg.sender] = block.timestamp; _mint(msg.sender, DRIP); emit Drip(msg.sender, DRIP); @@ -23,7 +24,9 @@ contract TestToken is ERC20 { /// Seconds until the caller may drip again (0 when it may). function dripWait(address who) external view returns (uint256) { - uint256 next = lastDrip[who] + DRIP_INTERVAL; + uint256 last = lastDrip[who]; + if (last == 0) return 0; + uint256 next = last + DRIP_INTERVAL; return block.timestamp >= next ? 0 : next - block.timestamp; } } diff --git a/contracts/dex/test/Dex.t.sol b/contracts/dex/test/Dex.t.sol index 9db6fff30..2d83e467e 100644 --- a/contracts/dex/test/Dex.t.sol +++ b/contracts/dex/test/Dex.t.sol @@ -84,8 +84,8 @@ contract DexTest { path[0] = address(tta); path[1] = address(ttb); uint256[] memory quoted = router.getAmountsOut(10 ether, path); - // 10 in at 0.3 percent on 500/500: 9.871580343970612988 out - require(quoted[1] == 9871580343970612988, "quote"); + // 10 in at 0.3 percent on 500/500: 10*997*500 / (500*1000 + 10*997) tokens + require(quoted[1] == 9775084808910328058, "quote"); (uint112 r0b, uint112 r1b,) = IgneumPair(pairAB).getReserves(); vm.prank(bob); uint256[] memory amounts = router.swapExactTokensForTokens(10 ether, quoted[1], path, bob, block.timestamp + 60); diff --git a/contracts/dex/test/Vm.sol b/contracts/dex/test/Vm.sol index 930eb7b97..68d0af156 100644 --- a/contracts/dex/test/Vm.sol +++ b/contracts/dex/test/Vm.sol @@ -6,6 +6,7 @@ interface Vm { function startBroadcast(uint256 privateKey) external; function stopBroadcast() external; function envUint(string calldata name) external view returns (uint256); + function envAddress(string calldata name) external view returns (address); function envOr(string calldata name, string calldata defaultValue) external view returns (string memory); function toString(address value) external pure returns (string memory); function toString(uint256 value) external pure returns (string memory); diff --git a/docs/contracts/devnet-3.json b/docs/contracts/devnet-3.json new file mode 100644 index 000000000..a495388de --- /dev/null +++ b/docs/contracts/devnet-3.json @@ -0,0 +1,24 @@ +{ + "network": "igneum-devnet-3", + "chain_id": 4463, + "chain_id_hex": "0x116f", + "note": "Devnet 3, test tokens, no value. The chain may reset; these addresses go with it.", + "rpc": "https://rpc.devnet.igneum.network", + "deployed_at": "2026-10-08T10:50:00Z", + "deployer": "0x07DD4DBca5c1a66755AF28BACCA1D901a2D209aA", + "source": "contracts/dex (Foundry, solc 0.8.28, evm_version paris, optimizer 200 runs)", + "contracts": { + "WIGN": { "address": "0x47447783D00e1760Eb815a34707b611D53f2A51e", "tx": "0x9fc48e0e3aa6ccdc41ff72381e1986d617722e22cd2b5f0f9f1e988eef60b2d6", "block": 26917 }, + "TTA": { "address": "0x9ad4F0435f9172A89C6765eE8631CAB2Db3B5052", "tx": "0xa16b6a6501d2705b7c7dc4cf01f8b65b87270e5f1f90b4638b19f120b4830dc0", "block": 26919, "name": "Test Token A", "faucet": "drip() mints 1,000 TTA to the caller once an hour" }, + "TTB": { "address": "0x996e3042089E80029348d7836472DB34C21841c1", "tx": "0x63eca8238d7f2c29e67c5aad33b0f1cb341d233f2ae0922621e4ba09d94eca61", "block": 26921, "name": "Test Token B", "faucet": "drip() mints 1,000 TTB to the caller once an hour" }, + "IgneumFactory": { "address": "0x09FF42dbb20448ddB673fD34F973a7D93E7139A8", "tx": "0xd25223b5f1adcba415e51a0936c1ba3e9ae1dcbb1299ec5c4a91293e41ace3bb", "block": 26923 }, + "IgneumRouter": { "address": "0x9a6fA842C4e58A87AEF1F3aD15233d99283002B7", "tx": "0xabce2fcac81f87fdca02ac4528d4029ebbe1e6d49304dbe6427e1320f2e17b27", "block": 26924 } + }, + "pairs": { + "TTA_WIGN": { "address": "0x5352Fc2723014Bb45A5b1767Df549DA755773A2B", "seeded": "500 TTA beside 100 IGN", "block": 26969 }, + "TTB_WIGN": { "address": "0x93Db70744e200D647f783C3953dF35d7394EdBC5", "seeded": "500 TTB beside 100 IGN", "block": 26970 }, + "TTA_TTB": { "address": "0xF35775db56dE4f1Cd8Ed7ea7CAb8D017d584a96F", "seeded": "500 TTA beside 500 TTB", "block": 26972 } + }, + "fee": "0.3 percent per swap, kept in the pool", + "gas_note": "Devnet 3 charges the proving dimension inside the execution gas: a local EVM estimate runs out (drip() failed at 133,603 gas in block 26925; the node's eth_estimateGas said 685,513). Send with the node's own estimate (forge script --skip-simulation, or eth_estimateGas before every send)." +} From 0d7ecc643274048aed60ecd3ff2a8277a163188a Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:14:50 +0000 Subject: [PATCH 10/19] Counter ASIC 3.0 status: the founder's load order on the four boxes and its placement (lane D's family sampling, the attack families at scale, lane C's packs, the pin matrix, the cross builds), the drift class settled by a fresh-walk node, the pin candidate e0644958 confirmed, the Windows installer class and the detach fix, the tiers' class-flip pair gated (081b3ba7), the rpc up, two reds on the release line from the first ten minutes of the load order (the missing 60x field; the detach's test literal) with their fixes and defaults Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index 51088a13b..602a7891c 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -465,7 +465,7 @@ Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of | 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | | 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | -The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. THE 0.3.25 PAIRS ON build-1 (the build-server lane, /srv/artefacts/0325-7bd2940f/): the seed pair at 10:34:44 BST (igneumd c7fc542b, igneum-miner 4494ecc4, glibc 2.34), the Windows pair at 10:36:13 (igneumd.exe 5d1dea23, igneum-miner.exe eee7bdfa), the hive package igneum-hive-0.3.25-7bd2940f.tar.gz at 10:37:49 (sha d977797f..., the three kit zips, smoked in ubuntu:20.04); the kit re-cut from 92f004f1 (sha 676240f6, 424,540 B) staged in both folders, the PC 1 host from it bc8d4f79 (in host.sha256 at the shipper's 24680e1d), the 0.3.25 Windows payload from 24680e1d cutting. Every pair of the cut exists by 10:38; the pin's gate set and the crossing are the only waits. FOUR NEW LANES ON THE FOUNDER'S ORDER (11:00 BST, "build all this today to close this gap"), mirrored by the coordinator as the shipper's clocks are: the explorer (a5ef1d5801084005b; explorer.igneum.network by 16:00), the canonical DEX and the Sepolia certificate verifier (a74a8267813d6ea34; the AMM by 14:00, the swap UI by 17:00, the verifier by 20:00), the builder pages, faucet and grants (adb29da59baf27898; /build and /grants by 15:00, the faucet by 16:00), three reference apps that only work on a proven chain (a2060899d2a27d31c; /light by 16:00, /receipt by 18:00, the Sepolia oracle demo by 21:00); the build-server lane stands up rpc.devnet.igneum.network by 12:00; they do not touch the 0.3.25 cut, the crossing or the fleet, sharing the boxes' lease pools (class measure) and the master-only deploy; a lane silent past 25 minutes gets the word from the coordinator and then main. THE FAST-TIME GATE ON THE 0.3.25 PAIR: SUMMARY PASS (cross-0325-39f127a1) at 09:54:40Z (10:54 BST) on the pair 39f127a1 (the node code and object byte for byte e0644958's; igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 09:42:25Z to 09:54:40Z, every check green (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8 at rung 1 at 9,985 bps, the stale node refused, the restart step resynced in 8 s, four sinks equal, 0 PoW rejections); the ceiling's two new fields absent from the 60x file so the ceiling stayed at never there (the node lane's note); to the shipper the same minute; the pin line names e0644958 and its gates. THE FOUNDER'S WORD AT 11:0x BST ("can we add in any more layers? class rotating? things that would render an ASIC useless as soon as it dropped"): the class v6 design opens today as a rotating family, the research lane and the hash lane under the coordinator, the design doc docs/design/class-v6-rotating-family.md by 18:00 BST with the chip-model rows beside each layer (what it does to k and capex for a fixed-function chip and to the per-joule edge for a GPU-like chip; what it costs every GPU tier, Apple included): (1) per-era draws of the class parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement), drawn from chain state like the program; (2) the state-derived dataset's size tracking chain-state growth with a floor, so fixed-memory silicon ages out; (3) scheduled family epochs by height (every 180 days by default) with no release; (4) the (c''') acceptance floor and the F8-form uniformity test generalised to each era's parameter draw, redraw on failure, so layers 1 and 3 need no per-era cryptanalysis. Per layer: the gate it needs (the family analysed as a family: the attack board's shape over the testnet period), the known-failed test, an honest line on what a fully general chip still gets. No consensus code this week; the document, the numbers and the gate plan. Per tier for the founder tonight: what each layer does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE ARC RE-READ IN ITS CHAIN (the hash lane, 10:57 UK): no clear came from the shipper, so the default ran at 10:50: the rotate-fold kit's fetch (sha 65b47211) published to PC 2 at 10:51:41, the run (run-ca3-pc2-v5-intel-bench-20261008, the v5 lane's script c0d398a1) in the publish chain behind another lane's publish-jobs.sh sign --deploy from the build-server worktree (the publisher serialises); the fingerprint line by 11:15 if the publisher frees inside ten minutes, else the blocking process named by 11:10. Queued on PC 1 behind the same publisher: run-ca3-pc1-v5lock-5090-20261008 (class v5 against v4 at unlocked, 1,300 and 1,200 MHz, the v5 kit's CUDA packs), its rows by 12:30; the AMD grid after it from about 11:25. The class-key work on ember-tiers-25 started; the v6 cost rows by 16:00 taken. THE TIERS' CLASS-FLIP STATE, THE UI HALF (the UI lane, 10:57 BST): tiers-class-25 at d949e274 on the mirror, off release-0.3.25's tip 24680e1d (the shipper having merged tiers-25 d3d0704a into release-0.3.25 at 111dae69), the crate gate GREEN 303 + 35 + 8 on build-1, the full pre-push GREEN 60, the UI tests 74 green known-failed first (the v4 tiers stayed on the buttons after the flip on d3d0704a); after the flip the table reads "re-measuring for class v5" on every button with the start minute or "queued (within ten minutes of the crossing)", the v4 watts never current, the strip's sentence naming the crossing; the knee note under the table when knee_loss_pct is over 2 percent; the captures tiers-flip-dark.png and -light.png; the fields tiers_class, program_class, tiers_remeasure_at, knee_loss_pct (the shape sent to the hash lane at 10:4x; the engine sha by 12:30); the default: the display rides 0.3.25 inert if the engine half is late and lights up on 0.3.26. THE FOUNDER'S WORD AT 11:1x BST: class v6 is DECLARED with the four layers as its spine (per-era parameter draws, the dataset tracking chain state, scheduled family epochs by height, the acceptance floor generalised to parameters), and deep past-and-future research opens now under the coordinator with serious resources ("see if anything can be optimised, added or invented"; reading public research is in-house, nothing paid or asked of anyone outside): four research lanes today, (A) history (every ASIC-resistant proof-of-work and how it fell or held: Ethash and the E3 and Linzhi chips, ProgPoW's review, RandomX and its chip analyses, Cuckoo, Equihash and the Z9, Argon2 and Scrypt and the Litecoin chips, KawPow, Autolykos, Octopus, kHeavyHash's chips; the exact mechanism each chip used and what the design missed, each mapped to Igneum's layers with "does v6 close it" as a sentence and a number), (B) the hardware future five years out (PIM and processing-near-memory, HBM3e and HBM4, LPDDR6, 3D DRAM, CXL memory pools, wafer-scale, chiplets, FPGA with HBM; for each the chip-model k band against a state-sized dataset and dependent random reads, and the one layer that would blunt it), (C) invention (layers beyond the four, each a paragraph, a known-failed test and a chip-model row: data-dependent program graphs, latency-bound dependent reads tied to the shard proof, randomised memory topology per era, VRAM-size ratchets, proof-carrying hashes sampled by the pool, time-locked parameter commitments, and what the lane invents; rejecting what costs GPUs more than chips), (D) the family gate (how a parameter family is cryptanalysed as a family: sampling bounds, coverage, the F8-form and (c''') tests over the parameter space, the attack board's shape over the testnet period, so layers 1, 3 and 4 can be automatic with a proof of what was tested). Resources: all four boxes under lease class measure, PC 1 by job for card rows, the rented fleet for one-shot measurements inside the ceiling. Deliverables: a first synthesis in docs/design/class-v6-rotating-family.md by 20:00 BST (the four layers priced, every finding from A to D with its number, a ranked list of what v6 adds beyond the four, the honest line on what a fully general chip still gets), the full report by 09:00 tomorrow, one line to main per lane as each lands; per tier at 20:00: what v6 does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE 0.3.25 APP TIP AND PIN CANDIDATE (the shipper, 10:58 BST): the app tip 9b93e649 (push gate GREEN; the crate unchanged from e0d4425f; the node-source pin to e0644958 and host.sha256 bc8d4f79); the pin candidate the node lane's ceiling cut e0644958 (digest 1b37cb9d, every gate green 10:53, the fast-time SUMMARY PASS 10:54, the floor at DAA 82,800 about 16:53 BST, a publish up to 14:53 without a second cut); the tiers' class-key halves: the UI lane's tiers-class-25 d949e274 green and inert alone, merged with the hash lane's engine sha the moment it lands (12:30), gated as a pair on the release tip, riding only if green by the 13:00 pin; the 0.3.24 Windows take 2 failed at a new place (Inno stopped the app and copied nothing); the update-return lane owns the fix on release-0.3.25 by 12:30, the default the 0.3.25 Windows entry waiting for a clean take 3 while Mac and HiveOS move at the minute. THE FOUR CLASS V6 RESEARCH LANES SPAWNED (the coordinator, 11:0x BST, each with its worktree, its box resources under lease class measure, its clocks and the rules): lane A history (a603a938582c43ab5; the first cut docs/analysis/class-v6/history.md by 15:00), lane B the hardware future (a4f73e2a6f2d1b757; hardware-future.md by 16:00), lane C invention (a5dfe95ee8c47cd0f; invention.md by 17:00), lane D the family gate (a07a99a3788566af2; family-gate.md by 17:00); each feeds the research lane's synthesis docs/design/class-v6-rotating-family.md by 20:00 (its outline by 13:00; the hash lane's per-tier rows by 16:00); the full reports by 09:00 tomorrow; the coordinator's lane mirror carries their clocks. A HELD PUSH AND ITS CAUSE (11:00 BST): the hash lane's push of ca3-v4-amend was refused at 10:58 by the gh-account hook reading the founder's gh (his personal login active again; nothing switched by any lane); the cause is the branch's own hook, which predates the per-process fix (34b0884d): the hook runs the branch's tools/ci, so every branch older than 09:58 must merge the mirror's master before its next push, under which the check reads Igneum's own gh directory and skips under the suspension marker; the rule to every lane. Live: the Arc re-read on PC 2 (published 10:59:47) and the v5lock job on PC 1 (published 10:53, about 12 minutes). THE CLASS V6 OUTLINE ON THE MIRROR (the research lane, docs/design/class-v6-rotating-family.md on counter-asic-4, the commit after fb61ed4b, pushed 10:5x UTC, two hours ahead of 13:00): section 0 the founder's table (per layer, what it does to a fixed-function chip and to a GPU-like chip on its release day, and the 5090, 5070 Ti and M5 Max columns, measured where the night's rows exist, the 5070 Ti scaled until the hash lane's row); the honest frame on top: the four layers render a FIXED-FUNCTION chip useless on the first era its wired value leaves (one tape-out lives one era) and move nothing for the stored-dataset chip with a programmable core except the core's size and the N5 project it forces; that chip keeps 3.6x at zero premium and 2.1x at k = 1 on a 5090 at its knee. The layer table (sections 1 and 2) names the bands each draw takes and the measured rows that set them: the mixer in {4, 8, 16} (x16 open), the op-mix weights within B = 4 with shuffle and mulhi capped (shfl 55.8 pJ per op), the read width in {1, 4} words (w64 excluded by the 5 October rows), the block shape 64 to 256 (never 1,024), N left to the ladder's signal (an unconditional draw retires the Apple tier at 200,000). Open numbers asked of the hash lane with defaults at 16:00: the 5070 Ti row (the rented 5070 scaled), the x16 mixer's verifier and build (the chip model's estimate), two re-weighted shadow packs for the op-mix band (the microbench arithmetic). Layers 2 to 4 and the gate plan are skeletons with their sources named, filling by 18:00 with the four research lanes' cuts, the synthesis by 20:00. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. THE 0.3.25 PAIRS ON build-1 (the build-server lane, /srv/artefacts/0325-7bd2940f/): the seed pair at 10:34:44 BST (igneumd c7fc542b, igneum-miner 4494ecc4, glibc 2.34), the Windows pair at 10:36:13 (igneumd.exe 5d1dea23, igneum-miner.exe eee7bdfa), the hive package igneum-hive-0.3.25-7bd2940f.tar.gz at 10:37:49 (sha d977797f..., the three kit zips, smoked in ubuntu:20.04); the kit re-cut from 92f004f1 (sha 676240f6, 424,540 B) staged in both folders, the PC 1 host from it bc8d4f79 (in host.sha256 at the shipper's 24680e1d), the 0.3.25 Windows payload from 24680e1d cutting. Every pair of the cut exists by 10:38; the pin's gate set and the crossing are the only waits. FOUR NEW LANES ON THE FOUNDER'S ORDER (11:00 BST, "build all this today to close this gap"), mirrored by the coordinator as the shipper's clocks are: the explorer (a5ef1d5801084005b; explorer.igneum.network by 16:00), the canonical DEX and the Sepolia certificate verifier (a74a8267813d6ea34; the AMM by 14:00, the swap UI by 17:00, the verifier by 20:00), the builder pages, faucet and grants (adb29da59baf27898; /build and /grants by 15:00, the faucet by 16:00), three reference apps that only work on a proven chain (a2060899d2a27d31c; /light by 16:00, /receipt by 18:00, the Sepolia oracle demo by 21:00); the build-server lane stands up rpc.devnet.igneum.network by 12:00; they do not touch the 0.3.25 cut, the crossing or the fleet, sharing the boxes' lease pools (class measure) and the master-only deploy; a lane silent past 25 minutes gets the word from the coordinator and then main. THE FAST-TIME GATE ON THE 0.3.25 PAIR: SUMMARY PASS (cross-0325-39f127a1) at 09:54:40Z (10:54 BST) on the pair 39f127a1 (the node code and object byte for byte e0644958's; igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 09:42:25Z to 09:54:40Z, every check green (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8 at rung 1 at 9,985 bps, the stale node refused, the restart step resynced in 8 s, four sinks equal, 0 PoW rejections); the ceiling's two new fields absent from the 60x file so the ceiling stayed at never there (the node lane's note); to the shipper the same minute; the pin line names e0644958 and its gates. THE FOUNDER'S WORD AT 11:0x BST ("can we add in any more layers? class rotating? things that would render an ASIC useless as soon as it dropped"): the class v6 design opens today as a rotating family, the research lane and the hash lane under the coordinator, the design doc docs/design/class-v6-rotating-family.md by 18:00 BST with the chip-model rows beside each layer (what it does to k and capex for a fixed-function chip and to the per-joule edge for a GPU-like chip; what it costs every GPU tier, Apple included): (1) per-era draws of the class parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement), drawn from chain state like the program; (2) the state-derived dataset's size tracking chain-state growth with a floor, so fixed-memory silicon ages out; (3) scheduled family epochs by height (every 180 days by default) with no release; (4) the (c''') acceptance floor and the F8-form uniformity test generalised to each era's parameter draw, redraw on failure, so layers 1 and 3 need no per-era cryptanalysis. Per layer: the gate it needs (the family analysed as a family: the attack board's shape over the testnet period), the known-failed test, an honest line on what a fully general chip still gets. No consensus code this week; the document, the numbers and the gate plan. Per tier for the founder tonight: what each layer does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE ARC RE-READ IN ITS CHAIN (the hash lane, 10:57 UK): no clear came from the shipper, so the default ran at 10:50: the rotate-fold kit's fetch (sha 65b47211) published to PC 2 at 10:51:41, the run (run-ca3-pc2-v5-intel-bench-20261008, the v5 lane's script c0d398a1) in the publish chain behind another lane's publish-jobs.sh sign --deploy from the build-server worktree (the publisher serialises); the fingerprint line by 11:15 if the publisher frees inside ten minutes, else the blocking process named by 11:10. Queued on PC 1 behind the same publisher: run-ca3-pc1-v5lock-5090-20261008 (class v5 against v4 at unlocked, 1,300 and 1,200 MHz, the v5 kit's CUDA packs), its rows by 12:30; the AMD grid after it from about 11:25. The class-key work on ember-tiers-25 started; the v6 cost rows by 16:00 taken. THE TIERS' CLASS-FLIP STATE, THE UI HALF (the UI lane, 10:57 BST): tiers-class-25 at d949e274 on the mirror, off release-0.3.25's tip 24680e1d (the shipper having merged tiers-25 d3d0704a into release-0.3.25 at 111dae69), the crate gate GREEN 303 + 35 + 8 on build-1, the full pre-push GREEN 60, the UI tests 74 green known-failed first (the v4 tiers stayed on the buttons after the flip on d3d0704a); after the flip the table reads "re-measuring for class v5" on every button with the start minute or "queued (within ten minutes of the crossing)", the v4 watts never current, the strip's sentence naming the crossing; the knee note under the table when knee_loss_pct is over 2 percent; the captures tiers-flip-dark.png and -light.png; the fields tiers_class, program_class, tiers_remeasure_at, knee_loss_pct (the shape sent to the hash lane at 10:4x; the engine sha by 12:30); the default: the display rides 0.3.25 inert if the engine half is late and lights up on 0.3.26. THE FOUNDER'S WORD AT 11:1x BST: class v6 is DECLARED with the four layers as its spine (per-era parameter draws, the dataset tracking chain state, scheduled family epochs by height, the acceptance floor generalised to parameters), and deep past-and-future research opens now under the coordinator with serious resources ("see if anything can be optimised, added or invented"; reading public research is in-house, nothing paid or asked of anyone outside): four research lanes today, (A) history (every ASIC-resistant proof-of-work and how it fell or held: Ethash and the E3 and Linzhi chips, ProgPoW's review, RandomX and its chip analyses, Cuckoo, Equihash and the Z9, Argon2 and Scrypt and the Litecoin chips, KawPow, Autolykos, Octopus, kHeavyHash's chips; the exact mechanism each chip used and what the design missed, each mapped to Igneum's layers with "does v6 close it" as a sentence and a number), (B) the hardware future five years out (PIM and processing-near-memory, HBM3e and HBM4, LPDDR6, 3D DRAM, CXL memory pools, wafer-scale, chiplets, FPGA with HBM; for each the chip-model k band against a state-sized dataset and dependent random reads, and the one layer that would blunt it), (C) invention (layers beyond the four, each a paragraph, a known-failed test and a chip-model row: data-dependent program graphs, latency-bound dependent reads tied to the shard proof, randomised memory topology per era, VRAM-size ratchets, proof-carrying hashes sampled by the pool, time-locked parameter commitments, and what the lane invents; rejecting what costs GPUs more than chips), (D) the family gate (how a parameter family is cryptanalysed as a family: sampling bounds, coverage, the F8-form and (c''') tests over the parameter space, the attack board's shape over the testnet period, so layers 1, 3 and 4 can be automatic with a proof of what was tested). Resources: all four boxes under lease class measure, PC 1 by job for card rows, the rented fleet for one-shot measurements inside the ceiling. Deliverables: a first synthesis in docs/design/class-v6-rotating-family.md by 20:00 BST (the four layers priced, every finding from A to D with its number, a ranked list of what v6 adds beyond the four, the honest line on what a fully general chip still gets), the full report by 09:00 tomorrow, one line to main per lane as each lands; per tier at 20:00: what v6 does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE 0.3.25 APP TIP AND PIN CANDIDATE (the shipper, 10:58 BST): the app tip 9b93e649 (push gate GREEN; the crate unchanged from e0d4425f; the node-source pin to e0644958 and host.sha256 bc8d4f79); the pin candidate the node lane's ceiling cut e0644958 (digest 1b37cb9d, every gate green 10:53, the fast-time SUMMARY PASS 10:54, the floor at DAA 82,800 about 16:53 BST, a publish up to 14:53 without a second cut); the tiers' class-key halves: the UI lane's tiers-class-25 d949e274 green and inert alone, merged with the hash lane's engine sha the moment it lands (12:30), gated as a pair on the release tip, riding only if green by the 13:00 pin; the 0.3.24 Windows take 2 failed at a new place (Inno stopped the app and copied nothing); the update-return lane owns the fix on release-0.3.25 by 12:30, the default the 0.3.25 Windows entry waiting for a clean take 3 while Mac and HiveOS move at the minute. THE FOUR CLASS V6 RESEARCH LANES SPAWNED (the coordinator, 11:0x BST, each with its worktree, its box resources under lease class measure, its clocks and the rules): lane A history (a603a938582c43ab5; the first cut docs/analysis/class-v6/history.md by 15:00), lane B the hardware future (a4f73e2a6f2d1b757; hardware-future.md by 16:00), lane C invention (a5dfe95ee8c47cd0f; invention.md by 17:00), lane D the family gate (a07a99a3788566af2; family-gate.md by 17:00); each feeds the research lane's synthesis docs/design/class-v6-rotating-family.md by 20:00 (its outline by 13:00; the hash lane's per-tier rows by 16:00); the full reports by 09:00 tomorrow; the coordinator's lane mirror carries their clocks. A HELD PUSH AND ITS CAUSE (11:00 BST): the hash lane's push of ca3-v4-amend was refused at 10:58 by the gh-account hook reading the founder's gh (his personal login active again; nothing switched by any lane); the cause is the branch's own hook, which predates the per-process fix (34b0884d): the hook runs the branch's tools/ci, so every branch older than 09:58 must merge the mirror's master before its next push, under which the check reads Igneum's own gh directory and skips under the suspension marker; the rule to every lane. Live: the Arc re-read on PC 2 (published 10:59:47) and the v5lock job on PC 1 (published 10:53, about 12 minutes). THE CLASS V6 OUTLINE ON THE MIRROR (the research lane, docs/design/class-v6-rotating-family.md on counter-asic-4, the commit after fb61ed4b, pushed 10:5x UTC, two hours ahead of 13:00): section 0 the founder's table (per layer, what it does to a fixed-function chip and to a GPU-like chip on its release day, and the 5090, 5070 Ti and M5 Max columns, measured where the night's rows exist, the 5070 Ti scaled until the hash lane's row); the honest frame on top: the four layers render a FIXED-FUNCTION chip useless on the first era its wired value leaves (one tape-out lives one era) and move nothing for the stored-dataset chip with a programmable core except the core's size and the N5 project it forces; that chip keeps 3.6x at zero premium and 2.1x at k = 1 on a 5090 at its knee. The layer table (sections 1 and 2) names the bands each draw takes and the measured rows that set them: the mixer in {4, 8, 16} (x16 open), the op-mix weights within B = 4 with shuffle and mulhi capped (shfl 55.8 pJ per op), the read width in {1, 4} words (w64 excluded by the 5 October rows), the block shape 64 to 256 (never 1,024), N left to the ladder's signal (an unconditional draw retires the Apple tier at 200,000). Open numbers asked of the hash lane with defaults at 16:00: the 5070 Ti row (the rented 5070 scaled), the x16 mixer's verifier and build (the chip model's estimate), two re-weighted shadow packs for the op-mix band (the microbench arithmetic). Layers 2 to 4 and the gate plan are skeletons with their sources named, filling by 18:00 with the four research lanes' cuts, the synthesis by 20:00. THE FOUNDER'S WORD AT 11:2x BST ("all builders are idle, load them up"): build-1 to build-4 filled now and kept above 80 percent all day under the lease pool, class measure behind the release gates, in this order of value: (1) the class v6 family gate's sampling runs for lane D (the F8-form census and the (c''') floor over the parameter bands: the mixer {4, 8, 16}, the op-mix weights within B = 4 with shuffle and mulhi capped, the read width {1, 4}, the block 64 to 256; thousands of drawn eras, the uniformity and bucket tests on each, so the family document carries measured coverage tonight); (2) the attack families at scale on the 0.3.25 pin candidate's igneum-pow (F8 to 256 seeds, F9 and F1 to 10^6 on the frozen 1c420786, the day-key scan to 2^28) as the record's strengthening lines; (3) the invention lane's candidate layers measured as packs as fast as it writes them; (4) the full suite matrix of the 0.3.25 pin on every box as the pre-pin check; (5) the Windows and hive cross builds and the sweep's reruns; the lease tool's pre-emption giving release-class work the cores when the pin's gates need them; one line to main at 12:00 with the load on each box and what runs there, then hourly only if a box drops idle. THE DRIFT CLASS SETTLED (the node lane, from the fleet's fresh-walk node, a shared-devnet node synced from an empty datadir to 191,441 chain blocks at 10:03:45Z): hub-1 and five standing boxes number the fresh chain exactly; seven boxes carry numbering inherited from an exec snapshot taken on a chain that later re-walked (+2: p1-4090, p1-a5000, pool-1, build-1's node1; +3: p2-3090-2; +4: p2-3090-4; +5: p1-5090 and p2-3090-3), and a restart on the kept datadir does not re-walk (p1-5090 at 09:22Z stayed +5); the cost: a prover on drifted numbering signs statements the hub vetoes, so the seven earn nothing from proving until they re-walk, the drift refusal stopping the waste. The node lane's word to the fleet: p1-5090 first, both snapshot files moved aside so the executor re-walks from the DAG, the re-walk timed and read against the fresh node, then the other six in series, hub-1 untouched, build-1's node1 after the 12:40Z move; if the re-walk reads over two hours the six wait for the node-side fix on the next node line (the loader re-numbering a resumed range against the DAG before serving). THE 0.3.25 PIN CANDIDATE CONFIRMED (the node lane): e0644958 on both mirrors (keygen, the re-announce, the ceiling switch at 82,800 in the Devnet 3 object, digest 1b37cb9d, the 0.3.24 pin refused both ways), every gate green at 09:53:01Z, the fast-time SUMMARY PASS at 09:54:40Z on the same object; the publish ceiling DAA 75,600 (14:53 BST); waiting only on the 68,400 crossing reading clean (about 12:54; the node lane's line the moment the DAA passes it); the shipper names the pin at 13:00; the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15. LANE C'S FIRST PACK (the invention lane, 11:0x BST by the Mac's clock; its own line read "12:1x", a clock to correct): build-1 takes the igneum-pow build from counter-asic-4 at 5984ffab, then the per-load shadow in its sound form (mx8+shl6912x1: 16 sub-blocks of 432, one pass, the form 20.2a named and never drew) as the first candidate: the acceptance census over 64 seeds and 16 drawn eras against the 16x27 form and the class v4 shape, the pack export, the F8 read at 2^24 and the verifier bench on a leased core, the first read by 13:30; build-2 next for the second candidate (warp-uniform data-dependent block selection); the candidates with no pack form (the VDF commitment, the VRAM ratchet, the pool-sampled witness, the state-tied reads) stay modelled and the 17:00 cut says so; the worktree igneum-wt-v6-invention on class-v6-invention. THE WINDOWS INSTALLER CLASS AND THE 0.3.25 TIP (the shipper, 11:08 BST): the app tip 139c147a (9b93e649 plus install-detach-25 52a34111, packaging/windows and tools/ci only, the crate unchanged; push gate GREEN); the 0.3.24 take 2 class: an installer started under the app's job runner is a child of the engine, and the engine's kill_tree on quit ended it between PrepareToInstall and the copy; the fix re-launches the installer as a one-shot scheduled task outside the job's tree; the 0.3.24 Windows entry skipped; the rule-14 take on PC 2 is the 0.3.25 installer over the running 0.3.21 app, queued ahead of the Arc re-read; the pin candidate e0644958, the DMG 501ba293 staged, the 13:00 pin and the 13:40 provisional minute standing. THE ATTACK FAMILIES AT SCALE (the attack-pass lane, cores held at 11:08 BST, every run under lease pool class measure): box 2 (88 cores): F8 seeds p66 to p257 (192 new, 256 with the gate's p2 to p65) at 2^24 on class v5 at the freeze 1c420786 (the gated binary 0f5c98dc, pairing e5a4ac5978462156; the leaves re-run on 8f481459 if the kit pairing flips), the window-model control, by-site, as three thirds of 64 seeds; box 4 (80 cores held, 16 asked): F9 to 10^6 on 1c420786 (seeds 100,000 to 999,999 in six chunks of 150,000 at 8 threads, four running), F1 to 10^6 class v5 programs on 1c420786 (one census at 40 threads with the progress line and flushed partials, re-drawing the record's first 10^5 on the way as a reproduction check), the F4 day-key scan to 2^28 on 8ca66afa's redraw rule at 8 threads; nothing on build-1 or build-3 (lane D's); the projections: F4 about 2 to 3 hours, F8's 192 seeds about 9 hours, F9's 900,000 and F1's 10^6 about 30 hours each, so the 17:00 default is partials for those two with the lane (d) rows carrying counts so far; any pre-emption by release-class work reported. THE RPC AND THE 0.3.25 PAIRS ON THE PIN CANDIDATE (the build-server lane, 11:0x BST): rpc.devnet.igneum.network up since 11:08 BST (the first of the founder's builder clocks, 52 minutes ahead); the 0.3.25 pairs on e0644958 running on build-1 since 11:06 (the app tip 139c147a), the Windows and hive crosses on build-2, build-3 and build-4 as reproducibility rows at class release by about 12:40; the sweep reruns' list not held by the lane, the default at 12:30: last night's sweep logs on build-1 read for rows that ended without a result line and those rerun at class measure. THE SWEEP RERUNS' LIST (the fleet lane to the build-server lane, 11:1x BST): the fleet ran nothing under the build boxes' lease pool last night (every fleet bench a rented GPU one-shot), so the rows the lease kills cut short are the hash and class lanes' and the build-server lane's default read on build-1 is the right one; the fleet's own rows without a result (A10, A40, A100 40 GB, H100 NVL, H100 PCIe, MI250, RTX 3050, RX 7800 XT, 7900 XT, 7900 XTX, 6900 XT) are provider gaps needing a GPU host, rerun the moment a provider lists one. THE CLASS-FLIP TIERS, BOTH HALVES (the UI lane, 11:13 BST by the Mac's clock, 1 h 47 min inside the 13:00 pin): tiers-class-25 at 081b3ba7 (the display d949e274 plus the hash lane's ember-tiers-25 0a838072, on release-0.3.25's 9b93e649; the field names matched exactly): the crate gate GREEN 305 + 35 + 8 on build-1, the pre-push GREEN 60, the UI tests 78 green known-failed first, the push gate GREEN; with the shipper. A RED ON THE RELEASE TIP, for the shipper and the update-return lane: release-0.3.25's 139c147a is red on one crate test (ota::return_tests::no_relaunch_while_an_installer_runs_and_a_relaunch_when_it_clears, 304 of 305): install-detach's 0c588b09 reshaped the installer's clear step into a multi-line block while the test asserts the one-line literal at app/igneum-app/src/ota.rs:1348; 9b93e649 passes; the fix is the test's literal on the install-detach line; the UI lane built on 9b93e649 so its tip is green alone. A RED FROM THE PIN MATRIX (the CI steward, 11:13 UK): the core suite fails on the pair (the node e0644958 with the app tree 9b93e649): config::params::tests::fast_time_60x_file_is_the_devnet_at_60x panics "override-60x.json lacks the field base_unit_decimals"; the field was added by 0e4ec18a on ca3-v4-node yesterday at 21:45 UK and reached neither master, release-0.3.25 nor the app tip while the node line's test demands it; so every box reads red on core, and the fix is one line on release-0.3.25 (the cherry-pick of 0e4ec18a, or "base_unit_decimals": 8 in infra/fast-time/override-60x.json); sent to the shipper; green so far pow and app on build-1 and build-3; the two new boxes' toolchains read the same as build-1 (Ubuntu 24.04.5, glibc 2.39, the pinned rustc, sccache and lease, no nvcc). The founder's fourth load item paid in its first ten minutes: a red no single-box gate had read. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From 4191027e04d2367149a033eec81c370f1e105ddf Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:16:33 +0000 Subject: [PATCH 11/19] fast-time: base_unit_decimals 8 in override-60x.json, the field the node line's test fast_time_60x_file_is_the_devnet_at_60x demands (0e4ec18a's change, added by hand after its cherry-pick conflicted; the 0.3.25 pre-pin matrix read core RED on every box without it, 8 October 2026, 11:1x UK) Co-Authored-By: Claude Fable 5.1 --- infra/fast-time/override-60x.json | 75 ++++++++++++++++++++++++++++--- 1 file changed, 68 insertions(+), 7 deletions(-) diff --git a/infra/fast-time/override-60x.json b/infra/fast-time/override-60x.json index 9c9a56549..db30b7f86 100644 --- a/infra/fast-time/override-60x.json +++ b/infra/fast-time/override-60x.json @@ -13,8 +13,15 @@ "mass_per_tx_byte": 1, "mass_per_script_pub_key_byte": 10, "mass_per_sig_op": 1000, - "block_mass_limits": { "compute": 500000, "storage": 500000, "transient": 1000000 }, - "block_lane_limits": { "lanes_per_block": 50, "gas_per_lane": 1000000000 }, + "block_mass_limits": { + "compute": 500000, + "storage": 500000, + "transient": 1000000 + }, + "block_lane_limits": { + "lanes_per_block": 50, + "gas_per_lane": 1000000000 + }, "storage_mass_parameter": 1000000000000, "deflationary_phase_daa_score": 0, "pre_deflationary_phase_base_subsidy": 50000000000, @@ -81,19 +88,73 @@ "signing_bonus_activation_daa": 18446744073709551615, "signing_bonus_bps": 1000, "finality_leave_activation_daa": 18446744073709551615, - "latency_ladder": [{"reps": 27, "admissible": true}, {"reps": 35, "admissible": true}, {"reps": 53, "admissible": true}, {"reps": 88, "admissible": false}, {"reps": 173, "admissible": false}, {"reps": 267, "admissible": false}], + "latency_ladder": [ + { + "reps": 27, + "admissible": true + }, + { + "reps": 35, + "admissible": true + }, + { + "reps": 53, + "admissible": true + }, + { + "reps": 88, + "admissible": false + }, + { + "reps": 173, + "admissible": false + }, + { + "reps": 267, + "admissible": false + } + ], "latency_ladder_activation_daa": 18446744073709551615, "latency_ladder_window_daa": 120, "exec_restart_state_root": "", - "emission": {"launch_rate": "3168808781", "ramp_seconds": 2592000, "ramp_start_percent": 10, "step_seconds": 63115200, "step_decay_q32": 2147483648, "tail": {"kind": "cap"}}, + "emission": { + "launch_rate": "3168808781", + "ramp_seconds": 2592000, + "ramp_start_percent": 10, + "step_seconds": 63115200, + "step_decay_q32": 2147483648, + "tail": { + "kind": "cap" + } + }, "proving_consensus_verify_daa": 18446744073709551615, "proving_shard_program_id": "", "proving_aggregator_id": "", "sig_scheme": 0, "sig_scheme_activation_daa": 18446744073709551615, "finality_succession_activation_daa": 18446744073709551615, - "latency_ladder_cache_rung": {"mib": 512, "admissible": false}, + "latency_ladder_cache_rung": { + "mib": 512, + "admissible": false + }, "latency_ladder_cache_rung_activation_daa": 18446744073709551615, - "fees": {"pgas": {"version": 1, "cycles_per_pgas": 1000, "intrinsic_pgas_per_tx": 300, "modexp_base": 10, "modexp_per_byte_numer": 1, "modexp_per_byte_denom": 10}, "block_proving_gas_limit": 120000, "shard_proving_gas_budget": 30000, "min_execution_base_fee_wei": 100000000000, "min_proving_base_fee_wei": 10000000000000, "initial_execution_base_fee_wei": 100000000000, "initial_proving_base_fee_wei": 10000000000000, "base_fee_change_denominator": 8}, - "peer_directory_activation_daa": null + "fees": { + "pgas": { + "version": 1, + "cycles_per_pgas": 1000, + "intrinsic_pgas_per_tx": 300, + "modexp_base": 10, + "modexp_per_byte_numer": 1, + "modexp_per_byte_denom": 10 + }, + "block_proving_gas_limit": 120000, + "shard_proving_gas_budget": 30000, + "min_execution_base_fee_wei": 100000000000, + "min_proving_base_fee_wei": 10000000000000, + "initial_execution_base_fee_wei": 100000000000, + "initial_proving_base_fee_wei": 10000000000000, + "base_fee_change_denominator": 8 + }, + "peer_directory_activation_daa": null, + "base_unit_decimals": 8 } From 93e55f9d435c5f3d5e8fa0797beaf891f5ad1d1b Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:17:50 +0000 Subject: [PATCH 12/19] Class v6 research lane B: the hardware future, five years out (first cut and the full report in one file) docs/analysis/class-v6/hardware-future.md: one table over PIM and PNM (HBM-PIM, AiM, LPDDR5X-PIM, UPMEM), HBM3E, HBM4 and the custom HBM4E base die, fine-grained DRAM on logic, LPDDR6, 3D DRAM, RLDRAM 3 and Folded Banks, FPGA with HBM2e, wafer-scale, CXL, optical I/O, chiplets and UCIe, and SRAM at N2: energy per dependent read, latency, cost per GB, availability to a non-hyperscaler, the k_read and shadow k bands, the edge per joule at zero shadow against the 5090 and the M5 Max, and the one layer of the four that blunts each with a number. The three findings: a 2 GiB SRAM full store on one N2 reticle is the strongest five-year chip (13x to 17x per joule, about USD 0.3 per MH/s) and no layer reaches it (layer 2 moves its capex, not its joules; the shadow at k = 0.5 holds 4.8x); per-bank PIM is blind (1.6 percent of reads in-bank at 2 GiB) and the custom HBM4E base die is the f = 1 chip with its controller in the stack (6.5x to 14x), untouched by layers 1 to 4; the denominator is the wrong card (the M5 Max at 0.78 microjoules halves to thirds every chip edge). Section 5 carries the k bands for the research lane's chip rows; section 7 the four decisions with defaults and deadlines; section 9 the sources with URLs and dates. tools/ci/export-exclude.txt: docs/analysis/class-v6 is research, not public export. Co-Authored-By: Claude Fable 5.1 --- docs/analysis/class-v6/hardware-future.md | 443 ++++++++++++++++++++++ tools/ci/export-exclude.txt | 3 + 2 files changed, 446 insertions(+) create mode 100644 docs/analysis/class-v6/hardware-future.md diff --git a/docs/analysis/class-v6/hardware-future.md b/docs/analysis/class-v6/hardware-future.md new file mode 100644 index 000000000..2ac76e04d --- /dev/null +++ b/docs/analysis/class-v6/hardware-future.md @@ -0,0 +1,443 @@ +# Class v6 research lane B: the hardware future, five years out + +Lane B of the class v6 rotating-family research (the founder's word of 8 October 2026, 11:1x UK: "see if anything can be +optimised, added or invented"). First cut landed 8 October 2026, 11:5x UK; the full report fills the same file. Every +figure carries a label: **measured** (a number read off an instrument in this repository, with the file), **claimed** +(a vendor's or a paper's number, with the URL and the date read), **modelled** (arithmetic on claimed figures by the +method of `docs/analysis/chip-model-v3.md` section 5), **approximate** (from memory or an estimate; the sensitivity is +given). Nothing here is a measurement of a chip. Reading public research is in-house; nothing was paid for or asked of +anyone outside. + +The question, as the coordinator put it: for each memory or packaging line, what does it do to a chip's cost per +dependent random read over a dataset of 1 to 8 GiB that grows with chain state (energy per read, latency, capacity cost +per GB, availability to a non-hyperscaler), what k band does it give the chip five years out, and which ONE of the four +v6 layers (1: per-era parameter draws; 2: the state-sized dataset with a floor; 3: scheduled family epochs; 4: the +(c''') acceptance floor and the F8 uniformity test per era) blunts it, with a number. Where a line beats every layer, +this file says so with the number. + +## 0. One page + +The reads are the hash. Under class v3 the RTX 5090 spends 2.40 microjoules per hash on 128 dependent reads, 18.8 nJ +per read all-in, and the memory system itself spends 2.0 nJ of that (chip-model-v3 5.3, modelled); the card's own +marginal per dependent DRAM read is 10.9 nJ unlocked and 8.7 nJ at the 1,300 MHz lock (measured 8 October 2026, +counter-asic-4-research 15.1a). Every chip in this file is a machine that pays the memory's nanojoule and not the +card's ten, plus whatever the shadow (the program work drawn into the memory wait) forces it to pay at `k` times the +GPU's cost per op. That identity does not change with any technology below; what changes is the memory's nanojoule, +the rate a chip can read at, what a GB costs, and who can buy it. + +**The three findings that change v6's design** + +1. **The strongest five-year chip is not a DRAM chip. It is a 2 GiB SRAM full store on one reticle of merchant N2, and + none of the four layers reaches it.** TSMC N2 reads 38 Mb/mm^2 of SRAM (claimed, IEEE Spectrum, 12 December 2024, + volume in 2025), so 2 GiB is about 452 mm^2 of macro, one die under the 858 mm^2 reticle, roughly USD 400 to 600 of + silicon at a USD 30,000 wafer (approximate). It has no activate ceiling, so its rate is power-bound: about 2,100 + MH/s at 300 W, 0.14 microjoules per hash, **13x to 17x the 5090 per joule at zero shadow and about USD 0.3 per + MH/s** against the 5090's 14.7 and the GDDR7 chip's 2.8 (modelled; the wire energy, 0.5 to 2.0 nJ per read, is the + sensitivity). Layer 2's floor moves its capex, not its joules: at 4 GiB it is two dies, at 8 GiB four, USD 1,000 to + 2,500 of silicon, and its edge per joule falls only from 17x to about 13x because an inter-die hop costs 0.27 nJ + (UCIe, claimed 0.5 pJ/bit). A floor that would blunt it on joules (16 to 32 GiB, eight to sixteen reticles) retires + every honest card under 32 GB first. The only lever that reaches it is the shadow at `k`: with the shadow core on + the same N2 die the ALU band is 0.3 to 0.8 (the record's), and the class v4 premium holds **4.8x at k = 0.5, 2.7x + at k = 1**; at the 5090's whole ALU budget (about 330,000 ops per hash, 575 W) 3.7x and 2.0x. The design change: + layer 1's program-length draw is sized against this chip, not the GDDR7 board, with its lower bound at the shadow + that holds the record's 2.1x today and its upper bound at the honest cards' full latency shadow, re-based every + family epoch (layer 3) on the cards then mining; and the public "2x" line is not reachable in this model against + this chip at any `k` under 1. What slows it is money and time (an N2 project, USD 100 M to 500 M and 18 to 24 + months, approximate), which is the clock of Counter ASIC 3.0 item 4, not a hash property. + +2. **Per-bank processing-in-memory is structurally blind to this hash; the real near-memory threat is the custom + HBM4E base die, and layers 1 to 4 do nothing to it.** HBM-PIM, AiM, LPDDR5X-PIM and UPMEM put a compute unit beside + each bank (or a DPU per 64 MB); a dependent read's next address is uniform over the dataset, so it lands in the same + bank with probability bank bytes over dataset bytes: **1.6 percent at 2 GiB on a 32 MB bank, 0.4 percent at 8 + GiB**, and UPMEM has "no direct communication channel among DPUs" (claimed, the PrIM paper), so the other 98 + percent of reads go to the host. The unit that can follow the chain across banks is a controller on the stack's + base die, which is exactly what TSMC's custom C-HBM4E is: "the custom base die will integrate memory controllers and + PHY" on N3P at 0.75 V, "2x the power efficiency", Micron production 2027, SK hynix HBM4E in 2026 (claimed, + TrendForce, 1 December 2025). That is the `f = 1` chip of the record with its controller moved into the stack: about + 0.9 to 1.0 nJ per read, and HBM4's 32 channels (JEDEC JESD270-4, April 2025) double the activate-bound ceiling per + stack, so **6.5x to 14x per joule at zero shadow** (the ceiling unmeasured, as the record's HBM rows are). The four + layers act on the program, the item map and the capacity; this chip runs any program and holds 36 to 64 GB. What + brakes it until about 2028 is availability (HBM allocated to AI, 20 to 26 week leads, Samsung asking USD 4 to 5 per + Gbit for HBM4 against 1.5 for HBM3E, October 2026) and the shadow at `k`: 4.4x at k = 0.5, 2.6x at k = 1. + +3. **The denominator is the wrong card.** Every chip edge in the record is quoted against the 5090 at 2.40 microjoules. + The Apple M5 Max measured 0.78 microjoules per hash at the GPU-plus-DRAM meter (latency-shadow-2026-10-06, 6 + October 2026), 3.1x the 5090 per joule, and LPDDR6 SoCs (JEDEC JESD209-6, 2025; 14.4 Gbps, 32-byte atoms) are that + tier's next step. Against the M5 Max the GDDR7 chip reads 1.7x, one HBM3 stack 2.4x, the HBM4 base die 3.6x to + 4.4x, the N2 SRAM die about 5x, and with the shadow at k = 0.5 the SRAM die reads about 3x. The honest joule, not + the 5090's, is the chain's resistance, and the same chips are 2x to 3x less frightening against it. The design + change: v6's acceptance floor (layer 4) and the shadow sizing (layer 1) are scored per card tier with the + unified-memory SoC tier as the reference joule, the dataset is kept inside 16 GB unified memory (8 GiB at the top of + the schedule does that), and the public text states the edge over the best honest joule, which is the number a + miner can act on. + +The honest line, in one sentence: five years out a 2 to 8 GiB dataset fits in one to four reticles of merchant SRAM at +USD 500 to 2,500 of silicon, every DRAM line converges on the same 0.5 to 1.0 nJ per read with its controller in the +stack, and no layer of the four touches either; the shadow at the measured `k` band holds 3x to 5x, the honest tier's +own efficiency halves that again, and the clock (project cost against daily issuance) is the wall that is left. + +## 1. The method and the denominators + +| Quantity | Value | Label | Source | +|---|---|---|---| +| The 5090 at class v3 | 136.1 MH/s at 326 W, 2.40 microjoules per hash, 17.5 G dependent reads per second, 415 ns at 256 lanes, a 32-byte sector per 4-byte read | measured | `docs/bench-log.md`; chip-model-v3 5.1 | +| The 5090 per dependent DRAM read, the whole card's marginal | 10.9 nJ unlocked, 8.7 nJ at the 1,300 MHz lock (dram_chase_1g, 18.17 G reads per second) | measured, 8 October 2026 | counter-asic-4-research 15.1a | +| The 5090 L2 hit, per dependent read | 2.4 nJ unlocked, 1.4 nJ at the lock | measured | the same | +| The 5090 per counted ALU op (the class v4 shadow's mix) | 11.3 pJ unlocked, 6.2 pJ at the lock (int_arx); the shadow's own 10.4 and 6.5 | measured | the same; counter-asic-4-research section 1 | +| The 5090 at the 1,300 MHz lock, class v3 | 134.6 MH/s at 223 W, 1.67 microjoules | measured, 7 October 2026 | counter-asic-3-status 7c, the clock grid | +| The class v4 premium `F` on the 5090 | 1.10 microjoules unlocked (145 W), 0.65 at the lock (82 to 88 W) | measured | the same | +| Apple M5 Max, class v3, GPU plus DRAM channels | 27.08 MH/s at 21.0 W, 0.78 microjoules; class v4 at 100,000 ops 1.40 (+16 W); 6.9 pJ per counted op | measured, 6 October 2026 (the SoC's other rails and the wall are not in the 21 W) | `docs/analysis/latency-shadow-2026-10-06.md` section 3 | +| RX 9070 XT, class v3 | about 18.6 MH/s at about 304 W, 10.6 microjoules; 2.4 G reads per second | rate measured, watts approximate | `docs/analysis/horizon/algorithm.md` 5.1 | +| The memory system's own cost per random 32-byte read | GDDR7 2.0 nJ, HBM3 1.2 nJ | modelled on O'Connor et al. 2017 Tables 2 and 3 and Samsung's pJ per bit roadmap | chip-model-v3 5.3 | +| The record's `f = 1` chips at zero shadow | GDDR7 board 166 MH/s at 77.6 W, 0.466 microjoules, 5.1x; one HBM3 stack 83.6 MH/s at 26.8 W, 0.321, 7.5x; eight stacks 666 MH/s at 174 W, 0.262, 9.2x | modelled | chip-model-v3 5.4 to 5.6 | +| The chip's `k` on the shadow's work | ALU 0.3 to 0.8; int8 tile 0.03 to 0.3 (the worse lever); L2 hit 0.1 to 0.3; shuffle 0.4 to 0.7 | the GPU side measured, the chip side claimed | counter-asic-4-research 15.1a | + +The chip's energy per hash is `E = 128 x E_read + P_static / rate + k x F`, the rate the smaller of the memory's +activate-bound ceiling over 128 and the power budget over the per-hash energy, and the edge is the card's microjoules +over the chip's. Two `k` columns appear below: `k_read`, the memory's energy per dependent read over the 5090's measured +10.9 nJ (what the technology itself buys, independent of any program), and the shadow `k` band, the chip's cost per +forced op over the GPU's, which no memory technology changes (it is a logic-node question: a chip core at N2 against a +GPU at N3 or N4 sits at the band's low end, 0.3; at the same node, 0.5 to 0.8; approximate). + +Reads are counted at the part's atom: 32 bytes on GDDR7, HBM3, HBM4 and LPDDR6 (each gives a 32-byte minimum access; +JEDEC, claimed), 64 bytes on an SRAM macro or a line-based part. The class v6 read-width draw (4 to 64 bytes per read +under layer 1) does not move any row: a chip's controller fetches the atom whatever the hash asks for, as the 5090 +fetches its 32-byte sector for a 4-byte load, and the record's w64 reading (the 5090 bandwidth-bound at 71.9 MH/s, +chip-model-v3 5.7) says a wider honest read costs the card and gives the chip nothing. The dataset's size enters only +through capacity cost and die count; its growth (layer 2) enters through the number of dies or stacks a chip must carry +at each family epoch. + +## 2. The table + +Energy is per dependent random read at the part's atom. Latency is the read's own (controller to data), not the GPU's +queueing. Cost per GB is factory-gate where a source gives it, with contract pricing about 2x (Silicon Analysts, October +2026). "Edge" is per joule against the 5090 at class v3 and zero shadow; the bracket is against the M5 Max's 0.78. +"Year" is when a non-hyperscaler could put the part on a board. Every chip-side figure is modelled unless labelled. + +| Technology | Year | Energy per dependent read | Latency | Cost per GB | Availability to a non-hyperscaler | `k_read` (over 10.9 nJ) | Edge at zero shadow, per joule | The layer that blunts it, and by how much | +|---|---|---|---|---|---|---|---|---| +| GDDR7 on a PCB, 16 devices, a 28 nm controller (the record's `f = 1` chip) | now | 2.0 nJ (modelled); 4.5 pJ per bit streaming (claimed, Micron) | about 55 ns controller, tRC about 45 ns | USD 10 (2 GB parts, ending) to 20 to 23 (3 GB parts at USD 60 to 70), September 2026 (claimed, TrendForce) | anyone, through distribution; the 2026 DRAM price cycle (LTAs USD 7.8 to 21 per GB, May 2026, claimed) roughly triples the chip's memory bill against 2025 | 0.18 | 5.1x (1.7x); 166 MH/s at 78 W | none of the four; the shadow at k = 1 holds 2.1x (the record); the dataset's size does not matter because the chip over-provisions capacity for channels (16 devices whatever the dataset) | +| GDDR7 at 36 to 48 Gbps, 4 and 6 GB devices | 2027 to 2028 (claimed, Micron roadmap via Guru3D and OC3D) | the same 2.0 nJ: the activate ceiling and the row energy do not move with the pin rate | the same | per-GB falls, per-channel rises: a random-read chip buys channels, not bytes | anyone | 0.18 | 5.1x | none; denser devices HURT the chip (fewer channels per GB), so this line goes the chain's way | +| HBM3E, one stack, a 28 nm controller on a one-stack interposer | now | 1.2 nJ (modelled); 4.05 pJ per bit streaming (claimed, Samsung) | about 50 ns | USD 8.3 factory gate (USD 300 per 36 GB), about 2x on contract (claimed, Silicon Analysts, October 2026); plus about USD 200 of interposer | Tier-1 volume pricing, 20 to 26 week leads, "supply constrained through 2026" (claimed); a one-stack buyer pays broker prices | 0.11 | 7.5x (2.4x); 84 MH/s at 27 W (ceiling unmeasured, 10.7 G reads per second; 2.3 G at the JEDEC tFAW, which would read 1.8x) | none; the shadow at k = 1 holds 2.4x (the record) | +| HBM4 (JESD270-4, April 2025): 2,048-bit, 32 channels x 2 pseudo-channels, 8 Gbps, 2 TB/s, 36 to 64 GB, VDDQ 0.7 to 0.9 V, a TSMC N12 base die at 0.8 V "1.5x efficiency" (claimed) | 2027 to 2028 for a non-hyperscaler | 1.0 to 1.1 nJ (modelled: the I/O term at 0.8 V against 1.1 V, the 909 pJ row activation unchanged) | about 50 ns | USD 15.3 (USD 550 per 36 GB, October 2026 estimate); Samsung asking USD 4 to 5 per Gbit for 2027 (USD 32 to 40 per GB) (claimed) | allocated to AI accelerators through 2027; "mid-to-high $4 per gigabit" in this month's negotiations (claimed, 2 October 2026) | 0.10 | 4.5x to 11x (1.5x to 3.6x): the 32 channels double the activate-bound ceiling per stack if tFAW is per channel (unmeasured) | none of the four; the shadow at k = 0.5 holds 4.4x, at k = 1 2.6x | +| Custom HBM4E base die (C-HBM4E): the controller and PHY in the stack on N3P at 0.75 V, "2x the power efficiency" (claimed, TrendForce, 1 December 2025); Micron 2027, SK hynix 2026 | 2028 or later for a non-hyperscaler | 0.9 to 1.0 nJ (modelled: no interposer crossing for the controller's traffic) | about 45 ns | HBM4E class, USD 20 to 40, plus a custom base die (USD 50 to 100 per stack, approximate) | by design a per-customer product of the three DRAM makers; the named customers are NVIDIA and Google; a miner-maker of Bitmain's size could commission one (approximate) | 0.09 | 6.5x to 14x (2.1x to 4.4x) | **beats all four layers**; the shadow at k = 0.5 holds 4.4x, at k = 1 2.6x; the shadow core sits on the same N3P base die | +| Fine-grained, hybrid-bonded DRAM on logic (FGDRAM-class 256-byte rows, "tFAW effectively eliminated", 51 percent lower energy per access, 4x bandwidth, claimed, O'Connor et al., MICRO 2017; hybrid-bonded stacks and 3D DRAM on the 2028 to 2030 roadmaps, approximate) | 2029 to 2031 | 0.5 to 0.7 nJ (modelled: 230 pJ activation for a 256-byte row plus about 1.2 pJ per bit of movement over bonded pads) | about 40 ns | HBM class, USD 15 to 40 (approximate) | custom, the three DRAM makers, top customers first | 0.05 | 12x to 20x (4x to 6x); about 125 MH/s per stack at 19 W, 1,000 MH/s at 150 W for eight | **beats all four layers**; the shadow at k = 0.5 holds 5.1x, at k = 1 2.8x | +| Per-bank PIM: Samsung HBM-PIM (a PCU per bank, 2021), SK hynix GDDR6-AiM (16 Gbps, 1.25 V, 2022) and AiMX (32 GB card), Samsung LPDDR5X-PIM (Hot Chips, 25 August 2026, 614 GB/s internal) and LPDDR6-PIM (JEDEC work "substantial progress") | now to 2027 | the host memory's own: 1.8 nJ (HBM2), 2.0 (GDDR6 class); the PIM unit never sees the read it would need | the same | the host part's plus a premium | HBM-PIM and AiM were samples and prototypes; LPDDR5X-PIM mass production "as early as 2027" (claimed) | n/a: no path for a cross-bank dependent read | none beyond the base-die row: the unit serves 1.6 percent of reads at 2 GiB (a 32 MB bank), 0.4 percent at 8 GiB | layer 2, trivially: any dataset over a bank; the number is bank bytes over dataset bytes | +| UPMEM DPU-in-DRAM: 128 DPUs per 8 GB DIMM, 64 MB MRAM per DPU, 350 MHz, MRAM by DMA at most 628 MB/s per DPU, "no direct communication channel among DPUs" (claimed, the PrIM paper, 2021); 1.2 W per 4 Gb chip, 10x DRAM's price at sampling, 1.5x projected (claimed, The Next Platform, 2020) | now | about 0.3 microseconds per 64-byte DMA at 350 MHz (approximate from the paper's alpha-plus-beta model); 2 GiB spans 32 DPUs and every cross-DPU hop is a host round trip of microseconds | microseconds | 1.5x to 10x DDR4 | anyone, in server DIMMs | n/a | none: dead at any dataset over 64 MB | layer 2's floor alone (2 GiB = 32 DPUs with no path between them) | +| LPDDR6 (JESD209-6, 2025): 24-bit channels as two 12-bit sub-channels, 32-byte minimum access, 10.7 to 14.4 Gbps, 4 to 64 Gb dies (claimed, JEDEC via PCWorld and HotHardware) | 2026 to 2028 | 1.5 to 2.0 nJ (approximate: short rows, low-voltage I/O over a PCB, 16 to 32 banks per channel) | about 60 ns | USD 8 to 21 in the 2026 cycle (LTAs, claimed); about USD 3 to 4 in 2024 (approximate) | anyone; it is also the honest SoC tier's memory | 0.15 to 0.18 | about 4x to 5x for a 32-channel controller chip (1.3x to 1.6x against the M5 Max, which already IS this part with a GPU) | none needed: the honest tier converges on it; see finding 3 | +| 3D DRAM (Samsung VS-DRAM 16 layers, VCT 4F2; Neo 3D X-DRAM 230 layers; SK hynix and Micron stacked cells), "from 2030" (claimed, heise, 29 May 2024) | 2030 or later | no vendor claims a latency or activate change; the row energy class is the planar part's (approximate) | the same | lower per GB after 2030 | the three makers | 0.1 to 0.2 | the DRAM rows above | none needed inside five years; it lowers the chip's and the card's capacity cost alike | +| RLDRAM 3 (Micron, 2011): tRC under 10 ns, 16 banks, no activate command, "SRAM-like random access" (claimed, Micron); 576 Mb and 1.125 Gb devices at 2,133 Mb/s (approximate) | now, in small volume | 2 to 4 nJ (approximate: a full small-row access per read; no energy figure found, the datasheet's power calculator exists) | under 10 ns | USD 400 to 800 (USD 50 to 100 per 1.125 Gb device, approximate, a networking part) | anyone, catalogue | 0.2 to 0.4 | about 2 G random reads per second PER DEVICE (16 banks over 8 ns), 32 G for 16 devices: 250 MH/s per board, about 5x per joule (approximate) | layer 2: USD 3,000 to 6,000 of memory at 8 GiB; the line to watch is AMD's "Folded Banks" (ISCA 2025): 8x the activate parallelism in HBM gives 6.7x the irregular bandwidth (claimed, the abstract) | +| FPGA with HBM2e: AMD Alveo V80 (Versal XCV80, 32 GB, 820 GB/s, 190 W, USD 9,495 MSRP, May 2024, claimed), Versal HBM VH1782, Altera Agilex 7 M (two HBM2e stacks, 820 GB/s), Alveo U55C USD 4,747 | now | HBM2 class, 1.8 nJ, but the fabric's controller reaches only the JEDEC tFAW ceiling: 2.3 to 2.4 G reads per second per stack (Shuhai, FCCM 2020; the record) | about 100 ns | USD 300 per GB of card | anyone, 10 to 24 week leads | 0.17 | 0.6x (0.2x): 4.8 G reads per second for two stacks, about 37 MH/s at 150 W, at 5x the 5090's price | none needed; no HBM3E FPGA was found announced (unverified) | +| Wafer-scale SRAM: Cerebras WSE-3, 44 GB SRAM, 21 PB/s, 900,000 cores, 46,225 mm^2 on N5 (claimed, Cerebras and The Next Platform, March 2024); CS-3 about 23 kW and "maybe $2.5 million" (claimed, approximate) | now, by the system | about 7 nJ (approximate): a 512-bit reply crossing about 140 mm of mesh on average at about 0.1 pJ per bit per mm | about 0.5 microseconds across the wafer | USD 50,000 per GB | by the system only | 0.6 | 0.1x to 0.4x: a uniformly random dependent chain is bisection-bound on a 2D mesh (about 110 G reads per second per wafer, up to 500 G with the dataset replicated twenty times), 5 to 22 M reads per second per watt against the 5090's 54 M; 1/1,000 per dollar | none needed; layer 2 removes the replicas as the dataset grows | +| CXL memory pools (CXL 2.0 and 3.x; about 70 ns of controller on top of local DDR5, 100 to 160 ns on Xeon 6 against 75 local, claimed, Introl, February 2026; USD 4 to 7 per GB before the 2026 cycle) | now | DDR5's 2 to 3 nJ plus the link | 170 to 250 ns through a host | USD 4 to 21 | anyone | 0.2 to 0.3 | 0.4x: 500 M 64-byte transactions per second per x8 device at about 25 W (approximate) | none needed; a capacity tool, not a random-read engine | +| Optical I/O: Ayar Labs TeraPHY (UCIe-compliant, about 5 pJ per bit, USD 500 M raised 3 March 2026, claimed); Celestial AI Photonic Fabric (6.2 pJ per bit, about 120 ns round trip, claimed) | 2027 or later | 6x to 8x the interposer's 0.8 pJ per bit per bit moved | plus 120 ns | n/a | n/a | worse than copper for this traffic | no edge: a pooling fabric; the dependent chain wants the memory closer, not farther | none needed | +| Chiplets and die-to-die: UCIe 0.25 to 0.5 pJ per bit (claimed, UCIe via SNIA); BoW 0.5 to 0.7; CoWoS-S USD 600 to 900 per H100-class package, a one-stack interposer about USD 200 (the record) | now | adds 0.27 nJ per 544-bit read that crosses a die boundary | plus 5 to 10 ns per hop | the package: USD 100 to 200 organic with UCIe-S, USD 200 to 900 for 2.5D | UCIe IP from several vendors; CoWoS capacity booked by AI through 2027 (approximate) | n/a | moves the SRAM chip's multi-die rows (below) and lets a 28 nm controller sit beside memory on an organic package | none needed | +| SRAM scaling at 2 nm: TSMC N2 38 Mb/mm^2 HD, +11 percent over N3E (claimed, IEEE Spectrum, 12 December 2024); wafers about USD 30,000, booked to 2028 (claimed, 2026) | now (N2 in volume from 2025) | 1.0 nJ for a 64-byte read from a 452 mm^2 array (approximate; 0.5 to 2.0: the global wire at about 1.3 pJ per bit across a 24 mm die is the term) | 10 to 20 ns | USD 200 to 300 per GB of silicon (one 600 mm^2 die, about USD 400 to 600, holds 2 GiB) | N2 is a merchant node: any customer with a project (Apple, AMD, NVIDIA, MediaTek, Qualcomm; the Bitcoin chip makers are on N3 and N4 class already, approximate) | 0.09 | **13x to 17x (4x to 5x)**: no activate ceiling, power-bound at about 2,100 MH/s per die at 300 W, USD 0.3 per MH/s of silicon; two dies at 4 GiB and four at 8 GiB cost USD 1,000 to 2,500 and read about 13x | **beats layers 1, 3 and 4; layer 2 cuts its capex, not its joules**: one reticle at 2 GiB, two at 4 GiB, four at 8 GiB; the shadow at k = 0.5 holds 4.8x, at k = 1 2.7x; on the M5 Max's joule about 3x at k = 0.5 | + +Density against the schedule: SRAM gains 6 to 11 percent per node every two to three years (N3E +6, N2 +11, claimed), +the dataset doubles every four years (spec 1.13.3: 2 GiB at genesis, 4 GiB at year 4). The SRAM chip loses that race +slowly: its die count doubles each doubling and its per-read energy rises 0.1 to 0.3 nJ per extra hop. It does not lose +it inside five years. + +## 3. What each line does to the four layers, read the other way + +| Layer | What it does to the lines above | Number | +|---|---|---| +| 1. Per-era draws (mixer rounds, op-mix weights, read width, program length, shadow placement) | The read width and the mixer draws move no chip row: every chip in the table stores the dataset and reads the atom. The program length IS the shadow, and it is the only draw that reaches the memory-system chips | at `F` = 1.10 microjoules (class v4) the strongest DRAM chip reads 2.6x to 2.8x at k = 1 and 4.4x to 5.1x at k = 0.5; at the full 5090 shadow (`F` about 2.0) 2.0x and 3.7x | +| 2. The state-sized dataset with a floor | Kills per-bank PIM and UPMEM outright; raises the SRAM chip's die count and capex; does nothing to any DRAM chip, which carries 24 to 64 GB per stack and over-provisions capacity for channels anyway | PIM serves 1.6 percent of reads at 2 GiB, 0.4 at 8; the SRAM chip's silicon USD 500 at 2 GiB, 1,000 at 4, 2,500 at 8, its joules 17x to 13x | +| 3. Scheduled family epochs every 180 days | None of the chips in the table is fixed-function; the controller chip runs any program and the SRAM chip's shadow core is a sequencer | 0 | +| 4. The (c''') floor and the F8 uniformity test per era | Keeps the hot-set cache at the record's 1.067x ceiling; the L2 hit at 2.4 nJ on the 5090 against 0.2 to 0.5 on a chip means the uniformity test is what stops a small cache from being the chip's edge | 1.067x at the ceiling (the record) | +| A fifth: the denominator | The honest tier's own operating point is the resistance: the 5090 at the lock reads 1.67 microjoules, the M5 Max 0.78; every chip edge halves to thirds against them | the SRAM die 5x and the base die 3.6x to 4.4x against the M5 Max at zero shadow; about 3x with the shadow at k = 0.5 | + +## 4. The lines in detail + +### 4.1 Processing-in-memory and processing-near-memory + +Samsung's HBM-PIM (Aquabolt-XL, February 2021) places a programmable computing unit inside each bank of an HBM2 stack +and reports 2.5x system performance and over 60 percent energy saved on a Xilinx Alveo host (claimed, Samsung, 24 +August 2021). SK hynix's GDDR6-AiM (February 2022) adds compute to a 16 Gbps GDDR6 die at 1.25 V, claims up to 16x on +some AI operations and 80 percent less power, and the AiMX card (2023) carries 32 GB of it (claimed, SK hynix). Samsung's +LPDDR5X-PIM (Hot Chips, 25 August 2026) reaches 614 GB/s inside the package against 76.8 GB/s over the external +interface at LPDDR5X-9600, 3x the tokens per second on Llama 3.1 8B, with mass production "as early as 2027" and +LPDDR6-PIM in JEDEC work (claimed, TrendForce and Sammyfans, 25 to 26 August 2026). UPMEM's DPU-in-DRAM ships: 128 +DPUs per 8 GB DIMM, 64 MB per DPU, 350 MHz, MRAM reached only by DMA with a fixed cost plus a per-byte cost and at most +628 MB/s per DPU for 2,048-byte transfers, and "there is no direct communication channel among DPUs" (claimed, the PrIM +characterisation paper, arXiv 2105.03814, read 8 October 2026). The academic line (IMPICA, ICCD 2016: a pointer-chasing +engine on a 3D stack's logic layer, 1.2x to 1.9x and 10 to 41 percent less energy on linked lists, hash tables and +B-trees, claimed) puts the chaser on the base die, not in the bank, for the same reason this hash defeats the bank +units: the next address is anywhere. + +What it does to the chip's cost per dependent read: nothing good for the attacker at the bank level. The PIM unit's +arithmetic is the wrong kind (FP16 SIMD, not 32-bit integer ARX) and the wrong place: with uniform addresses the chain +leaves the bank after one read with probability 1 minus bank bytes over dataset bytes (98.4 percent at 2 GiB on a 32 MB +bank), and in UPMEM's case leaves the DPU with no path but the host. The near-memory version (a chaser with its own +controller on the base die) is the record's `f = 1` chip, and that is where PNM becomes real: see 4.2. + +Blunted by: layer 2, by the dataset's size alone. k: none (no path). + +### 4.2 HBM3E, HBM4 and the custom base die + +JEDEC's JESD270-4 (16 April 2025; read via eeNews Europe, 18 April 2025, and the search summaries, the Business Wire +and All About Circuits pages refusing the fetch) doubles the channel count to 32 with two pseudo-channels each on a +2,048-bit interface at up to 8 Gbps, 2 TB/s per stack, 4 to 16-high stacks of 24 or 32 Gbit dies up to 64 GB, VDDQ 0.7 +to 0.9 V and VDDC 1.0 to 1.05 V (claimed). TSMC builds the standard HBM4 base die on N12 at 0.8 V for "roughly 1.5x" +efficiency and the custom C-HBM4E base die on N3P at 0.75 V for "2x the power efficiency of today's DRAM +manufacturing", and "the custom base die will integrate memory controllers and PHY components typically housed +separately"; Micron targets 2027 production, SK hynix a first tailored HBM4E in the second half of 2026 with 12 nm for +mainstream and 3 nm for NVIDIA and Google premium designs, Samsung 4 nm now and 2 nm for custom HBM (claimed, +TrendForce, 1 December 2025 and 23 January 2026). Prices: HBM2e USD 120 per 16 GB, HBM3 200 per 24, HBM3E 300 per 36, +HBM4 about 550 per 36 (estimate), factory gate, with contract about 2x and 20 to 26 week leads (claimed, Silicon +Analysts, October 2026); Samsung is asking "mid-to-high $4 per gigabit" for 2027 HBM4 against about 1.5 for HBM3E +(claimed, Sammyfans, 2 October 2026). + +What it does to the chip's cost per read: the row activation (909 pJ per 1 KB row, HBM2, O'Connor Table 3) does not +move; the movement and I/O terms fall with the voltage and the base-die node; a 32-channel stack doubles the activate +parallelism the record's HBM rows are bound by (10.7 G reads per second per HBM3 stack, unmeasured; 2.3 G at the JEDEC +tFAW). The arithmetic on the record's method: HBM4 1.0 to 1.1 nJ per read and 21.3 G reads per second per stack (166 +MH/s at about 36 W, 0.22 microjoules, 11x; at the JEDEC-tFAW ceiling 36 MH/s at 19 W, 0.53, 4.5x); the custom base die +0.9 to 1.0 nJ with its controller inside (166 MH/s at 29 W, 0.18, 14x; 6.5x at the low ceiling). The latency stays in +the 45 to 50 ns class; the capacity (36 to 64 GB) is 4x to 8x any floor layer 2 could set without retiring the honest +cards. + +Who can buy it: through 2027 the stacks are allocated to AI accelerators at Tier-1 volume terms; a custom base die is a +per-customer engagement with the DRAM maker. A chip maker of Bitmain's or Canaan's size (the history's rows: tape-outs +on 7 nm-class nodes, R&D in the tens of millions of dollars a year) could commission one from 2028 (approximate); a +USD 5 M startup cannot. The brake is money and queue, not physics, and it expires. + +Blunted by: nothing among the four. The shadow at `k` = 0.5 holds 4.4x, at `k` = 1 2.6x; the shadow core is logic on +the N3P base die and sits at the band's low end against a GPU on an older node. + +### 4.3 LPDDR6 + +JESD209-6 (2025) gives 10,667 to 14,400 MT/s on a 24-bit channel split into two 12-bit sub-channels, a 32-byte minimum +access with 32 and 64-byte bursts, 4 to 64 Gb dies, lower voltages than LPDDR5, a dynamic efficiency mode and on-die +ECC (claimed, JEDEC via PCWorld, HotHardware and MicrocontrollerTips, 2025). The bank count per channel and the tFAW +are not in the public summaries read (unverified); LPDDR5's 16 banks per channel and a tFAW near 20 ns are the +assumption (approximate). A 32-channel, 64-sub-channel controller chip then reads about 12.8 G dependent reads per +second (approximate), 100 MH/s, at 1.5 to 2.0 nJ per read: about 4x to 5x the 5090 per joule, the GDDR7 board's class +at a lower price per channel. The dies are the cheapest random-access memory a non-hyperscaler can buy outside the +2026 price cycle (USD 3 to 4 per GB in 2024, approximate; USD 8 to 21 in 2026 LTAs, claimed). + +The point of this line is not the chip. The M5 Max already is an LPDDR5X part with a GPU, at 0.78 microjoules per hash +measured at the GPU-plus-DRAM meter, 3.1x the 5090 per joule; the Windows-class LPDDR5X SoCs (NVIDIA's and Qualcomm's +desktop parts, approximate) and the LPDDR6 generation after them are the honest tier's floor. A controller chip on the +same memory beats that tier by 1.3x to 1.6x at zero shadow. The resistance of the chain is set by this tier, and v6 +should say so (finding 3). + +Blunted by: none needed. + +### 4.4 3D DRAM + +Samsung's VS-DRAM (VLSI 2023), 16 stacked layers demonstrated against Micron's 8, VCT 4F2 cells as the stepping stone +with prototypes in 2025 and commercial 3D DRAM "by approximately 2030"; Neo Semiconductor's 3D X-DRAM at 230 layers +and 128 Gbit per die as a concept (claimed, heise 29 May 2024, Yole, ComputerBase). No source read claims a latency or +activate-rate change; the gain is capacity per area (about 3x). For a chip that reads 1 to 8 GiB at random, capacity +per die is not the constraint (channels are), so 3D DRAM lowers the honest card's and the chip's cost per GB alike and +changes no row. Outside the five-year edge. + +### 4.5 CXL memory pools + +CXL 2.0 expanders on Xeon 6 measure 100 to 160 ns against 75 ns local DDR5; a CXL 3.1 controller adds about 70 ns; +pooled DDR5 was USD 4 to 7 per GB before the 2026 cycle (claimed, Introl, 1 February 2026, and the search summaries). +A dependent chain through a host CPU and a PCIe-class link at 64 bytes per transaction is bound by the link's +transaction rate (about 500 M per second per x8 device at about 25 W, approximate): 0.4x the 5090 per watt. A capacity +tool. No edge, no layer needed. + +### 4.6 Wafer-scale + +The WSE-3 holds 44 GB of SRAM at 21 PB/s across 900,000 cores on 46,225 mm^2 of N5; the CS-3 draws about 23 kW and +costs "maybe $2.5 million" (claimed, Cerebras and The Next Platform, 14 March 2024; the power from the search summaries). +A 2 GiB dataset spread over the wafer is read by a dependent chain whose next address is uniformly random across 215 +mm of mesh: the traffic is all-to-all, the mesh is bisection-bound, and the average reply crosses about 140 mm of wire. +At about 0.1 pJ per bit per mm (approximate) a 512-bit reply costs about 7 nJ before routers, 3x the GDDR7 board's +2.0; at about 950 links across the bisection at 32 bits per cycle and about 1 GHz (approximate), the wafer completes +about 110 G reads per second, 500 G with the dataset replicated twenty times in regions, 5 to 22 M reads per second per +watt against the 5090's 54 M. Per joule 0.1x to 0.4x, per dollar one thousandth. The wafer is a streaming machine; this +hash is not streaming. No layer needed; layer 2 removes the replicas as the dataset grows. + +### 4.7 Chiplets, interposers and optical I/O + +UCIe gives 0.25 to 0.5 pJ per bit by package type (claimed, UCIe consortium via SNIA SDC 2022 and 2025 pages); BoW 0.5 +to 0.7; CoWoS-S is USD 600 to 900 per H100-class package and a one-stack interposer about USD 200 (the record). A +read that crosses a die boundary pays about 0.27 nJ (544 bits at 0.5 pJ) and 5 to 10 ns, which is what makes the +multi-die SRAM chip of 4.9 cost 1.1 to 1.3 nJ per read instead of 1.0. Optical I/O (Ayar Labs' TeraPHY at about 5 pJ +per bit, UCIe-compliant, a USD 500 M round on 3 March 2026; Celestial AI's Photonic Fabric at 6.2 pJ per bit and about +120 ns round trip; claimed) is 6x to 8x the interposer's energy per bit and adds latency: it pools memory across +packages, which this traffic never wants. No row moves. + +### 4.8 FPGA with HBM + +The Alveo V80 (Versal XCV80, 32 GB HBM2e as two 16 GB stacks, 820 GB/s, 190 W, USD 9,495 MSRP, May 2024) and the U55C +(USD 4,747) are what a non-hyperscaler can buy today with HBM on it; Altera's Agilex 7 M-series carries the same two +HBM2e stacks at 820 GB/s (claimed, AMD, Wccftech, The Next Platform). The record's reading stands: an HBM2 stack under a +soft controller reaches the JEDEC tFAW ceiling, 2.3 to 2.4 G random reads per second (Shuhai, FCCM 2020, Figure 7), +so two stacks give about 4.8 G, about 37 MH/s at about 150 W (approximate), 0.6x the 5090 per joule at 5x its price. No +HBM3E FPGA was found announced in the pages read (unverified). No layer needed. + +### 4.9 SRAM at 2 nm: the full store on one reticle + +TSMC's N2 reads 38 Mb/mm^2 of high-density SRAM, 11 percent over N3E (claimed, IEEE Spectrum, 12 December 2024; +volume from 2025); wafers are about USD 30,000 and "booked to 2028" (claimed, tech-insider, 2026). The arithmetic: 2 +GiB is 17,180 Mbit, 452 mm^2 of macro; with periphery, a controller, the lanes' registers and a shadow core, a 550 to +650 mm^2 die under the 858 mm^2 reticle; about 95 gross dies per wafer, 55 to 70 percent good with row and column +repair (approximate), USD 400 to 600 of silicon. No DRAM, no interposer, an organic package. Under class v5 the +dataset's items are leaves of the chain state refreshed per window; the chip rewrites 2 GiB per window at on-die +bandwidth, the same 32 ms every GPU pays (the record, chip-model-v3 5.10), and holds a node or shares one across a farm +as the record prices. + +Energy per read: the macro's 64-byte read, about 0.1 nJ (approximate), plus the global wire. The record took 0.6 pJ per +bit of wire across a 128 mm^2 array; scaling with the side of the die gives about 1.3 pJ per bit across 600 mm^2, 0.68 +nJ for 512 bits, so about 1.0 nJ per read with the controller, range 0.5 to 2.0 (the sensitivity of every SRAM figure +here). Latency 10 to 20 ns. No activate ceiling, no tFAW, no refresh: the rate is power-bound. At 300 W with 30 W of +static and controller power: 270 W over 128 nJ per hash is about 2,100 MH/s per die, 0.14 microjoules per hash, 17x +the 5090 at zero shadow (13x at 1.3 nJ, 8x at 2.0, 30x at 0.5); USD 0.25 per MH/s of silicon, about 0.4 with the +board. Two dies at 4 GiB: half the reads cross one UCIe hop, 1.14 nJ, 15x, USD 1,000. Four dies at 8 GiB: 1.3 nJ, 13x, +USD 2,000 to 2,500, a 4-die organic package. The honest cards hold 8 GiB fine (16 GB and up); a floor that pushed the +dataset past what a package can hold (16 to 32 GiB, eight to sixteen reticles at USD 5,000 to 10,000 and 1.5 to 2.0 +nJ per read, still 7x to 9x) would retire every honest card under 32 GB first. Layer 2 therefore sets this chip's +capex and die count, not its joules, and the floor's number is a card-lifetime decision, not a chip decision. + +The project: the history's IBS figures (5 nm USD 416 M to 542 M, 3 nm 590 M) price an SoC; an SRAM array with a +controller and a sequencer core is simpler and the startup figure ("$50M to $75M" for 7 nm, SemiAnalysis) is the +better guide, so USD 100 M to 500 M at N2 (approximate) and 18 to 24 months to a first chip (approximate). That is the +clock of Counter ASIC 3.0 item 4 and the daily-issuance threshold of the history (chips at USD 20 K to 50 K of daily +issuance for compute-bound hashes; 32 months for Ethash at the largest prize): the SRAM chip arrives when the prize +pays for an N2 project, and nothing in the hash moves that date. + +Blunted by: layer 2 on capex only (USD 500 to 2,500 across the floor's range); the shadow at `k` on joules: with the +shadow core on the same N2 die against a GPU on N3 or N4 the ALU band sits at 0.3 to 0.5; at `F` = 1.10 the chip reads +4.8x at k = 0.5 and 2.7x at k = 1; at the 5090's whole latency shadow (about 330,000 ops per hash, `F` about 2.0, +575 W) 3.7x and 2.0x; against the M5 Max's joule about 3x at k = 0.5. + +### 4.10 RLDRAM and the activate-free line + +RLDRAM 3 (Micron, 2011; ISSI second-sourced) has a tRC under 10 ns, 16 banks per device, no separate activate +command and "SRAM-like random access" (claimed, Micron's product page and the 2011 announcements), in 576 Mb and 1.125 +Gb devices at up to 2,133 Mb/s (approximate). One device completes about 2 G random reads per second (16 banks over 8 +ns), six times a GDDR7 device's share of the 5090 board's 21.3 G; sixteen devices, 2 GiB, about 32 G reads per second, +250 MH/s per board. The energy per read is not published in anything read (a power calculator exists); a full +small-row access per read at an old node reads 2 to 4 nJ (approximate), so about 5x per joule, at USD 400 to 800 per +GB (approximate, a low-volume networking part). It is the proof that an activate-free DRAM exists, and AMD's "Folded +Banks" (ISCA 2025, with AMD Research: 8x the activate parallelism in a 3D-stacked HBM gives 6.7x the irregular +bandwidth, claimed from the abstract; the PDF refused the fetch) is the same idea on the HBM roadmap. If a DRAM maker +ships it in HBM, the activate ceilings in the record's HBM rows rise 6x to 8x and the HBM chip's rate per stack with +them; its energy per read falls by the row-size term (O'Connor's FGDRAM: 51 percent). That is the DRAM-on-logic row. + +Blunted by: layer 2 on RLDRAM's capacity cost (USD 3,000 to 6,000 at 8 GiB); nothing on the HBM version. + +## 5. The k bands for the research lane's chip rows + +For `docs/design/class-v6-rotating-family.md`. The shadow `k` is the chip core's energy per forced op over the GPU's +at the same operating point (the record's ALU band 0.3 to 0.8 on the 5090's measured 6.2 to 11.3 pJ per op); the +memory `k_read` is the technology's energy per dependent read over the 5090's measured 10.9 nJ. "Edge" is per joule at +zero shadow; "with the shadow" is at the class v4 premium `F` = 1.10 microjoules on the 5090 (`E_chip` = `E_mem` + `k F`). + +| Chip row | Year | `E_read` nJ | Rate per chip, MH/s | `E_hash` at zero shadow, microjoules | Edge over the 5090 (2.40) | Edge over the M5 Max (0.78) | `k_read` | Shadow `k` band | With the shadow, k = 0.5 / 1 | Silicon and memory, USD per MH/s | +|---|---|---|---|---|---|---|---|---|---|---| +| GDDR7 board, 28 nm controller (the record) | now | 2.0 | 166 | 0.466 | 5.1x | 1.7x | 0.18 | 0.5 to 0.8 (28 nm core: the high end) | 3.3x / 2.1x | 2.8 (2025 memory), about 7 in the 2026 cycle | +| HBM3E, one stack | now | 1.2 | 84 (ceiling unmeasured) | 0.321 | 7.5x | 2.4x | 0.11 | 0.3 to 0.8 | 3.8x / 2.4x | 6.6 | +| HBM3E, eight stacks | now | 1.2 | 666 | 0.262 | 9.2x | 3.0x | 0.11 | 0.3 to 0.8 | 4.1x / 2.5x | 4.0 | +| HBM4, one stack, N12 base die | 2027 to 2028 | 1.0 to 1.1 | 166 (36 at the JEDEC tFAW) | 0.22 (0.53) | 11x (4.5x) | 3.6x (1.5x) | 0.10 | 0.3 to 0.8 | 4.4x / 2.6x | about 5 | +| Custom HBM4E base die, N3P, controller in the stack | 2028 or later | 0.9 to 1.0 | 166 (36) | 0.18 (0.37) | 14x (6.5x) | 4.4x (2.1x) | 0.09 | 0.3 to 0.5 (an N3P core) | 4.4x / 2.6x | about 5 | +| DRAM on logic, FGDRAM-class rows, hybrid bonded | 2029 to 2031 | 0.5 to 0.7 | 125 per stack, 1,000 for eight | 0.15 | 16x (12x to 20x) | 5.2x | 0.05 | 0.3 to 0.5 | 5.1x / 2.8x | about 4 (approximate) | +| SRAM full store, one N2 reticle, 2 GiB | 2027 to 2028 (an N2 project) | 1.0 (0.5 to 2.0) | about 2,100 at 300 W | 0.14 | 17x (8x to 30x) | 5.6x | 0.09 | 0.3 to 0.5 (an N2 core) | 4.8x / 2.7x | 0.25 to 0.4 | +| SRAM full store, two N2 dies, 4 GiB | the same | 1.14 | about 1,850 | 0.16 | 15x | 4.9x | 0.10 | 0.3 to 0.5 | 4.6x / 2.7x | 0.55 | +| SRAM full store, four N2 dies, 8 GiB | the same | 1.3 | about 1,600 | 0.185 | 13x | 4.2x | 0.12 | 0.3 to 0.5 | 4.5x / 2.6x | 1.3 to 1.6 | +| LPDDR6 controller chip, 32 channels | 2027 | 1.5 to 2.0 | about 100 | 0.25 to 0.30 | 4x to 5x | 1.3x to 1.6x | 0.15 to 0.18 | 0.3 to 0.8 | 3.1x / 2.0x | about 3 (approximate) | +| Per-bank PIM, UPMEM, FPGA with HBM2e, wafer-scale, CXL, optical | | | | | under 1x or no path | | | | | | + +Arithmetic, the SRAM row: 270 W over (128 x 1.0 nJ) = 2.11 G hashes per second; 300 W over 2.11 G = 0.142 +microjoules; 2.40 over 0.142 = 16.9x; with the shadow at k = 0.5: (0.142 + 0.55) over (2.26 + 1.10) = 0.692 over 3.36 = +4.86x; at k = 1: 1.242 over 3.36 = 2.71x. The base-die row: 166 M x 128 x 0.95 nJ = 20.2 W plus 4 static plus 5 +controller = 29.2 W; 29.2 over 166 M = 0.176 microjoules; 2.40 over 0.176 = 13.6x. The M5 Max column divides 0.78 by +the same `E_hash`. Every chip-side figure is modelled; the GPU-side figures are the record's measurements. + +## 6. Consequences per tier + +| Tier | What this file means | What is being done | +|---|---|---| +| Home miner, one 8 GB card | Nothing changes today: no chip exists, and the first one in this file (an N2 SRAM die or an HBM4 base-die chip) is a USD 100 M-class project with a 2028-class date. When one lands it runs at 0.14 to 0.22 microjoules per hash against this card's 10 to 20; this tier is the first out, as the record says. The dataset's floor decides this tier's life more than any chip does: 4 GiB at year 4 (the spec's schedule) retires it then | the shadow sizing and the floor are the founder's numbers to set (section 7); the share-pattern detector (Counter ASIC 3.0 item 4) is what tells this miner a chip has arrived | +| One 16 GB card (9070 XT class) | Holds 8 GiB with room; AMD's 2.4 G reads per second at 304 W is 7x behind the 5090 per joule and 50x to 100x behind the chips here | the vendor-share metric; nothing in the hash moves AMD's dependent-read rate | +| One 24 or 32 GB card, the 5090 at the lock | 1.67 microjoules at the 1,300 MHz lock; the chips here are 8x to 12x ahead per joule at zero shadow and 2.5x to 4x with the class v4 shadow at k = 0.5 | the Ember knob carries the lock rows; the shadow's upper bound at the full ALU budget is the lever this file sizes | +| The unified-memory SoC (M5 Max, LPDDR5X and LPDDR6 desktops) | 0.78 microjoules at the GPU-plus-DRAM meter: the honest tier the chips beat least (1.7x to 5.6x at zero shadow, about 3x with the shadow). Capex-poor (27 MH/s per USD 4,000 machine) but joule-rich | finding 3: v6 scores the floor and the shadow per tier with this tier as the reference joule; the dataset stays inside 16 GB unified memory | +| A rig | A rig's cost is electricity; against an N2 SRAM chip at USD 0.3 per MH/s and 0.14 microjoules it earns 1/10 to 1/17 of a chip per watt and leaves when chips hold the hashrate | the issuance trigger: the bounty and the benchmark live before daily issuance crosses about USD 50 K (the record) | +| A pool user | A chip fleet is a few operators; the share-pattern detector is the warning | the detector on the observer, before the public testnet (the record) | +| The public claim | "Under 2x" is not reachable against any chip in this file at a `k` under 1. The honest sentence is: the strongest chip five years out beats a 5090 by 2.6x to 2.8x per joule at k = 1 and about 4.5x at k = 0.5 with the shadow on, and the best honest SoC by about 3x; and it costs an N2 project | the research lane's synthesis carries the number; nothing from this file goes to the site or the devnet | + +## 7. Decisions this raises for the founder + +Each carries a default and a deadline; silence means the default. + +1. **Size the shadow against the SRAM chip, not the GDDR7 board.** Layer 1's program-length draw gets a lower bound at + the length that holds the record's 2.1x on the GDDR7 chip today and an upper bound at the honest cards' full latency + shadow (the 5090's about 330,000 ops unlocked, about 150,000 at the lock; the M5 Max about 290,000; the 9070 XT + about 650,000; approximate from the record), re-based at every family epoch on the cards then mining. Default: the + research lane writes the draw with these bounds into the synthesis; the hash lane measures the rate and the watts at + the upper bound on the 5090 and the M5 Max before the first v6 era is cut. Deadline: 20:00 UK today (the synthesis). +2. **The floor (layer 2).** 2 GiB at genesis is one N2 reticle; 4 GiB two; 8 GiB four. The floor moves the chip's + capex (USD 500 to 2,500), not its joules (17x to 13x), and 8 GiB is the last size inside 16 GB unified memory and + the 16 GB card tier. Default: the spec's schedule stands (2 GiB plus 0.5 GiB a year, doubling at year 4), chosen on + card lifetime; the floor is not a chip lever and this file does not ask to raise it. Deadline: none; a note in the + synthesis. +3. **The reference joule (a fifth layer, or a rule).** Resistance is stated against the best honest joule (the + unified-memory SoC tier, then the 5090 at the lock), not the unlocked 5090. Default: the research lane adopts it in + the synthesis's chip rows (section 5's M5 Max column) and the public text, when there is one, carries the edge over + the best honest joule. Deadline: 20:00 UK today. +4. **The clock.** The chips in this file are USD 100 M-class projects with 2028-class dates; the issuance trigger, the + benchmark and the share-pattern detector of Counter ASIC 3.0 item 4 are what decide when they are built. Default: + unchanged from the record. Deadline: none. + +## 8. Unverified and owed + +- Every chip-side energy figure is modelled on the record's method (O'Connor's HBM2 breakdown, Samsung's pJ per bit + roadmap, the 909 pJ row activation standing in for HBM3, HBM4 and GDDR7); the SRAM wire figure (0.6 pJ per bit at 128 + mm^2, scaled with the die's side) is from memory and moves the SRAM rows by 2x either way; the UCIe hop (0.5 pJ per + bit) is the consortium's claim; the FGDRAM row size (256 bytes) and its 51 percent are the paper's simulation. +- HBM4's banks per pseudo-channel and tFAW per channel are behind the JEDEC paywall (the Business Wire and All About + Circuits pages refused the fetch); the "2x the activate ceiling" rests on 32 channels with tFAW per channel, as the + record's HBM rows rest on the same assumption at 16; both unmeasured. The AWS F2 hour the record names (chip-model-v3 + 5.3) is still the one measurement that would settle the HBM2 figure, and an HBM3 or HBM4 part is not rentable at the + controller level by anyone outside a hyperscaler today. +- LPDDR6's bank count and tFAW are not in the summaries read; the row is LPDDR5's structure (approximate). +- RLDRAM 3's energy per read and its 2026 price are not published in anything read; the row is approximate. +- The Cerebras mesh figures (link width, clock, bisection) are approximate; the fabric bandwidth was not on the pages + read (The Next Platform gives only its change against WSE-2). +- Dates: the DRAM-on-logic and hybrid-bonded HBM timelines ("2028 to 2031") are approximate; the sources read give + 3D DRAM "from 2030" and C-HBM4E production in 2027; the hybrid-bonding pages were not fetched (the TrendForce page + refused). The N2 project cost and time are approximate. +- The web-search budget of this session ran out at about 11:3x UK after 26 searches; the remaining facts were read by + direct fetch of the pages named in section 9. Pages that refused (403): JEDEC's Business Wire release, All About + Circuits, ACM's Folded Banks page, OC3D's GDDR7 capacity note, All About Circuits' HBM-PIM note. Their figures are + carried from the search summaries and marked claimed. +- Nothing was run on the Mac; nothing was built or benchmarked anywhere. The one measurement this file would want next + is on the record's queue already: the 5090 and the M5 Max at the shadow's upper bound (decision 1). + +## 9. Sources (URL and the date read; all read 8 October 2026 unless a file is named) + +Repository: `docs/analysis/chip-model-v3.md` (sections 5.1 to 5.11, 6); `docs/analysis/counter-asic-4-research.md` +on branch counter-asic-4 at fb61ed4b (sections 1, 15.1a, 20.4); `docs/analysis/asic-resistance-history.md` (2.5, 2.6); +`docs/plans/counter-asic-3-status.md` (7c, the clock grid); `docs/spec/01-lottery-hash.md` (1.8, 1.13.3, 1.14); +`docs/analysis/latency-shadow-2026-10-06.md`; `docs/analysis/sram-mirror.md`. + +- JEDEC HBM4 (JESD270-4, 16 April 2025): https://www.eenewseurope.com/en/hbm4-standard-doubles-channel-count-for-ai-boost (18 April 2025); https://hothardware.com/news/jedec-finalizes-hbm4-spec; https://www.businesswire.com/news/home/20250416843598/en (refused) +- TSMC HBM4 and C-HBM4E base dies: https://www.trendforce.com/news/2025/12/01/news-tsmc-unveils-custom-c-hbm4e-details-n3p-logic-dies-reportedly-target-2x-efficiency-gain/ (1 December 2025); https://www.trendforce.com/news/2026/01/23/news-samsungs-custom-hbm4e-design-reportedly-aimed-for-mid-2026-parallels-sk-hynix-and-micron/ (23 January 2026) +- HBM prices: https://siliconanalysts.com/data/hbm-pricing (October 2026); https://www.sammyfans.com/2026/10/02/samsung-seeks-more-than-3x-hbm3e-pricing-for-hbm4/ (2 October 2026); https://www.trendforce.com/news/?p=62555 +- GDDR7 prices and roadmap: https://www.trendforce.com/news/2026/09/24/news-micron-reportedly-ends-2gb-gddr7-narrowing-supply-options-for-nvidias-rtx-50-series/ (24 September 2026); https://www.guru3d.com/story/micron-fiveyear-roadmap-shows-24gb-36gbps-gddr7-in-2026; https://overclock3d.net/?p=314548 (refused; the 4 and 6 GB parts in 2027 to 2028 from the search summary) +- DRAM prices 2026: https://wccftech.com/mobile-dram-prices-expected-to-increase-by-100-quarter-over-quarter-as-long-term-agreements-now-getting-signed-at-prices-as-high-as-21-gb/ (4 May 2026); https://tech-insider.org/ddr5-ram-prices-2026/ +- Samsung HBM-PIM: https://news.samsung.com/global/samsung-brings-in-memory-processing-power-to-wider-range-of-applications (24 August 2021); https://www.allaboutcircuits.com/news/beyond-high-bandwidth-memory-samsung-breaks-processing-in-memory-into-AI-applications/ (refused) +- SK hynix AiM and AiMX: https://news.skhynix.com/developed-processing-in-memory (2022); https://www.hc2024.hotchips.org/assets/program/conference/day1/11_HC2024.SKhynix.GuhyunKim.rev920240822.pdf +- Samsung LPDDR5X-PIM and LPDDR6-PIM: https://www.sammyfans.com/2026/08/25/samsung-unveils-lpddr5x-pim-dram/ (25 August 2026); https://www.trendforce.com/news/2026/08/26/news-samsungs-4nm-gaia-could-mark-first-pim-commercialization-in-ai-pcs-mass-production-as-early-as-2027/ (26 August 2026); https://en.fnnews.com/news/202609060934024282 +- UPMEM: https://arxiv.org/pdf/2105.03814 (the PrIM characterisation, text extracted with pdftotext: Table 1, section 3.2, the inter-DPU statement); https://www.nextplatform.com/2020/02/04/putting-in-memory-processing-through-the-paces/ (4 February 2020); https://old.hotchips.org/hc31/HC31_1.4_UPMEM.FabriceDevaux.v2_1.pdf +- IMPICA: https://ghose.cs.illinois.edu/papers/16iccd_impica.pdf (ICCD 2016) +- Fine-grained DRAM: https://www.cs.utexas.edu/~skeckler/pubs/MICRO_2017_Fine_Grained_DRAM.pdf (text extracted with pdftotext: 3.92 pJ per bit per HBM2 access, 1.21 of it activation, the 256-byte row, tFAW "effectively eliminated", 51 percent) +- Folded Banks (ISCA 2025): https://dl.acm.org/doi/10.1145/3695053.3731111 (refused today; the abstract as the record read it on 6 October 2026); https://wantongli.ucr.edu/news/announcement22-ISCA%202025 +- LPDDR6: https://www.pcworld.com/article/2845759/lpddr6-memory-standard-announced-as-ddr5-dram-takes-over.html; https://hothardware.com/news/jedec-lpddr6-standard-released; https://www.microcontrollertips.com/what-is-jesd209-6-and-why-is-it-important-for-edge-ai/ +- CXL: https://introl.com/blog/cxl-memory-expansion-pooling-disaggregated-memory-ai-data-center-2025 (1 February 2026); https://www.snia.org/sites/default/files/2025-09/SNIA-SDC25-Peethambaran-CXL-as-scalable-cost-effective-Memory.pdf +- Cerebras: https://www.cerebras.ai/chip; https://www.nextplatform.com/2024/03/14/cerebras-goes-hyperscale-with-third-gen-waferscale-supercomputers/ (14 March 2024); https://sacra.com/c/cerebras-systems ("a couple million per system") +- FPGA with HBM: https://wccftech.com/amd-announces-mass-production-of-the-alveo-v80-compute-accelerator-9495-price-tag/ (17 May 2024); https://www.amd.com/en/products/accelerators/alveo/u55c/a-u55c-p00g-pq-g.html; https://www.nextplatform.com/2022/03/08/a-cornucopia-of-memory-and-bandwidth-in-the-agilex-m-fpga; https://arxiv.org/abs/2005.04324 (Shuhai) +- 3D DRAM: https://heise.de/en/news/Huge-RAM-3D-DRAM-with-multiple-layers-planned-from-2030-9738064.html (29 May 2024); https://www.yolegroup.com/industry-news/samsung-reveals-16-layer-3d-dram-plans-with-vct-dram-as-a-stepping-stone/ +- RLDRAM 3: https://www.micron.com/products/memory/dram-components/rldram-memory; https://newelectronics.co.uk/content/news/micron-unveils-third-generation-rldram-technology (2011); https://arxiv.org/pdf/1810.07059 +- SRAM at N2: https://spectrum.ieee.org/tsmc-n2-2670436570 (12 December 2024); https://marklapedus.substack.com/p/intel-tsmc-tout-sram-breakthroughs; wafer prices https://tech-insider.org/tsmc-2nm-wafer-price-2026/ +- UCIe and BoW: https://www.snia.org/sites/default/files/2025-05/SNIA-SDC22-Sharma-Universal-Chiplet-Interconnect-Express.pdf; https://www.3dtested.com/news/new-ucie-chiplet-standard-supported-by-intel-amd-and-arm +- Optical: https://www.theregister.com/2026/03/03/ayar_labs_500m/ (3 March 2026); https://www.allpcb.com/allelectrohub/photonic-interconnects-aim-to-solve-ai-memory-bottlenecks (the Celestial AI figures) +- H100 pointer-chase latency (353 ns, claimed): https://arxiv.org/pdf/2608.15764 diff --git a/tools/ci/export-exclude.txt b/tools/ci/export-exclude.txt index 76af5a1a3..eeeb9b8f6 100644 --- a/tools/ci/export-exclude.txt +++ b/tools/ci/export-exclude.txt @@ -19,3 +19,6 @@ docs/analysis/mission # copied logs are research and operations documents, not public export; the public text is the served sentence main landed. docs/analysis/cryptanalysis docs/plans/cryptanalysis +# 8 October 2026: the class v6 research lanes (the hardware future, the rotating-family research): internal research written +# for the founder on the chip model, quoting the operations record; the public text is whatever main lands from the synthesis. +docs/analysis/class-v6 From 2b76937caf3b953d81073502aee977dededb8c1c Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:19:10 +0000 Subject: [PATCH 13/19] Class v6 history lane, first cut: every ASIC-resistant proof of work and its chips by mechanism, the four-layer mapping with a number per chip class, and the three lessons that bind on v6 docs/analysis/class-v6/history.md: 21 chip rows by mechanism (what each chip specialised, what the design missed, the timeline, the v6 layer, closed or not, the per-joule number); in-depth sections for Ethash (the E3's DDR3, Linzhi's 4.4 GB beside 64 compute units, the Jasminer X4's DRAM hybrid-bonded onto a 40/45 nm die, the E9), ProgPoW (the Least Authority and Rao audits read from the PDFs, Kik, Linzhi's claim, the adopters), RandomX (the X5's RISC-V board at 1.46x, the X9 announced December 2025 and withdrawn May 2026 with zero units, RandomX v2 released 25 March 2026 with activation pending, the Qubic episode as the detector's lesson), Cuckoo (the GRN1's 512 MiB SRAM die, the G32, the iPollo G1, the 2025 bounty), Equihash (the Z9's single die, the Fudan parameter-following design), Scrypt and Argon2 (the BM1485's on-die scratchpads, Percival's lookup gap, MTP), the no-chip hashes, kHeavyHash, CryptoNight, X16R, Lyra2REv2, the compute rows. The four layers against the history layer by layer, with the E3's death dates, Autolykos's growth rule, the design-cycle and FPGA-compile clocks, and the honest verdict that the stored-dataset chip (class C) is not closed by any layer: 5.1x per joule at zero premium, 2.1x with the shadow at k = 1. Two corrections to the 5 October file (the X9, RandomX v2) and one claim marked unverified (the 5x fork-surviving chip). First cut for 15:00 UK; the full report by 09:00 UK on 9 October. Co-Authored-By: Claude Fable 5.1 --- docs/analysis/class-v6/history.md | 300 ++++++++++++++++++++++++++++++ 1 file changed, 300 insertions(+) create mode 100644 docs/analysis/class-v6/history.md diff --git a/docs/analysis/class-v6/history.md b/docs/analysis/class-v6/history.md new file mode 100644 index 000000000..1d7b524f4 --- /dev/null +++ b/docs/analysis/class-v6/history.md @@ -0,0 +1,300 @@ +# Class v6 research lane A, history: every ASIC-resistant proof of work, how it fell or held, and what that binds on the four layers + +8 October 2026, branch `class-v6-history`, the Counter ASIC coordinator's history lane. The founder's word at 11:1x UK: class v6 is declared with four layers as its spine, and the research opens "see if anything can be optimised, added or invented". This file extends and corrects `docs/analysis/asic-resistance-history.md` (5 October 2026, the deep dive: 31 rows, 24 papers, ten lessons, seven ranked additions); it does not repeat that file's rows. What is new here: the exact MECHANISM of every chip (what it specialised: the memory, the hash core, the instruction mix, the parameter fixity), what each design missed, the timeline from announcement to chip to response, and for each one the mapping to class v6's four layers with "does v6 close it" in one sentence and one number. Every figure about another chain cites a URL with the date it was read, or is labelled approximate. Every Igneum figure names the repo file. Reading public research is in-house; nothing is paid or asked of anyone outside. + +First cut landed before 15:00 UK on 8 October (the table, the mechanisms, the three lessons); the full report by 09:00 UK on 9 October. A line goes to the coordinator and to the synthesis lane at each landing. + +## 0. One page + +**The four layers, as declared, against the history's chip classes.** Every chip that ever beat a resistant hash belongs to one of five classes; the table says which layer of v6 answers each class, and with what number. "Closes" means the chip's measured or claimed edge falls under 2x per joule by a mechanism the layer supplies; "does not close" means the layer does not touch the chip's edge, and the number stands. + +| Chip class (the rows of section 1 it covers) | What the chip specialised | Its best measured edge | Which v6 layer answers it | Does v6 close it | The number after v6 | +|---|---|---|---|---|---| +| A. Fixed-function pipeline on a compute-bound hash (X11, Blake-256, Blake2b, Blake2s, kHeavyHash, Eaglesong, Blake3, SHA512/256d, NexaPow, X16R's FPGA) | the hash's round function unrolled in silicon; the instruction mix fixed at genesis | 33x to 1,280x per joule (Antminer KA3, D3, KS5 Pro) | layers 1 and 3 (the op mix, program length and families drawn or unlocked per era; already closed since class v2 by the per-epoch random program) | yes, and it was already closed: a per-epoch program has no round function to unroll | 0 of 128 loads and 0 of 512 program ops are a fixed function; what remains is class E | +| B. SRAM-scale memory (Scrypt's 128 KB, CryptoNight's 2 MB, Lyra2REv2's sponge, Cuckatoo's edge bitmap, Equihash's 144 MB) | the hash's whole working set on die or across a few dies, with a time-memory trade-off the designer had not drawn | 19x to 1,100x (Scrypt), 40x to 50x (CryptoNight X3), 12x to 100x (Equihash Z9 to Z15), 4x (Cuckatoo G1) | layer 2 (the dataset above any SRAM die, growing with chain state with a floor; the 256 MiB cache growing with it) | yes: the recompute chip that holds the cache on die reads 0.92x at the op budget and 1.86x per joule, and the cache's size forces it onto a 7 nm or better node | `chip-model-v3.md` section 2 and 5.4: 0.31x bare, 0.92x with the 3x factor, 1.86x per joule at f = 0; the curve from f = 0 to 1 is monotone and the partial-store chip is worse than both ends | +| C. The memory system without the GPU (the Ethash chips: Linzhi Phoenix, Jasminer X4, Antminer E9; the f = 1 chip of the model) | a controller and PHY for the same DRAM the card carries, a 32-byte atom per read, no shader, no scheduler, no clock tree; 55 W of memory without the 271 W of GPU | 2.1x to 4.8x per joule measured (Ethash, 2020 to 2022); 5.1x (GDDR7) to 7.5x (one HBM3 stack) modelled for Igneum at zero premium | none of the four directly: every per-era draw is firmware to a chip that stores the dataset; layer 2 only when the dataset passes the chip's board | **NO.** The f = 1 chip keeps 5.1x per joule on GDDR7 at zero premium and 2.1x with the class v4 shadow at k = 1; v6's draws move neither figure. What moves it is the honest card's own joules (the operating point, 3.6x at the 1,300 MHz knee) and the shadow's premium | 5.1x / 3.6x / 2.1x (zero premium unlocked / knee / knee with the shadow at k = 1), `counter-asic-4-research.md` sections 0 and 20.4; the chip's USD 2.8 per MH/s against the card's 14.7 | +| D. The firmware-survivable chip against periodic change (Monero's chips across four forks, Vorick's "survives forks at under a 5x hit", the X16R FPGA, the Equihash chip "able to follow parameter forks") | a programmable sequencer over the hash's op set; the per-block or per-fork change read as a configuration | chips back at 85 percent of Monero's hashrate four months after the v8 fork; 1.3x on X16R | layers 1 and 3 are the automatic form of the change those chains made by hand; they cost no governance event | partly: v6 closes the GOVERNANCE failure (no fork, no reset of the hashrate to a rentable size) and taxes die area for the reserve; it does not close the chip, because a reserve and a draw band readable at genesis are built in from day one | the draw's cost to the chip is a recompile per epoch and a new dataset mapping per era; the number is the "GPU without graphics" of ledger M1, which is class C's chip with a sequencer beside it: 5.1x less the shadow's k | +| E. The hot set and the steered address (Kik's 64-bit seed on ProgPoW, Dinur-Nadler on MTP, AP-F8-1 on class v4, the weak-day MUL draw) | a small SRAM serving the reads a biased draw or a cooperating node concentrates | 1.067x at v4's ceiling (0.52 to 4.6 percent of reads on 0.1 percent of items); a memory skip on ProgPoW 0.9.3; MTP from 2 GB to under 1 MB | layer 4 (the (c''') floor and the F8-form uniformity test generalised to every era's draw, with a redraw on failure) | yes, for the class the test models: the floor at 0.995 refuses every live hot set the in-house pass found (nine, 0.9809 to 0.9919) at 0.7 percent of candidates | `docs/spec/01-lottery-hash.md` 1.4.7.2 (class v5); the open residue is the shadow-written concentrations at 0.9992 to 0.9997, worth about 1.0004x to a chip | + +**The three lessons that bind on v6** (section 3 has the evidence): + +1. **The chip that stores the dataset is firmware-immune to every draw; only joules and memory growth move it.** Every per-era parameter (mixer rounds, op weights, read width, program length, shadow placement) and every family epoch is read by the f = 1 chip's sequencer as a configuration, as Monero's chips read four forks and X16R's FPGA read the per-block order. The number v6 inherits is 5.1x per joule at zero premium (2.1x with the shadow at k = 1), and the lever is the honest card's operating point and the shadow's premium, not the draw. +2. **Automatic change beats the human fork only where it costs the chip a redesign, and the one parameter that does is the memory.** Grin's six-monthly tweaks held because each was a new algorithm the lane was scheduled to retire; Monero's forks lost on the second lap; Ethash's DAG growth is the one scheduled change in the record that killed a shipped chip (the E3, when the DAG passed its 4 GB of DDR3). Layer 2 is that lesson made a rule, and its rate decides everything: at 2 GiB plus 0.5 GiB a year a 32 GB chip board outlives the chain, so layer 2 as declared ages out the honest 8 GB card before any chip unless the floor is set against DRAM cost per gigabyte, not against chain state. +3. **A steered or biased address pattern is always found after launch unless the test lives in the acceptance rule, and every new draw needs its own null.** ProgPoW's seed, MTP's blocks and class v4's lossy sources were the same attack three times; layer 4 puts the test where it must be, and its cost is one census per era draw on the node (2.2 s per candidate at 2^20 today) with the null re-derived for every drawn parameter, because the window model that defines "uniform" changes with the read width and the program length. + +**What the history says to add** (section 5): the issuance clock and the share-pattern detector (unchanged from the 5 October ranking, still unbuilt); a layer 2 floor stated in DRAM dollars; the read width kept out of the era draw unless a width other than 4 bytes is measured latency-bound on every vendor; the reserve ordered by what a sequencer chip cannot fold into firmware. + +## 1. The chips, one row per mechanism + +Columns: the hash; the chip (vendor, model); the mechanism (what it specialised); what the design missed; the timeline (hash live, to the chip, to the response); the v6 layer that answers it; does v6 close it; the number (the chip's measured or claimed per-joule edge, and what the layer leaves). Every ratio is arithmetic on the cited rate and watt figures of the chip and the best consumer GPU of its year, approximate by construction; section 2 carries the sources. "Closed at v2" means the per-epoch random program already removed the mechanism and v6 inherits it. + +| Hash (chain) | Chip | Mechanism: what it specialised | What the design missed | Timeline | v6 layer | Closed by v6 | The number | +|---|---|---|---|---|---|---|---| +| Ethash (Ethereum) | Bitmain Antminer E3 | 18 chips with 4 GB of commodity DDR3; "a memory interface connected to a small compute engine" | the hash needs only a memory interface; and the chip's memory was under-sized | live Jul 2015; chip announced Apr 2018, shipped Jul 2018; died by DAG growth Mar to Oct 2020 | 2 (the only layer that reached it) | class C: no | 1.0x to 1.1x per joule; the DAG passed its 4 GB at 20 to 27 months | +| Ethash | Innosilicon A10, A10 Pro, A11 Pro | the same with faster, larger DRAM (6 to 8 GB) | as above | Jul 2018 to Dec 2021 | none | no | 1.4x to 2.5x | +| Ethash | Linzhi Phoenix (E1400) | 64 compute units and 72 mixers per board beside 4.4 GB of "decentralized memory inside its ASIC"; sized to the DAG | memory energy per bit is the whole cost; many small memories by many small cores cut it | announced Sep 2018; tapeout Sep 2019; samples Dec 2020; no mass production | none | no | 2.1x to 2.2x | +| Ethash | Jasminer X4 (Sunlune) | DRAM dies hybrid-bonded (wafer-to-wafer DBI) onto a 40/45 nm logic die, 5 GB per unit, 40 chips per server | off-package DRAM at 3 to 4 pJ per bit was the cap; bonding takes it under 1 | Jun 2021 chip; Oct 2021 ship; the Merge 11 months later | none | no | 5.1x (about 7.3 pJ per bit all in) | +| Ethash | Bitmain E9, E9 Pro | conventional, 6 to 7 GB | as the E3, four years of DRAM later | Jul 2022 (8 weeks before the Merge); Feb 2023 | none | no | 3.0x, 4.1x | +| ProgPoW (never on Ethereum; KawPow, FiroPoW, ProgPowZ, Quai) | none; Linzhi claimed 3x to 8x with no derivation; one academic VU35P FPGA at 5.6 MH/s | a sequencer over eleven ops reads the per-period program as configuration (unbuilt); the on-die cache chip priced at "<< 0.1x" energy (unbuilt) | the light-evaluation chip was left as a suggestion; a 64-bit seed (Kik) | EIP May 2018; audits Sep 2019; exploit Mar 2020; dead Mar 2020; no chip on any adopter in 6 years | 1, 3 (the period made an era draw), 4 (the seed class) | class D and E yes; the Rao chip is class C, no | claimed 1.1x to 1.2x for the compute chip; the memory-system chip never priced; Igneum's reads 5.1x | +| RandomX (Monero) | Bitmain Antminer X5 | many RISC-V chips on a board over commodity DRAM: an in-order machine for a fixed VM spec; insides never published | CFROUND's cost on x86, a scratchpad-only AES circuit, the latency slack of a fixed 256-instruction program (v2's own list) | live Nov 2019; private mining from about 2021; X5 Sep 2023; X9 announced Dec 2025 and withdrawn May 2026 (zero shipped); v2 released Mar 2026, activation pending | 1 (program length and op mix drawn, which v2 had to fork to change) | class D: yes for the class; the gap itself has no GPU analogue | 1.46x (X5, measured); 3.8x (X9, claimed, never shipped); 5.4x (Pinecone R1X, claimed, undelivered) | +| CryptoNight (Monero) | Bitmain Antminer X3 (180 x BM1700); Baikal Giant N; secret chips from early 2017 | one or two 2 MB scratchpads in on-die SRAM with a hardware AES round per chip (inference from 1.2 kH/s at 2.6 W per chip) | the working set was a 2014 L3, which is a die | live Apr 2014; secret chips about 33 months; X3 announced at 47 and bricked by the v7 fork before delivery; chips back inside 4 months of v8; none in CN-R's 9 months | 2 (the dataset above any die); 1 and 3 (CN-R's per-block program is Igneum's per-epoch one) | yes | 40x (X3 against a Vega 64); Igneum's f = 0 chip 1.86x per joule | +| Cuckatoo31 (Grin) | Obelisk GRN1 (cancelled) | one TSMC 16 nm die with 512 MiB of SRAM holding the node bits, lean mining; 150 GPS at 800 W per chip | the resource was SRAM size at 320 MB, and a die was designed for it | hash live Jan 2019; announced Jan 2019; sold Apr; cancelled Jul 2019 | 2 | yes (closed at v3) | 50x planned; the 2025 bounty: memory trades for time, not energy | +| Cuckatoo32 (Grin) | iPollo G1 (30 x 12 nm chips) | node bits on die, the 512 MB edge bitmap in serial DRAM (the design's geometry; the G1's own split unpublished) | half the working set stayed in DRAM | Dec 2020, 23 months | 2 | yes | 3x | +| Equihash 200,9 (Zcash) | Bitmain Z9 mini, Z9 (BM1740); A9; Z11; Z15 Pro | the whole 144 MB working set on one die ("about 128 MB", eDRAM or SRAM); the Fudan NDSS 2019 design: an on-die linear sorter with the lists in off-chip DDR4, parameter-independent | the 1,000x halving penalty had no proof; 144 MB fit a die; a chip was designed to follow any (n, k) | live Oct 2016; secret chips before May 2018; Z9 mini at 18 months; Z15 Pro at 80 | 2 (a 2 GiB set is not a die); the parameter-following chip is layer 1's warning | yes by layer 2 | 12x (Z9 mini) to 110x (Z15 Pro); the Fudan design 13x in simulation | +| Scrypt (Litecoin) | Innosilicon A2; KnC Titan; Antminer L3+ (BM1485); L7 (BM1489) | one 128 KB scratchpad of on-die SRAM per core, 12 cores per 28 nm chip (L3+), 7 nm with 480 chips (L7); Percival's own lookup gap (half the memory for 25 percent more work) | the scratchpad was a 2011 cache; the time-memory trade favoured the chip at 2x to 4x by the designer's note | live Oct 2011; A2 Apr 2014 (30 months); L3+ Jun 2017; L7 Nov 2021 | 2 | yes | 70x (A2) to 1,300x (L7); Igneum's f = 0 chip 1.86x | +| Scrypt-N (Vertcoin 2014) | none; dropped pre-emptively | an automatic N doubling a chip follows with more SRAM or a wider lookup gap | public and slow growth sizes the chip for years | Jan 2014 to Dec 2014 | 2 (the precedent) | n/a | no chip shipped | +| Lyra2REv2 (Vertcoin) | FPGA bitstreams (2018), then Dayun Zig Z1 | a sponge of about 200 KB per core on die (approximate); 28 nm | SRAM-scale memory inside a hash chain | live Aug 2015; FPGA 2018; Z1 Sep 2018 (37 months); fork Feb 2019 | 2 | yes | 20x | +| X16R (Ravencoin) | CVP-13 FPGA bitstreams; OW1 and SKC Turing R1 | sixteen fixed cores with per-block routing | the draw changed the order, not the hardware | live Jan 2018; FPGA named Sep 2018; 45 percent of blocks by Jul 2019; fork Oct 2019; chips "evident" again Jan 2020 | 1 (the warning) | the fixed-function lane closed at v2; the draw is governance | 0.75x (OW1) to 5x (R1) | +| MTP, Argon2d (Zcoin) | none; Dinur-Nadler's attack before launch | the prover steers the data-dependent addresses into under 1 MB at a 170x penalty | the attacker controlled the memory's contents | 2017 attack; live Dec 2018; replaced Oct 2021 | 4 | yes (the day key is a VDF of chain state; the floor refuses hot sets) | 2 GB to under 1 MB; Igneum's residue about 1.0004x | +| Yescrypt, yespower | none | L2-latency-bound sequential work; small prizes | n/a | 2014 on | 2 | n/a | no chip | +| KawPow, Autolykos v2, Octopus, Verthash, FishHash | none | DRAM-scale random reads on small prizes | untested at a prize that pays for a controller project | 2020 to 2024 | 2 | class C: not tested | no chip; the Ethash record is their test | +| kHeavyHash (Kaspa) | IceRiver KS0 to KS5L; Antminer KS3, KS5 Pro | a fixed 64 x 64 nibble matrix pipeline | compute with a matrix in it is the cheapest silicon there is | live Nov 2021; KS0 Sep 2023 (22 months); GPU share gone by late 2023 | closed at v2; the mm8 reserve's warning | yes | 170x to 720x | +| Eaglesong (Nervos) | Toddminer C1; Antminer K5, K7 | a fixed sponge pipeline | compute | 4 months | closed at v2 | yes | about 100x | +| Blake3 (Alephium); Blake2s (Kadena); Blake-256 (Decred); Blake2b (Sia); X11 (Dash); SHA512/256d (Radiant) | the 5 October file's rows | fixed pipelines | compute | 4 to 31 months | closed at v2 | yes | 33x to 6,500x | +| NexaPow (Nexa) | DragonBall A21 | a secp256k1 Schnorr signature per nonce in hardware | a wide multiplier chain is already what the GPU does well | live 2023; chip Dec 2024 | closed at v2 | yes | 2.2x: the one compute row where the chip's edge stayed small | + +What the table says, sorted by class: class A (fixed pipelines) 33x to 6,500x, closed at v2; class B (SRAM-scale sets) 12x to 1,300x, closed at v3 by the dataset and the drawn curve; class C (the memory system) 1.0x to 5.1x measured on Ethash, modelled 5.1x to 9.2x for Igneum, NOT closed by any layer; class D (firmware-survivable sequencers) 0.75x to 5x on the program side, closed as governance and open as a chip; class E (steered addresses) closed by layer 4 for the class it models. + +## 2. The chips, in depth + +One section per proof of work. Each carries the mechanism, the miss, the timeline, the layer mapping and the number. Where a lane's return says "not found" the row says so. + +### 2.1 Ethash (Ethereum, July 2015): the E3, the Innosilicon line, Linzhi, Jasminer and the E9 + +**The mechanism of the hash.** A DAG of 1 GB growing 8 MB per epoch of 30,000 blocks (about 0.7 GB a year), 64 random 128-byte reads per hash mixed by FNV, keccak at each end; bandwidth-bound by design. EIP-1057 states the miss in one sentence: Ethash "requires external memory due to the large size of the DAG. However that is all that it requires - there is minimal compute ... a custom ASIC could remove most of the complexity, and power, of a GPU and be just a memory interface connected to a small compute engine" (https://eips.ethereum.org/EIPS/eip-1057, read 8 October 2026). + +**The chips, and what each specialised.** + +| Chip | Announced, shipped | The mechanism, as far as any source states it | MH/s, W, MH per joule | Against the best GPU of its year (derived, approximate) | Source (read 8 October 2026) | +|---|---|---|---|---|---| +| Bitmain Antminer E3 | leaked March 2018, announced 3 to 4 April 2018 at USD 800 (five per customer), shipped 16 to 31 July 2018; later batches USD 1,800 | 18 Ethash chips on three boards with 4 GB of commodity DDR3; chip, node, controller and read width never published; Bitmain's own support called it "a 4G video card" whose DDR "is up to the upper limit"; Rao's audit classes it as the conventional strategy (compute in silicon, memory off chip) | 180 claimed, 190 to 200 shipped, at 800 W: 0.24 MH/J | 1.0x to 1.1x against a tuned GTX 1080 Ti (45 MH/s at about 200 W, 0.225); under Rao's best overclocked 2019 GPU (0.40) | https://cryptoslate.com/bitmain-e3-asic-ethereum-miner/ ; https://2miners.com/blog/asic-miners-for-ethereum-antminer-e3-vs-innosilicon-a10-eth-master-comparison/ ; https://coingeek.com/memory-limitations-prompt-bitmain-antminer-e3-to-halt-etc-support/ ; https://www.kryptex.com/en/hardware/nvidia-gtx-1080-ti/reviews | +| Innosilicon A10 ETHMaster, A10 Pro (6 GB), A10 Pro+ (7 GB), A11 Pro (8 GB) | A10 announced 23 July 2018 at 365, 432 and 485 MH/s (USD 3,800 to 5,000); A10 Pro June 2020; A10 Pro+ January 2021; A11 Pro presold March 2021 at 2,000 MH/s and 2,500 W, shipped December 2021 at 1,500 MH/s and 2,350 W ("20 percent less efficient"; broker quotes about USD 27,000) | the same conventional chip with a larger and faster memory system; the DRAM type is not stated on any page read (GDDR6 is the trade's assumption, approximate); node and read width not found | A10 485 at 850 W: 0.57; A10 Pro 500 at 860: 0.58; A10 Pro+ 750 at 1,350: 0.56; A11 Pro 1,500 at 2,350: 0.64 (0.80 claimed) | A10 2.5x against the 1080 Ti; A10 Pro 1.4x and A11 Pro 1.6x (2.0x claimed) against a tuned RTX 3090 (120 MH/s at about 295 W, 0.41) | https://www.criptonoticias.com/mineria/nuevo-minero-asic-innosilicon-procesa-485-mh-ethereum ; https://www.theblock.co/post/125871/innosilicon-ethereum-miner-a11-pos ; https://whattomine.com/coins/151-eth-ethash/asics | +| Linzhi Phoenix (E1400) | announced 13 to 14 September 2018 by Chen Min (ex-Canaan) at 1,400 MH/s and 1 kW for April 2019; planned tapeout December 2018, actual September 2019; sample rollout 21 December 2020; no mass production; price never published; the company now "studying new opportunities" | two E1400 boards, each 64 compute units and 72 "mixers" with 4.4 GB of "decentralized memory inside its ASIC design" (The Block); the DRAM type never published (one 2019 forum post says an interposer and stacked HBM dies, unverified); the 4.4 GB sized to the DAG rather than to a commodity 6 or 8 GB, which reads as memory sized per chip (inference); Linzhi claimed a ProgPoW chip would reach 3x to 8x | 2,600 claimed, 2,733 measured by F2Pool, at about 3,000 W: 0.87 to 0.91 | 2.1x to 2.2x against the tuned 3090 | https://www.theblock.co/post/88622/questions-new-ethash-asic-ethereum ; https://www.coindesk.com/tech/2020/12/21/linzhi-begins-rollout-of-long-awaited-ethereum-miner-phoenix ; https://bitcoinmagazine.com/business/new-mining-manufacturer-linzhi-announces-ethereum-asic-miner ; https://linzhi.io/ ; https://github.com/Souptacular/linzhi | +| Jasminer X4 (Sunlune) | chip announced 6 June 2021; X4 server announced 11 October 2021, first batch shipped 29 October 2021; launch price not found (USD 497 used today) | **the only Ethash chip that moved the memory**: TechInsights found "the first ever DRAM-to-Logic hybrid-bonding" (wafer-to-wafer DBI), DRAM dies bonded face to face onto a 32 mm by 21 mm logic die on XMC's planar 40/45 nm node; Jasminer's words: "3DIC technology, by integrating the data storage unit and the computing unit on the same chip"; 5 GB per unit, 40 chips per X4 server; the DRAM vendor, capacity per die and node not found | 2,500 at 1,200 W: 2.08 (the X4-1U 520 at 240 W: 2.17) | 5.1x against the tuned 3090; a physics check: 64 reads of 128 bytes is 65,536 bits a hash, so 2.08 MH/J is about 7.3 pJ per bit all in, under any off-package DRAM | https://www.techinsights.com/ko/node/51986 ; https://www.techinsights.com/ko/node/52149 ; https://semiconductor-digest.com/?p=22931 ; https://miningnow.com/asic-miner/jasminer-x4-2500mh-s/ | +| Bitmain Antminer E9, E9 Pro | teased 27 April 2021 as "3 GH/s, the work of 32 GPUs"; shipped July 2022 at 2.4 GH/s, eight weeks before the Merge; E9 Pro February 2023, Classic only | conventional off-chip DRAM at larger scale: E9 (model 240-E) 6 GB in six bins 2,100 to 2,400 MH/s; E9 Pro (260-E) 7 GB; memory type not found on any page read; no teardown | E9 2,400 at 1,920 W: 1.25; E9 Pro 3,680 at 2,200: 1.67 | 3.0x and 4.1x against the tuned 3090 | https://d-central.tech/antminer-e9-family/ ; https://www.asicminervalue.com/miners/bitmain/antminer-e9-2-4gh ; https://www.coindesk.com/tech/2021/04/27/bitmain-to-release-antminer-e9-asic-for-ethereum-mining | + +**Why the cap sits at 2x to 5x.** Rao's audit (6 September 2019): every DAG read is random at about 40 ns of latency "completely independent of the memory bandwidth or the computation engine"; "typical DRAM energy dissipation is 3 to 4 pJ per bit" and "the energy expended to move data from DRAM to compute are the same for GPU or ASIC", so the shipping chips showed "about 1.6x hashrate per watt over GPUs" (E3 0.24, A10 0.57, the best overclocked GPU 0.40 MH/W in his table); integrating memory with logic cuts the movement energy "much more than 10x" to "under 0.3 pJ per bit", "the looming threat" (https://github.com/ethcatherders/progpow-audit, the PDF's text, read 8 October 2026). Ren and Devadas (TCC 2017) give the bound: memory hardness bounds area, not energy; the energy of a memory access is comparable on a chip and a CPU, so bandwidth hardness is the only energy lever (https://eprint.iacr.org/2017/225). The derived ladder at 65,536 bits a hash: pure memory energy caps Ethash at about 0.76 MH/J on DDR3, 2.8 on GDDR6 and 3.9 on HBM2 (O'Connor et al., MICRO 2017: HBM2 3.92 to 3.97 pJ per bit, GDDR5 14.0); the shipped chips sit at 0.24 (E3), 0.6 (Innosilicon), 0.9 (Linzhi), 1.25 to 1.67 (E9, E9 Pro) and 2.1 (Jasminer, by leaving commodity packaging). The cap was the memory's own energy per bit, and the one chip that beat it moved the memory onto the die's face. + +**The timeline.** Hash live July 2015; the E3 at 32 months (announced) and 36 (shipped); the first chip over 2x at 65 months (Linzhi, December 2020); 5x at 75 months (Jasminer, October 2021); the Merge at 86 (15 September 2022). The E3's death by DAG growth: Classic first, at epoch 328 (DAG about 3.56 GB, March 2020), then Ethereum, with a 30 March 2020 firmware stretching the DDR to about block 11.4 million (about October 2020): 20 to 27 months after shipping. Ethereum's responses: Zamfir's April 2018 poll (57 percent for an anti-chip fork); EIP-1057 created 2 May 2018, a 93 percent community vote in April 2019, audits delivered September 2019, "accepted" on 21 February 2020, EIP-2538's opposition on 25 February, then stagnant; the share claim in the EIP's own text: "as much as 40 percent of the Ethereum network may now be secured by ASICs" (undated inside a 2018 to 2020 document; no year-by-year series exists). After the Merge: Classic's hashrate went 64 to 183 TH/s in one day; today Classic reads 129.9 TH/s and ETHW 2.15; at USD 0.10 per kWh every Ethash chip in the table loses money (E9 Pro minus USD 5.28 a day), and a later wave (iPollo V1 3.6 GH/s at 3,100 W, June 2022; Jasminer X16-P 5.8 GH/s at 1,900 W, August 2023) holds Classic (https://hashrateindex.com/blog/how-much-ethereum-mining-hashrate-can-other-blockchains-absorb/ ; https://2miners.com/etc-network-hashrate ; read 8 October 2026). + +**The mapping to v6.** Ethash is class C in full, and its chips are the f = 1 chip of `chip-model-v3.md` section 5 at three points on the packaging ladder: commodity DRAM on a board (E3, E9: 1x to 4x), memory sized and placed per chip (Linzhi: 2x), DRAM bonded to the logic (Jasminer: 5x, the model's HBM3 row). None of v6's four layers touches a chip of this class: the program, the mixer, the op mix, the family schedule and the acceptance floor are all firmware or configuration to a controller that stores the dataset; the only layer that reaches it is layer 2, and only when the dataset passes the chip's board, which at 2 GiB plus 0.5 GiB a year is year 60 for a 32 GB board (section 4.2). Does v6 close it: **no**. The number: 5.1x per joule on GDDR7 and 7.5x on one HBM3 stack at zero premium against the 5090 unlocked, 3.6x at the 5090's 1,300 MHz knee, 2.1x at the knee with the class v4 shadow at k = 1 (`counter-asic-4-research.md` section 0); the history's measured band for exactly this chip class is 1.0x (E3) to 5.1x (Jasminer), and Jasminer's number is the model's HBM-class row reached in 2021 on a 40 nm logic die. The one thing Igneum has that Ethash did not: the honest card is latency-bound at 4-byte reads, not bandwidth-bound at 128, so the chip's energy per read is the activate's 909 pJ plus a 32-byte atom (2.0 nJ on GDDR7 against the card's measured 8.7 to 10.9 nJ marginal), which is where the 5.1x comes from, and the shadow is the only term on the card's side of that ratio. + +### 2.3 RandomX (Monero, 30 November 2019): the chips, RandomX v2, and the X9's withdrawal + +**The mechanism of the hash.** A VM running 8 chained programs of 256 instructions, 2,048 iterations each, over a 2 MiB scratchpad in three tiers (16 KiB, 256 KiB, 2 MiB) and a 2,080 MiB dataset derived from a 256 MiB cache by SuperscalarHash, a random superscalar program of about 450 instructions with 155 64-bit multiplies per function, tuned to a 170-cycle latency to match DRAM; double-precision floating point in all four rounding modes; the light-mode chip (cache on die) pays 760 cycles and 1,240 multiplies per item, "energy comparable to loading 64 bytes from DRAM" (https://github.com/tevador/RandomX/blob/master/doc/design.md and doc/specs.md, read 8 October 2026). Its DRAM argument: "DRAM cannot do more than about 25 million random accesses per second per bank group", about 1,500 H/s per bank group. + +**The chips, and what each specialised.** + +| Chip | Date | What is known of the inside | Rate, watts | Per joule against the best CPU | Source (read 8 October 2026) | +|---|---|---|---|---|---| +| Bitmain Antminer X5 | announced 27 August 2023, shipped September 2023 | "Bitmain's first RISC-V architecture CPU" (the reseller's only line); Spagni: not an ASIC but "a board containing multiple RISC-V CPU chips"; SChernykh: the chips were likely in use from about 2021, two years before sale, and do not beat Ryzen rigs per joule; core model, count, node, DRAM type and amount: NOT FOUND, no teardown | 212 kH/s at 1,350 W (157 H/W) | 1.46x over a Ryzen 9 7950X (107.5 H/W on Kryptex); a 100 W-capped Ryzen 9 9950X at 199 H/W beats it | https://criptonoticias.com/mineria/bitmain-lanza-antminer-x5-mineria-monero-asic ; https://bt-miners.com/products/bitmain-antminer-x5-monero-miner-212k-bt-miners/ ; https://pool.kryptex.com/en/device/cpu/amd/ryzen-9-7950x | +| Bitmain Antminer X9 | sales opened 26 December 2025 at USD 5,600, shipping scheduled July 2026; WITHDRAWN by mid-May 2026, refunds within hours, zero units shipped, "technical adjustments and a new strategy" through resellers, no Bitmain statement | "custom RISC-V cores specifically optimized for RandomX" (Bitmain's claim as relayed in Monero issue 10270); nothing else | 1,000 kH/s at 2,472 W (404 H/W), claimed | about 3.8x over the 7950X, claimed, never measured | https://bitmain.com.vc/news/bitmain-launches-antminer-x9 ; https://github.com/monero-project/monero/issues/10270 ; https://oneminers.com/blogs/news/whatever-happened-to-the-antminer-x9-bitmain-monero-miner (2 October 2026) | +| Pinecone INIBOX R1X | launched March 2026; shipping windows slipped from August to 10 to 18 October 2026; USD 3,200 to 4,950 | "built from ground up silicon", "optimized memory architecture for RandomX"; cores, node, DRAM: NOT FOUND; no delivered unit tested | 1,200 kH/s at 2,055 W (584 H/W), claimed | about 5.4x over the 7950X, claimed, undelivered | https://pineconebox.com/product/3 ; https://millionminer.com/news/monero-mining-guide-2026-antminer-x5-x9-pinecone-r1x-randomx | +| tevador's own "possible ASIC design" (24 December 2018, pre-release, marked outdated) | | 4 GiB of HBM for the dataset, 64 MiB of SRAM for 256 parallel 256 KiB scratchpads, 256 decoder and scheduler cores, 28 single-instruction workers; about 120,000 programs a second at about 300 W, "9 times more power efficient than a CPU" | | 9x, by the designer's own estimate of the pre-release design | https://github.com/tevador/RandomX/issues/11 | + +**What the design missed, in its authors' words.** RandomX v2 (PR 317 by SChernykh, merged 17 February 2026; v2.0 released 25 March 2026; Monero mainnet activation in PR 10038, open since August 2025, no date) names the three gaps a chip or a "specially designed CPU" took: (1) CFROUND, the rounding-mode switch, "costs up to 10 percent of hashrate on Ryzen CPUs" and "this is where an ASIC or a specially designed CPU can get an easy advantage"; v2 switches rounding 16 times less often; (2) the scratchpad initialisation was the only AES, so "a dedicated circuit for scratchpad initialization" paid off; v2 puts 16 AES operations per iteration in the main loop; (3) "while CPU cores got faster over the years, RAM latency stayed basically the same", about 50 to 55 ns from tuned DDR4 in 2019 to tuned DDR5 in 2026, so the fixed 256-instruction program left latency slack a faster core could not fill; v2 lengthens the program to 384 and prefetches two iterations ahead. Work per hash rises 52.9 percent; measured CPU hash rates move from minus 12.9 percent (a 100 W-capped 9950X) to plus 8 percent (a 28 W laptop part) (https://github.com/tevador/RandomX/blob/master/doc/design_v2.md and /pull/317, read 8 October 2026). The press reading: v2 "doesn't seem to be an attempt to eliminate every form of specialization", it removes "unintended advantages that benefited hardware in version 1.0" (https://www.coinpro.ch/en/?p=42081, 31 March 2026). + +**The timeline.** Live 30 November 2019; X5 at parity hardware 46 months later (September 2023), on chips SChernykh believes mined privately from about 2021 (21 months after launch); the X9 announced at 73 months and withdrawn at 78; v2 released at 76 months with a 52.9 percent work increase and no activation date; the R1X undelivered at 82 months. The X9's withdrawal is read by the trade press as Bitmain waiting for v2 rather than shipping a part the fork would hit (https://oneminers.com/blogs/news/antminer-x9-cancelled-what-bitmain-pulling-the-model-means-for-monero-mining, 31 July 2026). Corrections to the 5 October file's row 17: the X9 was not delivered in July 2026 (withdrawn, zero units), and "no fork as of October 2026" is now "v2 released, activation pending". + +**The mapping to v6.** RandomX is class D (the firmware-survivable machine): every X5 claim is a many-core RISC-V board over commodity DRAM, a better CPU for a fixed VM spec, and the v2 fixes are exactly the parameters class v6 layer 1 draws (program length, the op mix's cost on the honest machine, the memory latency slack). Two readings bind: first, RandomX's whole gap is the CPU's out-of-order overhead against an in-order many-core board, and Igneum's honest machine is already the in-order many-lane design, so the X5's 1.46x has no Igneum analogue; the Igneum analogue of "a better machine for the fixed spec" is class C's memory chip at 5.1x. Second, the v2 changes show what a per-era draw of program length and op mix buys: it closes the slack a faster honest core leaves (the latency-shadow argument of class v4 in RandomX's words), and it costs the honest machine up to 12.9 percent of rate at the power-capped point, which is the Igneum premium question in another chain's numbers. Does v6 close it: yes for the class (layers 1 and 3 draw what v2 had to fork to change), and the number is the X5's 1.46x, which becomes the shadow's premium arithmetic on Igneum (2.1x at k = 1 at the knee). + +**The Qubic episode (2025) as the detector's lesson.** Qubic's pool reached an average of 22.09 percent of Monero's hashrate over the campaign and 23 to 34 percent during ten withholding periods, with six-hour windows near 50 percent and never a daily 51 percent (Lee and Kim, arXiv 2512.01437, AFT 2026, read 8 October 2026); an 18-block reorg on 14 September 2025 invalidated 117 to 118 transactions. Detection rested on things the adversary controlled: one payout wallet, extra-nonce signatures in its coinbase, its own pool API; when Qubic encrypted its job messages and rotated keys those signals went, and Rucknium's warning stands that a miner split across addresses and solo-mining leaves only the orphan rate and double spends as signals. Qubic ran stock CPU miners, so the nonce-distribution method that found the 2018 and 2019 chips (MoneroCrusher: nonces clustered under about 1.34 billion of 4.3; 85.2 percent of the hashrate, about 5,400 machines at 128 kH/s) did not apply. For Igneum's detector (the 5 October addition 4) this means two instruments, not one: the per-program rate spread and nonce pattern for a chip, and a share-by-key-and-template pattern for a concentrated honest fleet; neither survives an adversary who randomises both. + +### 2.2 ProgPoW (EIP-1057, May 2018): the independent review, the exploit, and the adopters + +**The mechanism.** A random program re-drawn every PROGPOW_PERIOD (50 blocks in 0.9.2, 10 blocks, about 2 minutes, in 0.9.3) from the block number, so miners compile ahead; 16 lanes, a 32-register file per lane, 64 outer iterations each with 4 uint32 DAG loads per lane (256 bytes per lane-group read), 11 cache accesses into a 16 KB cache and 18 random math ops drawn from eleven (add, mul, mulhi, min, rotl, rotr, and, or, xor, clz, popcount) with KISS99 as the generator and FNV1a for merging; keccak-f800 with 32-bit words at both ends "to reduce impact on total power"; the stated aim is that "the algorithm's requirements match what is available on commodity GPUs", the stated chip gain "minimal, roughly 1.1 to 1.2x", with the remaining chip levers named as removing the graphics pipeline, the floating-point units and minor merge-function tweaks (https://eips.ethereum.org/EIPS/eip-1057 and https://github.com/ifdefelse/ProgPOW, read 8 October 2026). + +**The independent review, exactly.** Least Authority (report version 9 September 2019): no issues, five suggestions. Suggestion 2, the light-evaluation attack, in the report's words: "on-die scratchpad memory of around 100 MB is possible in ASICs, that we can fetch at least 128 bytes during a single read, and that such a read might have a latency in the range of one to some tens of cycles"; a chip replaces every DAG read with calc_dataset_item(cache, i) over an on-die cache, at a latency near DATASET_PARENTS x k1, "as low as about 300 cycles"; "the energy expended per bit to access DRAM is about 3 pJ per bit, but when the memory access is on-chip, it decreases to 0.3 pJ per bit, which is a 10x improvement"; conclusion: "efficient light-evaluation attacks may become possible within a few years. This is also an issue that applies to Ethash"; the mitigation offered: raise DATASET_PARENTS from 256 to 512 (which 0.9.4 did), and "for details on the related hardware advancements, please see Bob Rao's corresponding audit report". Suggestion 5: "hardware targeting machine learning is also useful for ProgPoW mining" (the PDF at https://leastauthority.com/static/publications/LeastAuthority-ProgPow-Algorithm-Final-Audit-Report.pdf, text extracted, read 8 October 2026). Bob Rao (6 September 2019): "the only meaningful metric is Energy per Hash"; Ethash chips "about 1.6x hashrate per watt over GPUs"; "10/7 nm processes provide up to 25 Mbits per mm^2 of SRAM and 100 M transistors per mm^2"; "with sufficient on-chip memory available, ProgPOW ASICs with << 0.1x E/H over GPUs can be built"; three chip approaches costed (the whole DAG on die, "possible in 2025+"; a custom stacked memory; the cache only, "about 51 MB as of 30 August 2019", with the item recomputed, "512 MB SRAM" on the slide); the DAG-on-die economics on a three-year Moore cadence: a single die holding a 6.49 GB DAG in 2024 to 2025 at 532 mm^2 and USD 221 per good die (USD 34 per GB), or sixteen dies of 33 mm^2 at USD 6.51 each; "a die that can hold the logic and entire DAG at any point in time becomes cost effective at around 2025 and beyond"; the 16-die split "becomes cost effective today": 16 x USD 6.62 of silicon plus USD 16 of package plus USD 25 of PCB plus USD 25 of heatsink and interface, "about USD 172 total", against "about USD 240" for a GPU board with "8 GB GDDR6 about USD 150"; a 10 nm-class chip "USD 20 M+" and "1+ year to develop, can be done if there is a 150-day ROI to miners" (the PDF at https://github.com/ethcatherders/progpow-audit, text extracted, read 8 October 2026). + +**The exploit.** Kik, 4 March 2020: the 64-bit seed carried between the two keccak passes is too small; fix a seed and compute its mix once, grind an extra-nonce in the header to meet the difficulty on the final keccak, then scan nonces until keccak_progpow_64(header_hash, nonce) equals the seed; the memory path runs once per 2^64 nonces and the rest is keccak, "ASICs benefit most when network difficulty exceeds 2^50"; 0.9.4 widened the carried state from 64 to 256 bits (the digest of the first keccak plus the mix plus padding) (https://github.com/kik/progpow-exploit ; https://github.com/ifdefelse/ProgPOW ; read 8 October 2026). + +**Linzhi's claim.** 8 January 2019: "shocked" by ProgPoW with USD 4 M invested, and a stated intention "to study the feasibility, and then build, ProgPoW ASICs"; the repository recording their claim puts a ProgPoW chip at 3x to 8x (https://github.com/Souptacular/linzhi ; https://forklog.com/proizvoditel-majnerov-linzhi-vystupil-protiv-realizatsii-predlozheniya-progpow/ ; read 8 October 2026). No ProgPoW chip was ever shown. + +**The timeline and the adopters.** EIP created 2 May 2018; a 93 percent vote of 2.93 M ETH in April 2019; both audits September 2019; "accepted" on the 21 February 2020 call; EIP-2538's opposition 25 February; the 6 March 2020 call with "frustration but little progress"; stagnant since; the Merge 15 September 2022. KawPow (Ravencoin, 6 May 2020), FiroPoW (26 October 2021), ProgPowZ (Zano), Sero and Quai (January 2025) run ProgPoW variants; no chip is listed for any of them on WhatToMine or asicminervalue as of 8 October 2026 (https://whattomine.com/coins/234-rvn-kawpow/gpus ; https://www.asicminervalue.com/), on prizes that never reached the market caps at which the 2018 chips appeared (the 5 October file, section 2.5). + +**The mapping to v6.** ProgPoW is class D (a sequencer over eleven ops reads the per-period program as configuration) and class E (Kik). Its review is the one piece of the record that priced Igneum's own chips before Igneum did: the light-evaluation attack is the f = 0 recompute chip (M16, 0.92x at the op budget with the mixer at x8), and Rao's 16-die DAG holder at USD 172 is the f = 1 chip at USD 470 of memory and board. Does v6 close it: the class D half yes (layers 1 and 3 are ProgPoW's period change made an era draw, with no fork), the class E half yes (Igneum's seed is 256 bits through the VDF; layer 4 is the acceptance-side test the audits said to add), and the Rao half no (class C, section 2.1). The number: ProgPoW claimed 1.1x to 1.2x against a conventional compute chip and never priced the memory-system chip; Igneum's model gives that chip 5.1x. + +### 2.4 Cuckoo Cycle (Grin, January 2019): the GRN1, the G32, the iPollo G1 and the 2025 bounty + +**The mechanism of the hash.** Find a 42-cycle in a random bipartite graph of 2^31 or 2^32 edges from siphash; the lean solver keeps "1 bit per edge and 1 bit per node in one partition", bottlenecked by random node-bit access, which "requires tons of SRAM, which is lacking on CPUs and GPUs, but easily implemented in ASICs"; the mean solver keeps 33 bits per edge and is bandwidth-bound, about 4x faster; Tromp: "our primary PoW of Cuckatoo31+ is intended to be mined by ASICs"; the family is "a Proof of SRAM" (https://github.com/tromp/cuckoo ; https://forum.grin.mw/t/cuckatoo31-im-mutability/2442 ; read 8 October 2026). The memory geometry a chip needs (the 13 November 2018 feasibility thread): the 512 MB edge bitmap is accessed sequentially and can sit in external DRAM (96 GB/s with 128 MB on chip); the node bitmap must be SRAM; Cuckatoo31 fits one die at "256 + 64 = 320 MB of on-chip memory", Cuckatoo32 needs "at least 640 MB" or 512 MB of SRAM plus 512 MB of serial DRAM; "trimming is over 99 percent of the effort" (https://forum.grin.mw/t/cuckatoo32-feasibility/1199 ; https://forum.grin.mw/t/advice-on-cuckatoo-hardware-implementation/12042). The original "several orders of magnitude" time-memory claim fell to Andersen's edge trimming on 31 March 2014, two months after publication, and the design took it as its baseline. + +**The chips.** + +| Chip | Dates | Mechanism | Rate, watts, price | Against a GPU | Source (read 8 October 2026) | +|---|---|---|---|---|---| +| Obelisk GRN1 (Cuckatoo31) | announced 17 January 2019; chip details 20 March; sale 9 April (Mini 70 GPS at 400 W for USD 2,000; GRN1 420 GPS at 2,200 W for USD 10,000; Immersion 840 at 4,400 W for USD 20,000; shipping October 2019); CANCELLED 19 July 2019 with full refunds | one die, TSMC 16 nm, "a full 512 MiB of memory on board" (SRAM), 150 GPS at 800 W per chip, "thousands of hashing cores and memory banks", siphash plus blake2b, two sorter types; cancelled for the Cuckatoo31 phase-out, Grin under USD 2 in May 2019 and funding | 420 GPS at 2,200 W planned | about 50x per joule against a GTX 1080 Ti at about 0.9 GPS and 250 W on C31 (approximate), planned, never built | https://forum.grin.mw/t/obelisk-grn1-chip-details/4571 ; https://forum.grin.mw/t/obelisk-grn1-full-sale/4773 ; https://forum.grin.mw/t/grn1-cancellation-announcement/5624 | +| Innosilicon G32 (C31+ and C32+) | announced 17 April 2019 (G32-Mini 21.5 GPS at 140 W for USD 788; G32-1800 328 GPS at 1,800 W for USD 9,388; delivery from August 2019); CANCELLED 16 January 2020 citing foundry delays | about 100 chips at about 1.8 to 1.9 GPS each on C32 (an investor's figure) | never shipped | | https://forum.grin.mw/t/innosilicon-grin-miner-g32-preliminary-specification/4842 ; https://forum.grin.mw/t/innosilicons-grin-asics-canceled/6932 | +| iPollo G1 (Cuckatoo32) | December 2020 | 30 chips at 12 nm; whether the node bits sit in SRAM or the edge bitmap in DRAM is not published; Tromp doubts it is multi-chip in Innosilicon's sense | 36 GPS at 2,800 W, USD 9,000 | about 3x per joule against an RTX 4060 at 0.45 GPS and 110 W | https://pool.kryptex.com/device/asic/ipollo/g1 ; https://miningboard.com/algorithms/Cuckatoo32 | + +**The 2025 bounty.** Stephan Theisgen claimed the USD 10,000 linear time-memory trade-off bounty on 11 April 2025 (a solver at N/k bits at most 10k times slower, any k at or above 2), paid 30 April; the measured penalty about half an order of magnitude past linear; Tromp's reading: a chip with N/k bits must hash each edge "roughly (k + 1000) times" against under 6 for the lean miner, so memory can be traded for time but not for energy, and Cuckatoo "remains a Proof of SRAM" (https://forum.grin.mw/t/another-cuckatoo-bounty-succesfully-claimed/11739, read 8 October 2026). + +**The mapping to v6.** Class B. Cuckoo's resource was SRAM size at 320 to 640 MB per die, and a 16 nm die with 512 MiB of SRAM was designed, priced and sold before the economics killed it; the one chip that shipped sat at 3x because half its working set stayed in DRAM. Layer 2 answers it the way the chip model already does: Igneum's 256 MiB cache is the GRN1's die (128 mm^2 at N5, `sram-mirror.md`), and the 2 GiB dataset above it is what the lean solver never had to hold; the time-memory curve Cuckoo got wrong by 50x and then bounded in 2025 (linear in time, not in energy) is the curve `chip-model-v3.md` section 5 draws for Igneum (monotone; the f = 0 end pays 6.3 nJ and 9,360 ops per item against 1.2 to 2.0 nJ for a stored one), and its verdict is the same as Tromp's: memory can be traded for time, not for energy. Does v6 close it: yes, and it was closed at class v3. The number: 50x planned on C31 and 3x shipped on C32; Igneum's f = 0 chip 1.86x per joule and the f = 1 chip, which Cuckoo did not have because its memory was never a DRAM-scale random-read set, 5.1x. + +### 2.5 Equihash 200,9 (Zcash, October 2016): the Z9 and the parameter-following chip + +**The mechanism of the hash and its claim.** Wagner's generalised birthday problem with algorithm binding; the paper's claim: a PoW needing "700 MB of RAM" that "increases the computations by the factor of 1000 if memory is halved" (https://eprint.iacr.org/2015/946, read 8 October 2026). Zcash chose (200, 9), which solvers run in about 144 MB (approximate); no trade-off-resistance bound was ever proved for Equihash (Alcock and Ren, 2017, cited through the Fudan paper below). + +**The chips and the mechanism.** Bitmain's BM1740 (Z9 mini, announced 3 May 2018 at 10 kSol/s and 300 W, shipped June; Z9 September 2018, 42 kSol/s at 970 W on 48 chips) is "a single-chip Equihash miner", "presumably with around 128 MB of memory", eDRAM or SRAM "an open question" (Tromp, 7 June 2018, https://forum.z.cash/t/let-s-talk-about-asic-mining/27353/3332): the whole (200, 9) working set on one die, which the 1,000x claim had assumed impossible at that size. A (144, 5) solver needs over 1.6 GB and cannot fit; Tromp "physically inspected the product and did not find enough memory to handle (144, 5)". Vorick's architecture (13 May 2018) is the other route: "a basic architecture for equihash ASICs that would be able to successfully follow a hardfork that chose any set of parameters", with "massive speedups and efficiency gains over GPUs", because on a chip "you can merge the memory and computation together ... do most of your manipulating in-place" (the essay, through https://steemit.com/crypto/@waraa/the-state-of-cryptocurrency-mining). The published design of that route: Bai, Gao, Hu and Zhang, NDSS 2019, an adversary solver whose sort step is a linear-time insertion sorter of 2,048 "smartcell" flip-flop chains feeding merge stages buffered in off-chip DDR4 (list 1,600 Mib, pairs 2,016 Mib for (200, 9)), pair generation and XOR on small MCUs; simulated at SMIC 28 nm: 40.6 Sol/s at 500 MHz for 0.75 to 0.78 W, 52 to 54 Sol/J against about 4 for the best GPU software, "at least 10x" and parameter-independent (https://www.ndss-symposium.org/wp-content/uploads/2019/02/ndss2019_09-5_Bai_paper.pdf, read 8 October 2026). Later chips: Innosilicon A9 (June 2018, 50 kSol/s at 620 W, USD 9,999); Z11 (April 2019, 135 kSol/s at 1,418 W, 12 nm); Z15 (420 kSol/s at 1,510 W); Z15 Pro (June 2023, 840 kSol/s at 2,780 W). Against a GTX 1080 Ti at 735 to 785 Sol/s and 250 to 305 W (about 2.7 Sol/J): Z9 mini 12x, Z9 16x, A9 30x, Z11 35x, Z15 Pro 110x (https://www.asicminervalue.com/miners/bitmain/antminer-z9 ; https://www.asicminervalue.com/miners/bitmain/antminer-z15-pro ; https://en.wikibooks.org/wiki/ZCash_mining_GPU_Comparison/GPU_Mining ; read 8 October 2026). + +**The timeline.** Hash live 28 October 2016; three groups mining on secret chips before the Z9 announcement (Vorick); the Z9 mini at 18 months; the Zcash Foundation's statement of 8 May 2018 asked whether chips "could handle different parameters of Equihash" and the community vote of June 2018 went 45 to 19 against prioritising resistance; no fork; proof of stake announced November 2021; the Z15 Pro at 80 months. + +**The mapping to v6.** Class B, with a lesson for layer 1: Equihash's parameters (n, k) were the knob the forks turned (Bitcoin Gold to 144,5; Beam to 150,5; Flux to 125,4), and a chip was designed to follow "any set of parameters" by keeping the memory off die and the sort on it. A per-era draw of a parameter a chip can follow is a configuration to it; a draw of the memory SIZE is the one that forced the Z9's single die to fail on (144, 5), which is layer 2's mechanism, not layer 1's. Does v6 close it: yes by layer 2 (a 2 GiB set is not a die), and the parameter-following chip is the honest warning for layer 1's draws. The number: 12x at the first chip, 110x by 2023, against a hash whose 1,000x penalty claim never had a proof; Igneum's claim for its curve rests on a drawn curve (`chip-model-v3.md` 5.4) and the in-house pass's exact pebbling optimum (adv-cache-3), and still has no proof. + +### 2.6 Scrypt and Argon2: the Litecoin chips, the lookup gap, and MTP + +**Scrypt (Tenebrix and Litecoin, 2011; N = 1,024, r = 1, p = 1: a 128 KB scratchpad).** The mechanism of every scrypt chip is one 128 KB of on-die SRAM per hashing core and many cores per die: Watkins (2014) "there only needs to be 128 KB of memory per processor ... one core driving a 128 KB cache", SRAM chosen over every other memory as fastest, salsa20/8 about 60 percent of the runtime and memory access 38 percent (https://arxiv.org/pdf/2208.02160); the BM1485 of the L3+ (June 2017, 504 MH/s at 800 W): 12 cores per chip, "every BM1485 integrates on-die SRAM to hold that scratchpad", 28 nm, 288 chips per unit, about 1.7 MH/s per chip (so about 1.5 MB of SRAM per chip, arithmetic) (https://d-central.tech/mining-glossary/bm1485/ ; https://www.asicminervalue.com/miners/bitmain/antminer-l3-504mh); the L7 (November 2021, 9.5 GH/s at 3,425 W, USD 15,000, 0.36 W per MH against the L3+'s 1.58): the BM1489 at TSMC 7 nm, 480 chips (https://d-central.tech/mining-glossary/bm1489/ ; https://cryptoage.com/en/2550-bitmain-antminer-l7-is-a-new-asic-miner-for-litecoin-and-dogecoin.html); the Innosilicon A2 (21 April 2014, 28 nm, 1.6 to 1.8 MH/s per chip at 10 W, about 150 MH/s per box at 1 kW: https://www.design-reuse.com/news/34403/innosilicon-28nm-litecoin-asic-reference-miner.html); the KnC Titan (March 2014, 250 MH/s at 800 to 1,000 W, USD 9,995, "4 chips x 2,284 cores": https://www.coindesk.com/markets/2014/03/28/kncminer-updates-titan-spec-promises-250mhs/), whose 2,284 cores cannot each hold 128 KB on a 2014 die, so it shared scratchpads or took the time-memory trade-off (not confirmed). The trade-off itself is in the designer's record: Percival (18 November 2012) on storing every other scratchpad entry: memory halved for about 25 percent more BlockMix work, area-time about 0.625x, the trade favouring the attacker at 2x to 4x reductions, "already in the paper's cost estimates" (https://mail.tarsnap.com/scrypt/msg00092.html; all read 8 October 2026). Against an R9 280X at 700 to 740 kH/s and 340 to 450 W at the wall: A2 about 70x, L3+ about 300x, L7 about 1,300x per joule (derived, approximate). Scrypt-N (Vertcoin 2014: N doubling by timestamp up to 2^30) was dropped on 13 December 2014 for Lyra2RE "as a proactive defense against emerging Scrypt-N capable ASICs", because raising N "simply involves doing more iterations" and more SRAM or a larger lookup gap (https://vertcoinproject.org/vertcoin_whitepaper.pdf ; https://coincentral.com/what-is-vertcoin-a-beginners-guide/). + +**Argon2 as a proof of work (MTP, Zcoin, 10 December 2018).** Argon2d over 4 GB with a Merkle tree; Dinur and Nadler (2017): malicious proofs with under 1 MB, 1/3,000 of the honest memory, at a computation penalty of 170, "more than 55,000 times faster than what is claimed by the designers", with a 2^64 one-time precomputation, by injecting chosen blocks that steer Argon2d's data-dependent addresses (https://eprint.iacr.org/2017/497); MTP 1.2 patched it before launch; no Argon2 chip was ever built; Firo replaced MTP with FiroPoW on 26 October 2021 (https://firo.org/2021/10/01/firopow-and-instantsend-release.html). Yescrypt and yespower (GlobalBoost-Y 2014; Yenten, Cranepay, Tidecoin on yespower from 2018; L2-latency-bound sequential work: https://www.openwall.com/yespower/): no chip found, small prizes. + +**The mapping to v6.** Class B throughout, and the parameter-growth lesson for layer 2 in its oldest form: Scrypt-N's automatic growth was abandoned because a chip follows a scratchpad that grows by doubling the SRAM it already has, and Percival's own 2012 note says the time-memory trade favours the chip at 2x to 4x. Does v6 close it: yes by layer 2 (no die holds 2 GiB; the curve is monotone against partial stores); the lookup-gap lesson is the reason the dataset's chained cache has a drawn pebbling optimum (adv-cache-3) rather than a claim. The number: 1,300x for scrypt by 2021; Igneum's f = 0 chip at 1.86x per joule. MTP is class E (section 4.4). + +### 2.7 to 2.9 KawPow, Autolykos, Octopus: the no-chip hashes, and why + +| Hash | Live | Mechanism | Chip status, 8 October 2026 | Why no chip (the honest reading) | Source | +|---|---|---|---|---|---| +| KawPow (Ravencoin; Neoxa, Clore, Meowcoin, Neurai) | 6 May 2020 | ProgPoW 0.9.4 with a per-block program; "no additional future algorithm forks are envisaged" | none listed on WhatToMine or asicminervalue | class D on a small prize: Ravencoin's cap never reached the 2018 cluster's | https://whattomine.com/coins/234-rvn-kawpow/gpus ; https://github.com/RavenProject/Ravencoin/blob/master/roadmap/README.md | +| Autolykos v2 (Ergo) | February 2021 (v1 July 2019) | a k-sum (k = 32) over a Blake2b table of 2^26 elements of 31 bytes (2.08 GB) growing about 5 percent per 51,200 blocks from block 614,400 to a cap of 2,143,944,600 elements at block 4,198,400; v1's non-outsourceable puzzle removed because "large players could bypass this resistance using smart contracts" | none | class C territory (a table read per hash) on a small prize; its growth rule is the one automatic schedule in the record untested by a chip; f2pool and Ergo's own docs call it resistant with no chip named | https://docs.ergoplatform.com/mining/autolykos/ | +| Octopus (Conflux) | October 2020 | Ethash-style DAG (the "dense matrix step" unverified in the source) | none | as above; f2pool: "cannot be efficiently mined with FPGAs or ASICs" | https://f2pool.io/mining/guides/how-to-mine-conflux/ | +| Verthash (Vertcoin), FishHash (Iron Fish from April 2024, Karlsen) | January 2021; April 2024 | a 1.2 GB table from the chain's headers; a 4,608 MB constant dataset with Blake3 and 512 iterations | none | the same class as Ethash's chips, on prizes under the 2018 cluster | https://fips.ironfish.network/fips/fip-3-memory-hard-mining-algorithm | + +The reading for v6: "no chip" on a DRAM-scale random-read hash is an economic fact, not a design one (the 5 October file, section 2.5: compute-bound hashes got chips at USD 20 K to 30 K of daily issuance, Ethash at USD 7.6 M). Every hash in this table is class C and none of them has been tested at a prize that pays for a controller project; the Ethash record is the test, and it read 1x to 5x. + +### 2.10 kHeavyHash (Kaspa, November 2021): the KS chips + +**The mechanism.** cSHAKE256 of the header and nonce, a 64 x 64 matrix of 4-bit values generated from the pre-PoW hash, a nibble-wise matrix-vector multiply, XOR, a final cSHAKE (https://github.com/kaspanet/rusty-kaspa/blob/master/consensus/pow/src/lib.rs, read 8 October 2026); designed for optical and specialised hardware; the chip wires the multiply as a fixed pipeline. IceRiver KS0 (September 2023, 100 GH/s at 65 W), KS1 (1 TH/s at 600 W), Antminer KS3 (August 2023, 9.4 TH/s at 3,550 W), KS5 Pro (March 2024, 21 TH/s at 3,150 W), KS5L (April 2024, 12 TH/s at 3,400 W); against an RTX 4090 at 2.08 GH/s and 226 W: KS0 about 170x, KS5 Pro about 720x per joule (https://www.asicminervalue.com/miners/bitmain/antminer-ks5-pro-21th ; https://www.kryptex.com/overclocking/nvidia-rtx-4090-micron-24gb-medium-overclock ; read 8 October 2026). Timeline: 17 to 20 months to the first chip; the GPU share negligible by late 2023; seven forks left Kaspa to re-resist (the 5 October file, row 23). Node: not found. + +**The mapping to v6.** Class A. The matrix multiply is a warning for the reserve's mm8 family, not for the hash: a fixed 64 x 64 nibble multiply is the cheapest thing silicon does, and the measured rows agree (the 5090's int8 MAC at 1.4 to 4.1 pJ against a 5 nm array's claimed 0.04 to 0.4, `counter-asic-4-research.md` 15.1a). Does v6 close it: yes, closed since class v2 (no fixed function to unroll); the number, 720x, is what a fixed pipeline does to compute-bound work, and the reserve's ordering (section 4.3) keeps mm8 last for exactly this reason. + +### 2.11 CryptoNight (Bytecoin 2012, Monero 2014): the X3, the secret chips, and four forks + +**The mechanism of the hash.** 524,288 iterations of an AES round plus an 8-byte multiply over a 2 MB scratchpad sized to a 2014 per-core L3; latency-bound at SRAM scale. + +**The chips.** Bitmain Antminer X3, announced 15 March 2018 at 220 kH/s and 550 W (465 to 470 W measured), 180 BM1700 chips on three boards, USD 11,999 for batch 1 falling to USD 1,900; no teardown or vendor description of the BM1700 exists (node, SRAM, AES units not found). The arithmetic is the mechanism: 1.2 kH/s and about 2.6 W per chip is one or two 2 MB scratchpads in on-die SRAM with a hardware AES round per chip, and nothing else reaches that rate in that power (approximate, inference). Baikal Giant N, March 2018, 20 kH/s at 60 W. Against a Vega 64 at 2,009 H/s and about 200 W card power (about 10 H/W) the X3 is about 40x per joule; against a Threadripper 1950X at about 1,000 H/s and 185 W, about 75x (approximate) (https://www.asicminervalue.com/miners/bitmain/antminer-x3-220kh ; https://www.asicminervalue.com/miners/baikal/bk-n ; https://hothardware.com/reviews/monero-mining-with-amd-ryzen-threadripper ; read 8 October 2026). + +**The secret chips.** Monero's own 2018 review: by early 2018 "it was estimated that 80 to 90 percent of the network was specialized hardware" (https://web.getmonero.org/2019/02/12/2018-year-in-review.html); about half the hashrate (about 500 MH/s of 1,000) left at the v7 fork on 6 April 2018; Krawiec-Thayer's nonce study (24 November 2018) found half of all blocks with nonces in the lowest 0.002 percent of the space, patterns that "evaporated abruptly" at the fork (https://www.hackernoon.com/utter-noncesense-a-statistical-study-of-nonce-value-distribution-on-the-monero-blockchain-f13f673a0a0d). Vorick (13 May 2018, through secondary coverage): secret Monero ASIC mining "since early 2017, making up 50 percent of the hashrate". The timeline from the hash to the first secret chip is therefore about 33 months (April 2014 to early 2017), not the 43 the 5 October file gives (which counted to the fork); the announced chip came at 47. + +**The four forks and what each cost a chip.** v7 (6 April 2018): a one-byte tweak to the main loop; half the hashrate left; "a precaution and deterrent". v8 (18 October 2018, height 1,685,555): a whole-cache-line shuffle (4x the bandwidth demand) plus a 64:32 division and a 64-bit square root per iteration, 5 to 10 percent off CPU rate; the hashrate went from about 320 MH/s to just under 1,000 MH/s before March 2019 with chip nonce patterns visible from December: chips back inside two months, dominant inside four (https://decrypt.co/14421/ ; https://en.cryptonomist.ch/2019/05/08/mining-monero-hashrate-pow-change/). CryptoNight-R (9 March 2019, brought forward from April after the detection): a per-block random sequence of 60 to 69 integer instructions (63 on average; MUL 40 percent, XOR 23, SUB 12, ADD 12, ROR 8, ROL 6) over 9 registers, seeded by height so miners compile ahead; SChernykh's claim is a chip's minimum latency for the random math "at least 2.5 times higher" than the DIV plus SQRT it replaced (a chain of 15 multiplies against 6), with up to 1.5x more for an out-of-order chip; a hardware engineer in the PR thread estimated a chip could still do about 18 ns per iteration, comparable to a CPU (https://github.com/SChernykh/CryptonightR ; https://github.com/monero-project/monero/pull/5126). The hashrate fell from about 1 GH/s to 140 MH/s and settled at 300 to 350 MH/s; no CN-R chip is documented in its nine months. RandomX followed on 30 November 2019. + +**The mapping to v6.** CryptoNight is class B (SRAM-scale memory: layer 2 closes it, the 2 GiB dataset is 1,000 mm^2 of SRAM even at N5) and class D (the forks: v7 and v8 kept the machine's shape and the chips returned; CN-R changed what the machine had to be, a per-block random program, and no chip came in nine months, which is the per-epoch random program Igneum ships). The number: 40x per joule for the X3 against a GPU; the Igneum analogue, the f = 0 recompute chip that holds the 256 MiB cache on die, reads 0.92x at the op budget and 1.86x per joule (`chip-model-v3.md` 5.4), and CN-R's 2.5x latency claim is the fixed-shape mixer's 3x factor in the other direction. The honest residue: Monero's chips were found by nonce pattern four months after a fork at 85 percent of the hashrate; Igneum has no detector yet. + +### 2.12 X16R (Ravencoin, January 2018): a drawn order over fixed functions + +**The mechanism and the chips.** Sixteen hash functions in an order set by the previous block hash, a per-block automatic change with no fork. The sequencing changed; the sixteen primitives did not, so one large FPGA holding all sixteen cores needs only per-block routing: BittWare and SQRL's CVP-13 (Xilinx VU13P) was announced on 7 September 2018 naming "X17r, X16r and TimeTravel10" (https://www.cryptoninjas.net/2018/09/07/squirrels-research-labs-and-bittware-launching-new-fpga-crypto-mining-hardware/ ; https://www.bittware.com/cvp-13 ; read 8 October 2026). An unknown pool held 10 to 20 percent of blocks in February 2019, 30 to 40 in March, 45 by 11 July; the OW Miner OW1 (September 2019, 182 MH/s at 1,500 W, about USD 500) and SKC Turing R1 (680 MH/s at 800 W, about USD 1,500) were sold as ASICs, their insides unconfirmed; against a P102-100 at 35 MH/s and 219 W the OW1 is 0.75x and the R1 about 5x per joule (https://cryptoage.com/en/1782-asics-ow-miner-ow1-and-skc-miner-turing-r1-for-the-x16r-algorithm-exist.html ; https://miningboard.com/algorithms/X16R). X16Rv2 (1 October 2019) swapped one hash in; on 30 January 2020 Tron Black called the chips' return "evident"; KawPow followed in May 2020; Ravencoin's roadmap records "ASICs have been developed for X16R (and X16Rv2)" (https://cointelegraph.com/news/ravencoin-community-clash-over-mining-algorithm-continues ; https://github.com/RavenProject/Ravencoin/blob/master/roadmap/README.md). + +**The mapping to v6.** Class D in its purest form, and the warning for layer 1: a draw over a FIXED set is a configuration to a chip that holds the set. Igneum's op-mix draw is over twelve families a chip holds from genesis; what the draw does cost a chip is nothing, and what it costs the honest card is the per-vendor energy table. Does v6 close it: the fixed-function lane is closed by the per-epoch program (there is no sixteen-core pipeline to route), and the draw itself is a governance device. The number: 5x for the one X16R box with a plausible chip inside; the GPU-without-graphics sequencer's gain on Igneum's program side is bounded by that kind of figure, and its memory side by class C's 5.1x. + +### 2.13 Lyra2REv2 (Vertcoin, August 2015): FPGA first, then the Zig Z1 + +A memory-hard sponge (Lyra2 at T = 1, R = 8, C = 256, p = 1: a matrix small enough for on-die SRAM per core, on the order of 200 KB, approximate) inside a chain of hashes. The FPGA came first: an academic Lyra2 core on 16 July 2018 and a full FPGA miner at 2.6 to 3.7 MH/s and 323 to 432 nJ per hash, "significantly more energy efficient than both a GPU and a commercially available FPGA-based miner", which confirms commercial bitstreams before the chip (https://arxiv.org/abs/1905.08792); the Dayun Zig Z1 on 19 September 2018: 6.8 GH/s at 1,200 W, 28 nm, USD 8,000, "equivalent to 100 GeForce GTX 1080 Ti", about 20x per joule against a 1080 Ti at about 68 MH/s and 250 W (https://cryptoage.com/en/1231-first-asic-miner-lyra2rev2-dayun-zig-z1.html; read 8 October 2026). Vertcoin 0.14.0 forked at block 1,080,000 (1 February 2019) "to rid the network of the current generation of Lyra2REv2 ASICs and FPGAs" with Lyra2REv3 (R = 32, p = 4: 16x the memory), then Verthash in January 2021 (https://github.com/vertcoin-project/vertcoin-core/releases/tag/0.14.0). The 22 reorgs of October to December 2018 and the December 2019 attack came through rented hash on a hashrate the forks had reset (the 5 October file, row 7). Mapping: class B (layer 2 closes it) and lesson 5 (the fork reset the hashrate to a rentable size, which v6's layers 1 and 3 never do). The number: 20x; Igneum's f = 0 chip 1.86x. + +### 2.14 Eaglesong (Nervos, November 2019) and Blake3 (Alephium, November 2021): compute, embraced + +Eaglesong: a new sponge; the Toddminer C1 in February 2020 at 4 months, the Antminer K5 (April 2020, 1.13 TH/s at 1,580 W), the K7 at 63.5 TH/s and 3,080 W; about 100x per joule against an RTX 2080 Ti at about 1.5 GH/s and 220 W (approximate) (https://www.asicminervalue.com/miners/bitmain/antminer-k5-1130gh ; https://miningboard.com/algorithms/Eaglesong). Blake3: double Blake3, chosen as chip-friendly; the Goldshell AL-BOX (May 2024, 360 GH/s at 180 W), Antminer AL1 (15.6 TH/s at 3,510 W) and AL1 Pro (August 2024, 16.6 TH/s at 3,730 W), AL3 (8 TH/s at 3,200 W); about 100x to 220x per joule against an RTX 4090 at 6.0 GH/s and about 300 W (https://www.asicminervalue.com/miners/goldshell/al-box ; https://whattomine.com/asics/294-bitmain-antminer-al1-pro). Blake2s on Kadena: the Goldshell KD5 (March 2021, 18 TH/s at 2,250 W) and Antminer KA3 (166 TH/s at 3,154 W) against a GTX 1660 Super at 633 MH/s and 79 W: about 1,000x and 6,500x (https://www.asicminervalue.com/miners/goldshell/kd5 ; https://miningboard.com/algorithms/Blake%20%282s-Kadena%29). NexaPow (SHA-256 plus a secp256k1 Schnorr signature per nonce, "useful ASICs"): the DragonBall A21 (December 2024, 3.4 GH/s at 1,800 W, USD 6,999) is only about 2.2x per joule against an RTX 4090 at 320 MH/s and 380 W, because big-integer elliptic-curve arithmetic leaves a fixed pipeline little to strip (https://spec.nexa.org/mining/NexaPOW/ ; https://www.asicminervalue.com/miners/dragonball-miner/a21 ; all read 8 October 2026). Mapping: class A, closed since class v2; the numbers (100x to 6,500x) are what a fixed function costs, and NexaPow's 2.2x is the one compute-bound row where the chip's edge stayed small, because the work was already a wide multiplier chain, which is the shape of the ALU shadow's k band (0.3 to 0.8) in another chain's numbers. + +## 3. The three lessons, with the evidence + +**Lesson 1. The chip that stores the dataset is firmware-immune to every draw; only joules and memory growth move it.** Evidence: the five Ethash chips (section 2.1) never touched Ethash's compute and never needed to; Rao's audit said in 2019 that "the energy expended to move data from DRAM to compute are the same for GPU or ASIC" and that the only lever a chip has is the memory's own energy per bit, which Jasminer took by bonding the DRAM to the die; the Least Authority audit's conclusion that "the random math core likely prohibits the build of a light-evaluation based ASIC" was about the f = 0 chip and said nothing about the f = 1 chip, which is the one that shipped on Ethash in three forms. On Igneum's side the identity of `counter-asic-4-research.md` section 2 says it in one line: at zero premium the edge is E_card over E_mem and no hash change touches it. What v6 inherits: 5.1x per joule on GDDR7 at zero premium against the 5090 unlocked, 3.6x at its 1,300 MHz knee, 2.1x with the class v4 shadow at k = 1, and USD 2.8 against 14.7 per MH/s. What binds on v6: every per-era draw of layer 1 and every family epoch of layer 3 must be priced against this chip as a configuration change (a recompile per epoch, a new mapping per era) and never claimed as a cost to it; the public text's "under 2x" stays worded against the recompute chip, as the 6 October verdict already requires. + +**Lesson 2. Automatic change beats the human fork only where it costs the chip a redesign, and the one parameter that does is the memory.** Evidence: Grin's three Cuckaroo tweaks (17 July 2019, 15 January 2020, 16 July 2020) each changed the edge function and no chip ever shipped for the lane, but each was a hard fork with a new solver and the lane was scheduled to die; Monero's v7 and v8 kept the machine's shape and the chips were back inside four months, CN-R changed the machine and no chip came in nine months, RandomX changed it again and the X5 took 46 months; X16R's per-block order cost the FPGA nothing; Scrypt-N's public, slow growth was abandoned before a chip because the chip could be sized for years of it; Ethash's DAG growth is the one automatic rule in the record that killed a shipped chip, and it killed the one chip whose memory was sized to the card fleet's own limit (4 GB), 20 to 27 months after shipping. The compile and design cycles bound the race: Bitmain built the A3 in about 5 months and Halong the B52 in 9 (Vorick), a full Vivado compile on a mid-size part runs 42 to 160 minutes and hours on a large one (PRflow, FPT 2019), so an hourly program outruns every compile and a six-monthly change outruns no chip. What binds on v6: layers 1 and 3 close the governance failure (no fork, no hashrate reset to a rentable size, which cost Vertcoin two 51 percent attacks) and tax a sequencer chip die area, not architecture: Rao's own figure is about 1 M gates and 0.025 mm^2 at 10 nm for ProgPoW's whole inner loop, so a lane array carrying every reserve family is the class v4 shadow core's 30 mm^2 of N5 and USD 25 to 40, not a wall. Layer 2 is the one real lever, and its value is its floor and ceiling, not its tracking: the floor keeps class B closed (2 GiB is 1,000 mm^2 of SRAM at N5), and a ceiling under the honest tiers' memory is a requirement, because any rate that ages out a 32 GB chip board retires the 8 GB card first. + +**Lesson 3. A steered or biased address pattern is always found after launch unless the test lives in the acceptance rule, and every new draw needs its own null.** Evidence: Kik's exploit came five months after two audits that named the seed's keccak as a thing to scrutinise and three days after the EIP was declared dead; Dinur and Nadler found MTP's under-1 MB proof before launch only because the construction was published and reviewed, and the fix was a construction change; AP-F8-1 was found by the attack-pass lane one day after class v4 reached the devnet, in 96.6 percent of the class's programs, and took three sub-versions and class v5's floor to close; the in-house pass then found a program that passed every part of the rule and still read a live hot set, which is why the floor sits at 0.995 and not 0.98. What binds on v6: layer 4's test is a per-site ratio against the window model, and the window model is a function of the dataset size, the windows, the era stride and the read width, so a draw of any of those changes the null and the census must be re-derived per era (2.2 s per candidate at 2^20 on one box core; 0.7 percent of candidates refused at the floor); the residue the floor cannot reach without refusing most clean programs (the shadow-written concentrations at 0.9992 to 0.9997, about 1.0004x) moves with any draw of the shadow placement and needs its own ceiling per era, with the F8 gate's 1.2x-of-window shape. + +## 4. The four layers against the history, layer by layer + +### 4.1 Layer 1: per-era draws of the class parameters + +What is declared: the parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement) are drawn per era from chain state like the program. What the history says about each: + +| Parameter drawn | The precedent | What the draw costs a chip | What it costs the honest card and the verifier | Reading | +|---|---|---|---|---| +| Mixer round count (within the tested margin) | RandomX made the item derivation itself a random program so a chip could not hard-wire it (SuperscalarHash); CryptoNight-R's random math raised chip latency 2.5x | nothing on the f = 1 chip (it derives no item); on the f = 0 recompute chip the fixed shape is the 3x factor, and a drawn ROUND COUNT keeps the shape: the chip builds the widest count and gates the rest | the verifier's 10 ms gate caps the count (x8 is 2.1 ms per warp on the reference core, x16 about 3.7); the daily build 23 to 77 ms at x8 | a draw of the count within a margin the chip already covers is firmware; the lever against the recompute chip is a drawn SHAPE (the 5 October addition 2, reserve), and the recompute chip is not the one anyone builds (chip-model-v3 5.6) | +| Op-mix weights (within the measured safe band) | X16R drew the ORDER of sixteen fixed hashes per block and an FPGA served it at 1.3x within 20 months; ProgPoW drew the math per period and no chip exists on its adopters in eight years, on small prizes | a sequencer chip over the twelve families covers any weight table; the measured GPU cost per family is the real constraint (shfl 55.8 pJ against add 11.3 on the 5090: a shuffle-heavy draw taxes the card up to 5x per instruction with no better k) | the per-program hash-rate spread must stay under the 5 percent rule on every vendor across the band (the six-era spread was 1.3 to 3.2 percent on the era layout) | right as a governance device (no fork), neutral as a chip device; the band must be bounded by the per-vendor cost table of `counter-asic-4-research.md` 15.1a, not only by the rate spread | +| Read width | w16 moved the honest denominator 2.7 percent and the chip's cost not at all; w64 made the 5090 bandwidth-bound (71.9 MH/s); the 9070 XT pays a 64-byte line at every width | the chip pays the same 32-byte atom at w4 and w16; wider reads hand a custom controller the Ren-Devadas bandwidth lever (the Ethash chips' whole edge) | a 47 percent loss on the 5090 at w64 | the one parameter whose draw can move the memory physics the wrong way: keep the allowed set at {1} (as 1.13.1 already does) unless a wider width is measured latency-bound on all three vendors; a draw over {4 B, 16 B} is harmless and worthless | +| Program length | ProgPoW's loop count and Ethash's 64 accesses were fixed; RandomX's 8 chained programs of 256 instructions fixed; no chain drew its program length | a longer program is more shadow work per hash: the class v4 lever (the latency ladder) priced at 2.1x at k = 1; a chip builds the longest rung's core and idles it on short eras | the verifier rung 3 is inadmissible on the reference core (latency-ladder section 5); the card's premium per op 6.2 to 11.3 pJ measured | the draw must stay inside the admissible rungs (0 to 2); its value is the governance one (the ladder stepped by draw instead of by 90 percent signal), and the honest card pays the premium on every era | +| Shadow placement | no precedent in any chain; the per-load placement (16 blocks of 16 after every load) was DEAD as drawn on 7 October (acceptance in execution order accepts 1.4 percent of candidates; `counter-asic-4-research.md` 20.2a-close) | the per-load form would force the chip's ALU core inside every read's dependency (the USD 200 M break-even row, 16.2) if a sound form existed | compile-ahead at 16 sites; a class change | draw only over placements shown sound (today: the one block after the loads); the per-load form is the research item, not a draw value | + +### 4.2 Layer 2: the dataset's size tracking chain-state growth with a floor + +What is declared: the state-derived dataset's size tracks chain-state growth, with a floor, so fixed-memory silicon ages out. The history has exactly two scheduled memory-growth rules that ran against shipped hardware, and one of them killed a chip: + +| Precedent | The rule | What it did to chips | What it did to honest cards | Source | +|---|---|---|---|---| +| Ethash DAG growth | +8 MB per epoch of 30,000 blocks (about 0.7 GB a year; Rao's audit, slide "DAG size"); 1 GB at launch (July 2015), 3.0 GB by July 2019, 3.94 GB by November 2020 | the Antminer E3 (shipped July 2018, 4 GB of DDR3) ran out of DAG room on Classic at epoch 328 (about 3.56 GB, March 2020) and on Ethereum at about block 11.4 million (about October 2020) after a 30 March 2020 firmware stretched its DDR use: 20 to 27 months after shipping, 57 to 63 months after the hash went live; the A10 Pro (6 GB) and every later chip carried more memory than any card of its year and never aged out | the same rule retired 3 GB cards in 2018 and 4 GB cards by December 2020; Ethereum Classic cut the DAG to 2.47 GB (Thanos, ECIP-1099, block 11,700,000, 28 November 2020) to keep the 4 GB cards after the August 2020 51 percent attacks, and its own text says the fork kept "competitive mining across both GPU and ASIC hardware" | https://ethereumclassic.org/blog/2020-11-27-thanos-hard-fork-upgrade/ ; https://coingeek.com/memory-limitations-prompt-bitmain-antminer-e3-to-halt-etc-support/ ; https://cointelegraph.com/news/bitmains-antminer-e3-to-continue-mining-ether-with-new-update (all read 8 October 2026); Rao's audit, the DAG-size slide (https://github.com/ethcatherders/progpow-audit) | +| Autolykos v2 table growth | N = 2^26 elements of 31 bytes (2.08 GB) until block 614,400; then about 5 percent every 51,200 blocks (N doubles every 102,400 blocks, about 142 days at Ergo's 2-minute block) to a cap of 2,143,944,600 elements at block 4,198,400 (about 66 GB) | no chip has shipped for Autolykos as of today, on a small prize; the rule has never been tested against one | the table passes 8 GB in about 2 years of growth and 16 GB a year later (arithmetic on the rule); Ergo's GPU fleet thins by card memory on a schedule it chose | https://docs.ergoplatform.com/mining/autolykos/ (read 8 October 2026) | + +The arithmetic that binds. A chip's memory is bought by the board, and today's prices are the chip model's: GDDR7 about USD 20 per 2 GB device, so 32 GB on a 512-bit board is USD 320; one HBM3 stack 24 GB about USD 200 (`chip-model-v3.md` 5.1, September to October 2026 prices). Igneum's schedule as specified (2 GiB plus 0.5 GiB a year, doubling steps at years 4, 12, 28; spec 1.13.3) reaches 4 GiB at year 4 and 8 GiB at year 12. Against that schedule: + +| Memory the chip or card holds | Years until the dataset passes it | Who it is | +|---|---|---| +| 8 GB | 12 | the honest 8 GB card (the first tier out, `card-lifetime-2026-10-05.md`) | +| 12 GB | 20 | the honest 12 GB card | +| 16 GB | 28 | the 9070 XT class | +| 24 GB | 44 | one HBM3 stack; the 4090 and the M5 Max class | +| 32 GB | 60 | the f = 1 GDDR7 chip's board; the 5090 | + +So layer 2 as a rate "tracking chain state" ages out fixed-memory silicon only if the dataset grows faster than a chip generation's memory headroom, and every rate that does that retires the honest small cards first by the same table. The E3 is the only case in the record where a growth rule beat a chip, and it beat a chip that had under-provisioned memory by a factor the card fleet also hit (4 GB). The rule that would hurt the f = 1 chip is one that keeps the dataset above what one board of commodity DRAM holds at the chip's price point, and that rule is unaffordable for the honest fleet. The honest reading: layer 2 is the right lever class (the memory is the one parameter a stored-dataset chip cannot read as firmware), and its value is set by the floor and the ceiling, not by the tracking: a floor keeps the dataset above every SRAM die (class B stays closed: 2 GiB is 1,000 mm^2 of SRAM even at N5), and a ceiling keeps it under the honest tiers' memory. Between those two lines the chip's board holds whatever the card holds, and the growth rate changes nothing for it. What "tracking chain state" adds over the fixed schedule is governance (no release decides the size) and the class v5 link (the dataset is built from the state, so the size follows the state's record count naturally); it is not an anti-chip rate. Open for the synthesis lane: if the chain's state grows the way Ethereum's did (approximate, from memory: Ethereum's account and storage state passed 100 GB in its eighth year), a dataset that tracks it literally outgrows every consumer card inside the chain's first decade, so the ceiling is the number to fix, in gigabytes per tier per year, before the rate. + +### 4.3 Layer 3: scheduled family epochs by height, every 180 days, no release + +What is declared: a new instruction family goes live on a height schedule, every 180 days by default, with no release (the reserve of spec 1.13.2, ordered at genesis). The record of scheduled change against chips: + +| Chain | The change and its cadence | Human release needed | What it cost the chip | Outcome | Source | +|---|---|---|---|---|---| +| Grin, Cuckaroo lane | a new tweak of the edge function every six months (Cuckarood July 2019, Cuckaroom January 2020, Cuckarooz July 2020), each a hard fork; the lane's reward share scheduled from 90 percent to zero by January 2021 | yes, every time | a new edge function per tweak: a redesign, not a configuration | no chip ever shipped for the tweaked lane; the chip lane (Cuckatoo31+) got the iPollo G1 at about 4x in 23 months | the 5 October file rows 18 and 19, [S61] to [S66] | +| Monero | four algorithm forks in 20 months (v7 April 2018, v8 October 2018, CN-R March 2019, RandomX November 2019) | yes, every time | v7 and v8: a re-spin (85 percent of the hashrate vanished at v7 and chips were back at 85 percent four months after v8); CN-R: random math per block, chip latency up 2.5x; RandomX: a new class of machine, parity hardware after 46 months | the forks were events; the chips survived the ones that kept the shape and died on the one that changed the machine | rows 16 and 17; Vorick on a fork-surviving chip, [S47] | +| Ravencoin X16R | the ORDER of sixteen fixed hashes drawn per block from the previous block hash; no fork needed | no | nothing: a sixteen-core sequencer reads the order as a configuration | an FPGA at 1.3x within 20 months; the X16Rv2 fork (one hash swapped) was answered by bitstreams within weeks | row 9 | +| Ethereum ProgPoW | the random math re-drawn every PROGPOW_PERIOD (50 blocks in 0.9.2, 10 blocks in 0.9.3, about 2 minutes); the op set fixed (add, mul, mulhi, min, rotl, rotr, and, or, xor, clz, popcount) | no | a sequencer over the eleven ops and a 32-register file; the audit priced the conventional compute chip at little gain and the on-die cache chip at "<< 0.1x" energy per hash | never deployed on Ethereum; no chip on KawPow or FiroPoW in five to six years on small prizes | https://eips.ethereum.org/EIPS/eip-1057 and https://github.com/ifdefelse/ProgPOW (read 8 October 2026) | +| Igneum class v2 to v6 | a new program every epoch; era draws every 180 days; one reserve family per era | no | a recompile per epoch; the reserve families are in the shipped generator from genesis, so a chip that reads the reserve at genesis carries every family's datapath from day one | the governance failure is closed; the chip is not | this file | + +The clocks that bound the race (all read 8 October 2026): Bitmain built the A3 "in about 5 months" and Halong the B52 "in about 9" (Vorick, through https://www.nextbigfuture.com/2018/05/obelisk-explains-the-state-of-asics-and-crypto-mining.html); KnC taped out a 20 nm part "only 3 months after starting the project" in 2014 (https://www.design-reuse.com/news/34090/20nm-asic-for-bitcoin-mining.html); ASICMiner went from founding in July 2012 to 64-chip boards on 31 January 2013 (Taylor, IEEE Computer 2017, https://michaeltaylor.org/papers/Taylor_Bitcoin_IEEE_Computer_2017.pdf); Linzhi from founding (February 2018) to tested boards (December 2020) took 27 months; Rao's 10 nm-class project "1+ year". A full Vivado compile on a mid-size Xilinx part (ZCU102) runs 42 minutes typical and 160 worst (PRflow, FPT 2019, https://ic.ese.upenn.edu/abstracts/prflow_fpt2019.html) and "several hours" on a large device (a 2021 Paderborn talk); the Least Authority audit called a 2-minute period "impractical" for a bitstream; the one KawPow FPGA on record is an academic VU35P build at 5.6 MH/s (NTU 2022, https://tdr.lib.ntu.edu.tw/handle/123456789/84103?locale=en), under an eighth of a 2022 GPU. So an hourly program outruns every compile and every design cycle; a 180-day family epoch outruns no chip's design cycle and does not need to, because the family is in the generator from genesis. + +Reading. Scheduled change beat chips in exactly one shape: Grin's, where each change was a new function nobody could know in advance, and even then only because the lane was built to die. Every change a chip could read at genesis (X16R's order, ProgPoW's period, Igneum's reserve) became firmware. Layer 3 as declared is X16R's and ProgPoW's shape, automated and spaced at 180 days: it removes the fork (lesson 5 of the 5 October file, and the one thing that cost Vertcoin two 51 percent attacks), and it taxes a sequencer chip die area for families not yet live. The honest number for that tax is small: the reserve's candidates are integer ALU operations (shifts, bit-field extract, andn, byte permute, popcount, select, the second shuffle form; mm8 last), each a few thousand gates per lane; against the f = 1 chip's USD 470 of memory and board, a lane array carrying every reserve family is the same 30 mm^2 of N5 the class v4 shadow already forces (`counter-asic-4-research.md` 16.1: USD 25 to 40 of die). The one family that is not cheap for a chip to carry idle is one whose unit is large (mm8's tile engine), and that is exactly the family the measured rows say not to use for forcing (the 5090's int8 MAC at 1.4 to 4.1 pJ against a 5 nm array's claimed 0.04 to 0.4: `counter-asic-4-research.md` 15.1a). So layer 3 is right for governance and neutral for the chip; what would make a family epoch cost a chip a redesign is a family whose semantics are not knowable at genesis, which is the random item-derivation program of the 5 October addition 2 (a per-day SuperscalarHash-style derivation), the one RandomX idea Igneum has not taken, and it acts on the f = 0 chip only. + +### 4.4 Layer 4: the (c''') floor and the F8-form uniformity test per era draw + +What is declared: the (c''') acceptance floor (the per-site distinct-index ratio at or above 0.995 against the window model, spec 1.4.7.2) and the F8-form uniformity test generalised to each era's parameter draw, with a redraw on failure. The history has three attacks of this one class, and in every case the test that would have caught it lived outside the acceptance rule: + +| Attack | The steer | Found when | What the fix was | Source | +|---|---|---|---|---| +| Kik's ProgPoW exploit (4 March 2020) | the 64-bit seed between the two keccak passes: fix a seed, compute its mix once, then grind nonces until keccak_progpow_64(header, nonce) equals the seed; the memory path is run once per 2^64 nonces, so a chip that never touches the DAG wins once the difficulty passes 2^50 | after two tentative approvals and five months after both audits (September 2019) | ProgPoW 0.9.4 widened the carried state from 64 to 256 bits (the digest of the first keccak, plus the mix, plus padding) | https://github.com/kik/progpow-exploit and https://github.com/ifdefelse/ProgPOW (read 8 October 2026) | +| Dinur and Nadler on MTP (2017) | the prover controls the memory's contents, so by injecting blocks it steers Argon2d's data-dependent addresses into a set it can hold in under 1 MB at a 170x compute penalty, in place of 2 GB | before launch, by cryptanalysis, not by the design's own test | MTP 1.2 patched the construction | the 5 October file [P18] [S38] | +| AP-F8-1 on class v4 (7 October 2026) | a load site whose source register was last written by a lossy op (or, mul, mulhi) saturates to all-ones or zero at a known rate, and the era map sends the constant to one item; 96.6 percent of class v4 programs carried a lossy-sourced load; the worst seed read 29x the window model on one item | in the attack-pass lane's census, one day after the class shipped to the devnet | sub-versions 1 to 3 (the freshness fixpoint, the executed shadow block, the (c'') ratio at 0.98), then class v5's (c''') at 0.995 | `docs/plans/counter-asic-3-status.md` section 7c; spec 1.4.7.2 | + +Reading. Layer 4 is the right answer to this class and the only one of the four layers that acts on a mechanism the record shows beating hashes after launch. Two things bind on it. First, the test can only see what its null models: (c''') measures distinct-index ratios against the window model, which is a function of the dataset size, the per-site windows, the era stride and the read width; a draw of the read width or the program length changes the null, so "generalised to each era's draw" means the window model is re-derived per era and the census re-run per draw, not one floor reused. The cost is known: 2.2 s per candidate at 2^20 on one box core, once per epoch draw on a node, and 0.7 percent of candidates refused by the floor (spec 1.4.7.2). Second, the residue the floor cannot reach without refusing most clean programs is the shadow-block-written concentrations at 0.9992 to 0.9997, worth about 1.0004x to a chip (spec 1.4.7.2), and a per-era draw of the shadow placement moves that residue, so the redraw rule needs its own ceiling stated (the 1.2x-of-window gate of F8 is the right shape; the number per era is the census's to set). The Kik lesson is separate and already closed: Igneum's seed is 256 bits through the VDF and the program is the epoch's, so there is no 64-bit state to grind; the header-locality search (the 5 October check 1) remains the nearest analogue and was measured by adv-accept-2 in the in-house pass. + +## 5. What to add, optimise or invent, ranked (first cut; the full report re-ranks with the other lanes' findings) + +| Rank | What | Why the history says so | Cost to the honest card | Where | +|---|---|---|---|---| +| 1 | **State layer 2's floor and ceiling in gigabytes per tier, before its rate.** The floor above every SRAM die (today's 2 GiB holds); the ceiling under the honest tiers' memory on a stated glide (the 8 GB tier's life is the first number) | the E3 is the only chip a growth rule ever killed and it was the chip with the fleet's own memory limit; Scrypt-N was abandoned because its growth was public and slow; Autolykos's growth is untested; a dataset that tracks chain state literally outgrows every card if the state grows the way Ethereum's did (approximate) | none at the floor; everything at the ceiling | genesis rule, with the card-lifetime table | +| 2 | **The clock and the detector**, unchanged from the 5 October ranking and still unbuilt: the per-program rate spread and nonce pattern on the observer (the method that found Monero's chips at 85 percent), plus a share-by-key-and-template instrument (what found Qubic until it randomised), plus the issuance trigger at about USD 50 K a day | every chip in the record was on its chain before it was announced (Monero 2017, Zcash's three groups, SChernykh's 2021 reading of the X5) | none | before the public testnet | +| 3 | **Keep the read width out of the era draw** unless a width other than 4 bytes is measured latency-bound on all three vendors; a draw over {4 B, 16 B} is harmless and worthless (w16 moved the chip's cost not at all) | the Ethash chips' whole edge was the bandwidth lever Ren and Devadas name; w64 made the 5090 bandwidth-bound | a 47 percent loss on the 5090 at w64 | spec 1.13.1's allowed set stays {1} | +| 4 | **Bound the op-mix draw by the per-vendor energy table, not only by the rate spread**: the 5090 pays 55.8 pJ per shuffle against 11.3 per add, so a shuffle-heavy era taxes the honest card up to 5x per instruction for no better k | X16R's drawn order cost the chip nothing and the fleet nothing; Igneum's draw can cost the fleet watts | up to 2x the premium per instruction at the band's edge | the band's definition in the class v6 spec | +| 5 | **Order the reserve by what a sequencer cannot fold into firmware**, mm8 last; and name the random item-derivation program (the 5 October addition 2) as the one reserve item whose semantics are not knowable at genesis | the kHeavyHash chips and the 5090's own 1.4 to 4.1 pJ per int8 MAC; RandomX's SuperscalarHash is the one idea Igneum has not taken, and it acts on the f = 0 chip only | none at launch | reserve ordering, genesis | +| 6 | **Generalise layer 4 with a null per drawn parameter**: the window model re-derived per era, the census per draw, and a stated ceiling for the shadow-written residue per shadow placement | lesson 3 | 2.2 s per candidate once an epoch on a node | the class v6 acceptance rule | +| 7 | **The per-load shadow placement as a research item, not a draw value**, until a sound construction is drawn (the 16 x 27 form accepted 1.4 percent of candidates) | it is the one placement that would force the chip's ALU core inside every read's dependency (the USD 200 M break-even row) | compile-ahead at 16 sites | `counter-asic-4-research.md` 20.2a | + +## 6. Consequences per user tier + +| Tier | What this history means for it | What is being done | +|---|---|---| +| Home miner, one 8 GB card | Layer 2 is the one layer whose rate reaches this tier first: at 0.5 GiB a year the 8 GB card is out at year 12, and any rate fast enough to age out a 32 GB chip board is out of this card's life inside two years. No chip of class A, B or E touches this miner under v6; the class C chip (the memory system without the GPU) reaches it as it reached Ethash's 4 GB miners: by price per MH/s, 5x | the layer 2 ceiling stated per tier (section 4.2) before the rate; the issuance clock and the detector (section 5) | +| One 12 GB or 16 GB card | as the 8 GB tier with 20 and 28 years on the schedule; the 9070 XT's 10.6 microjoules per hash is 23x behind the GDDR7 chip per joule on the model, so this tier's card is the first the class C chip displaces | nothing in the hash fixes AMD's dependent-read rate; the vendor-share metric is the warning | +| One 24 or 32 GB card (5090, M5 Max) | the honest best: 1.69 microjoules at the 5090's knee against the chip's 0.47; 3.6x at zero premium, 2.1x with the shadow at k = 1; the M5 Max at 0.78 microjoules is 1.7x behind the GDDR7 chip with no shadow at all | the operating point as the shipped default (Ember); the shadow's rung | +| A rig | the Ethash precedent in full: chips at 2x to 5x per joule and 5x per dollar took the hashrate over four years, and the ASIC share stayed small only while the chips were not cheap enough at scale; the model says this chip is (USD 2.8 against 14.7 per MH/s) | the break-even cap row (USD 100 M in years 1 to 2 with the N5 shadow core) is the real wall; the clock | +| A pool user | MoneroCrusher found chips at 85 percent of Monero's hashrate by the share pattern, four months after a fork; Igneum's detector is the same method on the observer, unbuilt | the detector before the public testnet (section 5) | +| Every tier, on governance | no fork, ever, for a draw or a family: the history's clearest lesson (Vertcoin's two 51 percent attacks after forks, Monero's four, Ethereum's two-year ProgPoW fight) is the one v6's layers 1 and 3 close outright | nothing further | + +## 7. Unverified and owed + +- The research lanes' returns for the chip mechanism columns (section 1 and 2) are the first cut's owed content; where a lane reports "not found" the row says so and carries the 5 October file's figure. +- The Vorick post (13 May 2018) is read through secondary coverage (davidgerard.co.uk, zycrypto.com, nextbigfuture.com, a steemit copy; read 8 October 2026): the secret-Monero-ASIC claim ("since early 2017, making up 50 percent of the hashrate"), the three Zcash groups, the Equihash fork-following architecture, "about 5 months" for Bitmain's A3 and "about 9 months" for Halong's B52, the A3 under USD 10 M with USD 20 M of orders in eight minutes, the manufacturer withdrawal that cost Obelisk "north of USD 2 million". The 5 October file's "13 months for a startup" and "a chip able to survive Monero's forks at under a 5x hit" were NOT found on any fetched page and are carried here as unverified; the Monero fork-survival fact that is verified is the record itself (chips back inside four months of v8). +- The KawPow fork block and its 3-block period are approximate (the minerstat and Tron Black pages answered 403). +- The Ethereum DAG date of passing 4 GB on the main chain (about December 2020) is approximate; the Classic figures (3.94 GB at epoch 376, 27 November 2020) are cited. +- The Ethereum state-size figure in section 4.2 is from memory, approximate, and is the open question handed to the synthesis lane. +- Every per-joule ratio for a chip against a GPU is arithmetic on the cited rate and watt figures of both and is approximate by construction. +- Nothing here is a measurement; the Igneum figures are the repo's measured rows as cited, and the chip figures are the chip model's, modelled. + +## 8. Sources + +Every URL is cited inline at the row that uses it, with "read 8 October 2026" at the row or the section; the 5 October file's [S], [P], [E] and [L] lists are cited by their tags and not repeated. The primary documents read in full by this lane (text extracted where the page is a PDF): EIP-1057 (https://eips.ethereum.org/EIPS/eip-1057); the ifdefelse ProgPOW README (https://github.com/ifdefelse/ProgPOW); the Least Authority audit (https://leastauthority.com/static/publications/LeastAuthority-ProgPow-Algorithm-Final-Audit-Report.pdf, report version 9 September 2019); Bob Rao's hardware audit (https://github.com/ethcatherders/progpow-audit, 6 September 2019); Kik's exploit (https://github.com/kik/progpow-exploit); RandomX design.md, design_v2.md, specs.md, PR 317, release v2.0 and issue 11 (https://github.com/tevador/RandomX); Tromp's README and the Grin forum threads named in 2.4 (https://github.com/tromp/cuckoo ; https://forum.grin.mw); the Ergo Autolykos docs (https://docs.ergoplatform.com/mining/autolykos/); the Thanos post (https://ethereumclassic.org/blog/2020-11-27-thanos-hard-fork-upgrade/); the TechInsights Jasminer notes (https://www.techinsights.com/ko/node/51986 and /52149); the Fudan NDSS 2019 paper (https://www.ndss-symposium.org/wp-content/uploads/2019/02/ndss2019_09-5_Bai_paper.pdf); Percival's lookup-gap note (https://mail.tarsnap.com/scrypt/msg00092.html); Lee and Kim on Qubic (https://arxiv.org/html/2512.01437v2); PRflow (https://ic.ese.upenn.edu/abstracts/prflow_fpt2019.html). Papers of 2024 to 2026 found: Blocki and Smearsoll, "Provably memory-hard proofs of work with memory-easy verification", ePrint 2025/1456 (Omega(N^2 / log N) cumulative memory with polylog verification: https://eprint.iacr.org/2025/1456); Condrey, PoSME, arXiv 2604.15751; Yang et al., PHICOIN, arXiv 2412.17979 (a resistance claim with no algorithm in the abstract). Pages that refused every lane (403, 404, DNS): Vorick's original post on Medium and sia.tech and its archive copy; Linzhi's and ifdefelse's Medium posts; MoneroCrusher's Medium post (figures taken from criptonoticias coverage); bitmain.com's product list; support.bitmain.com's Z9 page; minerstat; Tron Black's posts; innosilicon.global; jasminer.com (an empty shell); cryptomining-blog.com; the Yole DBI report. The session's web-search budget ran out at 11:0x UK; everything after that is direct fetches of known URLs, and "not found" in this file means not found on a fetched page. From eca96c4799f8e3fc65f73732a1ce63eb0a937b2a Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:15:45 +0000 Subject: [PATCH 14/19] Bench table: the fleet's watts are class v3 (the sweep lane's read of its pack, 8 October 2026: program_class v3, generator 3): every model-sweep row and every 7 October fleet row carries its watts as "(class v3)", its MH per watt hollow and unranked, the per-watt lede claims drawn from class v4 rows only, the sweep line stating the fact and the measured gap (a 5070 Ti at 141 W on v3 and 224 W on v4 at one rate); the rates stand (memory-bound, the same under both) Co-Authored-By: Claude Fable 5.1 --- site/build.mjs | 7 +- site/miner-bench.json | 227 ++++++++++++++++++++++++++++-------------- site/miners.html | 8 +- 3 files changed, 161 insertions(+), 81 deletions(-) diff --git a/site/build.mjs b/site/build.mjs index f1dcd61db..f891e01c6 100644 --- a/site/build.mjs +++ b/site/build.mjs @@ -434,6 +434,7 @@ for (const [file, active] of PAGES) { const tunedState = (r) => { const v = r.tuned || 'stock, mining'; if (/Ember Tune|core lock/.test(v)) return ['tuned', 'Tuned']; if (/no lever/.test(v)) return ['nolever', 'No lever']; return ['stock', 'Stock']; }; const effCell = (r) => { if (r.mh_per_w == null) return ['not measured', -1]; + if (r.watts_class === 'v3') return ['○ ' + fmt3(r.mh_per_w) + '', -1]; if (r.watt_basis === 'chip') return ['○ ' + fmt3(r.mh_per_w) + '', -1]; return [(best && r === best ? '' : '') + fmt3(r.mh_per_w) + '', r.mh_per_w]; }; @@ -473,12 +474,14 @@ for (const [file, active] of PAGES) { const earlierTable = render(earlier, 'bench-earlier'); // the lede recomputes from the data (8 October 2026): the best desktop card by rate, then the best stock and the best tuned // MH per wall watt as separate claims, so a rented stock row never reads as a tuned one and a tuned lock never reads as stock - const ranked = (list) => list.filter(r => r.mh_per_w != null && r.watt_basis !== 'chip').sort((a, b) => b.mh_per_w - a.mh_per_w); + // a row whose watts were read under another class (watts_class 'v3': the fleet's bench pack, 7 and 8 October 2026) is ranked by rate, + // never by MH per watt; the per-watt claims come from rows whose watts are class v4 + const ranked = (list) => list.filter(r => r.mh_per_w != null && r.watt_basis !== 'chip' && !r.watts_class).sort((a, b) => b.mh_per_w - a.mh_per_w); const bestStock = ranked(cur.filter(r => tunedState(r)[0] === 'stock'))[0]; const bestTuned = ranked(cur.filter(r => tunedState(r)[0] === 'tuned'))[0]; const name = (r) => esc(r.card.replace(/ \(.*$/, '')); // the rented model sweep of 8 October 2026 (23 rows on 22 models, every class v5 fingerprint matched): its two page-level findings - const sweepLine = `

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found class v4's rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held class v4's rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it; the table's rates are class v4.

`; + const sweepLine = `

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found the rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held the rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it. The sweep's watts, and the 7 October fleet rows' watts, are class v3 figures: the fleet's bench pack ran the class v3 program, and class v4 draws more for the same rate (a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4, 8 October 2026). Those rows are ranked by rate alone; the per-watt claims above come from rows measured under class v4; the class v4 watts rerun lands on each row with its date, the v3 watts kept beside it as the shadow premium per card.

`; const bestLine = best ? `

Best desktop card: ${name(best)}, ${esc(fmt(best.mh_s))} MH/s (measured, ${esc(best.date)}).` + (bestStock ? ` Best stock MH per wall watt: ${name(bestStock)}, ${esc(fmt3(bestStock.mh_per_w))} at ${esc(fmt(bestStock.watts))} W, untuned (${esc(bestStock.by)}, ${esc(bestStock.date)}).` : '') + (bestTuned ? ` Best tuned: ${name(bestTuned)}, ${esc(fmt3(bestTuned.mh_per_w))} at ${esc(fmt(bestTuned.watts))} W (measured, ${esc(bestTuned.date)}).` : '') + '

' : ''; diff --git a/site/miner-bench.json b/site/miner-bench.json index 3673fdc95..05de69e18 100644 --- a/site/miner-bench.json +++ b/site/miner-bench.json @@ -281,7 +281,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep, a rented card)", "by": "measured by the fleet", - "note": "424.0 W maximum; 98 percent of its random-read ceiling like the 5090; 1.78x the 5090's hash at 1.15x the tuned 5090's hash per watt and a third of the hash per rented dollar", + "note": "424.0 W maximum; 98 percent of its random-read ceiling like the 5090; 1.78x the 5090's hash at 1.15x the tuned 5090's hash per watt and a third of the hash per rented dollar. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04", @@ -291,7 +291,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "385.6 (class v3)" }, { "card": "NVIDIA RTX 5080 (16 GB)", @@ -328,7 +330,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 595.71.05, Ubuntu 24.04", @@ -338,7 +340,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "178.3 (class v3)" }, { "card": "NVIDIA RTX 3070 (8 GB)", @@ -394,7 +398,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "283.0 W maximum", + "note": "283.0 W maximum. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580.65.06, Ubuntu 24.04", @@ -404,7 +408,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "267.3 (class v3)" }, { "card": "NVIDIA RTX 3090 (24 GB)", @@ -438,7 +444,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580.65.06, Ubuntu 24.04", @@ -448,7 +454,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "249.5 (class v3)" }, { "card": "NVIDIA RTX 4090 (24 GB)", @@ -482,7 +490,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "114.5 W maximum; the Devnet 3 boxes read 23.7 to 25.7 MH/s beside their nodes", + "note": "114.5 W maximum; the Devnet 3 boxes read 23.7 to 25.7 MH/s beside their nodes. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04", @@ -492,7 +500,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "111.6 (class v3)" }, { "card": "NVIDIA RTX 3060 Ti (8 GB)", @@ -504,7 +514,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -514,7 +524,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "129.5 (class v3)" }, { "card": "NVIDIA RTX 4060 Ti (8 GB)", @@ -526,7 +538,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -536,7 +548,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "77.5 (class v3)" }, { "card": "NVIDIA RTX 4070 Ti (12 GB)", @@ -548,7 +562,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -558,7 +572,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "107.3 (class v3)" }, { "card": "NVIDIA RTX 5060 (8 GB)", @@ -570,7 +586,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -580,7 +596,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "75.4 (class v3)" }, { "card": "NVIDIA RTX 5070 (12 GB)", @@ -592,7 +610,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -602,7 +620,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "102.8 (class v3)" }, { "card": "NVIDIA RTX 5070 Ti (16 GB)", @@ -614,7 +634,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -624,7 +644,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "145.6 (class v3)" }, { "card": "NVIDIA RTX A5000 (24 GB)", @@ -658,7 +680,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -668,7 +690,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "240.7 (class v3)" }, { "card": "NVIDIA A100 PCIe (80 GB)", @@ -680,7 +704,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -690,7 +714,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "299.6 (class v3)" }, { "card": "NVIDIA A100 SXM (80 GB)", @@ -702,7 +728,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -712,7 +738,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "266.3 (class v3)" }, { "card": "NVIDIA H200 SXM (141 GB)", @@ -724,7 +752,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -734,7 +762,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "432.9 (class v3)" }, { "card": "NVIDIA B200 (180 GB)", @@ -746,7 +776,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -756,7 +786,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "855.6 (class v3)" }, { "card": "NVIDIA RTX PRO 6000 Blackwell (96 GB)", @@ -768,7 +800,7 @@ "date": "2026-10-07", "source": "bench log: 7 October 2026, every rentable card on the hash (the fleet's sweep on rented cards; hive package 0.3.20, Ubuntu 24.04)", "by": "measured by the fleet", - "note": "", + "note": ". Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, not tuned)", "driver_os": "NVIDIA driver 580, Ubuntu 24.04", @@ -778,7 +810,9 @@ "pl_w": null, "label": "stock (no measured tune point; the 5080 and 9070 XT passes and the class v4 efficiency pass land theirs when read)" }, - "group": "datacentre" + "group": "datacentre", + "watts_class": "v3", + "watts_display": "288.7 (class v3)" }, { "card": "NVIDIA RTX 5070 Ti (16 GB)", @@ -790,7 +824,7 @@ "date": "2026-10-08", "source": "model sweep 2026-10-08, row 1 (a rented card, result cb2-rtx-5070-ti-16-gb-7324-result.json; 09:23Z)", "by": "measured by the fleet", - "note": "rented single-card host, 0.06 h; power.draw mean 135.9 W (instant 139.0, max 140.0, limit 300 W), SM 2,795 MHz mean, memory 13,801 MHz, 47 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the same rate on this card (78.72 MH/s, fingerprint matched); read ceiling 79.6 MH/s, so the hash runs at 0.99 of it", + "note": "rented single-card host, 0.06 h; power.draw mean 135.9 W (instant 139.0, max 140.0, limit 300 W), SM 2,795 MHz mean, memory 13,801 MHz, 47 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the same rate on this card (78.72 MH/s, fingerprint matched); read ceiling 79.6 MH/s, so the hash runs at 0.99 of it. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "v4_cost": "not measured (class v4 program only)", "tuned": "stock, bench only (rented, no clock control on the host)", "driver_os": "NVIDIA driver 595.84, Ubuntu 24.04 (CUDA 12.8.1 image)", @@ -800,7 +834,9 @@ "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" }, - "group": "buy" + "group": "buy", + "watts_class": "v3", + "watts_display": "135.9 (class v3)" }, { "generator": "v2", @@ -822,7 +858,9 @@ "mh_per_w": 0.289, "driver_os": "NVIDIA driver 615.71.09, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 2 (a rented card, result cb2-rtx-4070-ti-super-7325-result.json; 09:26Z)", - "note": "rented single-card host, 0.10 h; power.draw mean 142.6 W (max 145.2, limit 285 W), SM 2,885 MHz mean, memory 10,251 MHz, 47 C, the SW thermal bit set with no rate effect (three runs within 0.02 percent); self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (41.37 MH/s, fingerprint matched); read ceiling 41.6 MH/s, the hash at 0.99 of it. The Ada 192-bit bus class: the same MH per watt as the 4070 Ti; the extra 4 GB buys headroom, not rate" + "note": "rented single-card host, 0.10 h; power.draw mean 142.6 W (max 145.2, limit 285 W), SM 2,885 MHz mean, memory 10,251 MHz, 47 C, the SW thermal bit set with no rate effect (three runs within 0.02 percent); self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (41.37 MH/s, fingerprint matched); read ceiling 41.6 MH/s, the hash at 0.99 of it. The Ada 192-bit bus class: the same MH per watt as the 4070 Ti; the extra 4 GB buys headroom, not rate. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", + "watts_class": "v3", + "watts_display": "142.6 (class v3)" }, { "generator": "v2", @@ -844,7 +882,9 @@ "mh_per_w": 0.201, "driver_os": "NVIDIA driver 580.65.06, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 3 (a rented card, result cb2-rtx-3080-ti-12-gb-efll-result.json; 09:28Z)", - "note": "rented single-card host, 0.04 h; power.draw mean 293.5 W (max 299.6, limit 320 W), SM 1,936 MHz mean, memory 9,251 MHz, 74 C, the SW power-cap bit set 26 W under the limit; self-test PASS (96 of 96 vector lanes); the class v5 kit runs 2.5 percent faster on this card (60.38 MH/s, fingerprint matched); read ceiling 58.9 MH/s, the hash at 1.00 of it. Matches the 7 October row (58.95 MH/s) across hosts; the worst MH per watt of the sweep and the cheapest MH per rented hour, so a buyer on a power bill prefers the 5070 Ti and a renter this card" + "note": "rented single-card host, 0.04 h; power.draw mean 293.5 W (max 299.6, limit 320 W), SM 1,936 MHz mean, memory 9,251 MHz, 74 C, the SW power-cap bit set 26 W under the limit; self-test PASS (96 of 96 vector lanes); the class v5 kit runs 2.5 percent faster on this card (60.38 MH/s, fingerprint matched); read ceiling 58.9 MH/s, the hash at 1.00 of it. Matches the 7 October row (58.95 MH/s) across hosts; the worst MH per watt of the sweep and the cheapest MH per rented hour, so a buyer on a power bill prefers the 5070 Ti and a renter this card. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", + "watts_class": "v3", + "watts_display": "293.5 (class v3)" }, { "generator": "v2", @@ -866,7 +906,9 @@ "mh_per_w": 0.317, "driver_os": "NVIDIA driver 595.84, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 4 (a rented card, result cb2-rtx-4080-16-gb-7347-result.json; 09:28Z)", - "note": "rented single-card host, 0.15 h; power.draw mean 128.6 W (max 130.5, limit 320 W), SM 2,776 MHz mean, memory 10,801 MHz, 38 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (40.80 MH/s, fingerprint matched); read ceiling 41.0 MH/s, the hash at 0.99 of it. One percent under the 4070 Ti Super at 14 W less: the Ada 16 GB tier is a 41 MH/s tier whichever card" + "note": "rented single-card host, 0.15 h; power.draw mean 128.6 W (max 130.5, limit 320 W), SM 2,776 MHz mean, memory 10,801 MHz, 38 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (40.80 MH/s, fingerprint matched); read ceiling 41.0 MH/s, the hash at 0.99 of it. One percent under the 4070 Ti Super at 14 W less: the Ada 16 GB tier is a 41 MH/s tier whichever card. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", + "watts_class": "v3", + "watts_display": "128.6 (class v3)" }, { "generator": "v2", @@ -882,13 +924,15 @@ "mh_per_w": 0.261, "driver_os": "NVIDIA driver 580.65.06, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 5 (a rented card, result cb2-rtx-3070-8-gb-qh9j-result.json; 09:32Z)", - "note": "rented single-card host, 0.06 h; power.draw mean 142.3 W (max 146.7, limit 225 W), SM 2,002 MHz mean, memory 6,801 MHz, 57 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (37.14 MH/s, fingerprint matched); read ceiling 37.2 MH/s, the hash at 1.00 of it. Against the 3070 Ti: 5 percent less rate for 25 percent less power, the better owner card of the two; 8 GB holds the class v4 set with room", + "note": "rented single-card host, 0.06 h; power.draw mean 142.3 W (max 146.7, limit 225 W), SM 2,002 MHz mean, memory 6,801 MHz, 57 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (37.14 MH/s, fingerprint matched); read ceiling 37.2 MH/s, the hash at 1.00 of it. Against the 3070 Ti: 5 percent less rate for 25 percent less power, the better owner card of the two; 8 GB holds the class v4 set with room. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "142.3 (class v3)" }, { "generator": "v2", @@ -904,13 +948,15 @@ "mh_per_w": 0.22, "driver_os": "NVIDIA driver 595.71.05, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 6 (a rented card, result cb2-rtx-3070-ti-8-gb-8528-result.json; 09:33Z)", - "note": "rented single-card host, 0.07 h; power.draw mean 177.1 W (max 183.6, limit 290 W), SM 1,955 MHz mean, memory 9,251 MHz, 67 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (39.12 MH/s, fingerprint matched); read ceiling 39.1 MH/s, the hash at 1.00 of it. Repeats the 7 October row on another host (38.98 then, 38.98 now; 178 then 177 W), so the class v4 number is host-independent to 0.1 percent", + "note": "rented single-card host, 0.07 h; power.draw mean 177.1 W (max 183.6, limit 290 W), SM 1,955 MHz mean, memory 9,251 MHz, 67 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (39.12 MH/s, fingerprint matched); read ceiling 39.1 MH/s, the hash at 1.00 of it. Repeats the 7 October row on another host (38.98 then, 38.98 now; 178 then 177 W), so the class v4 number is host-independent to 0.1 percent. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "177.1 (class v3)" }, { "generator": "v2", @@ -926,13 +972,15 @@ "mh_per_w": 0.253, "driver_os": "NVIDIA driver 595.71.05, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 7 (a rented card, result cb2-rtx-4060-ti-16-gb-7364-result.json; 09:33Z)", - "note": "rented single-card host, 0.22 h; power.draw mean 79.2 W (max 86.2, limit 165 W), SM 2,836 MHz mean, memory 8,751 MHz, 63 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (20.10 MH/s, fingerprint matched). Repeats the 7 October number (20.10 then, 20.07 now); the one card of the sweep well under its probed read ceiling (0.65 of 30.9 MH/s: the 128-bit bus with the large L2 makes the chase probe optimistic), so 20 MH/s at 79 W is the card's rate; 16 GB holds any class with room", + "note": "rented single-card host, 0.22 h; power.draw mean 79.2 W (max 86.2, limit 165 W), SM 2,836 MHz mean, memory 8,751 MHz, 63 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (20.10 MH/s, fingerprint matched). Repeats the 7 October number (20.10 then, 20.07 now); the one card of the sweep well under its probed read ceiling (0.65 of 30.9 MH/s: the 128-bit bus with the large L2 makes the chase probe optimistic), so 20 MH/s at 79 W is the card's rate; 16 GB holds any class with room. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "79.2 (class v3)" }, { "generator": "v2", @@ -948,13 +996,15 @@ "mh_per_w": 0.468, "driver_os": "NVIDIA driver 580.173.02, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 8 (a rented card, result cb2-rtx-5070-12-gb-8750-result.json; 09:35Z)", - "note": "rented single-card host, 0.09 h; power.draw mean 110.8 W (max 112.0, limit 250 W), SM 2,851 MHz mean, memory 13,801 MHz, 60 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (51.98 MH/s, fingerprint matched); read ceiling 52.3 MH/s, the hash at 0.99 of it. Repeats the 7 October row (52.01 then, 51.82 now) at 8 W more on this host; second to the 5070 Ti in MH per watt among consumer cards (GDDR7 lifts Blackwell here, not SM count); 12 GB holds the class v4 set with 10 GB spare", + "note": "rented single-card host, 0.09 h; power.draw mean 110.8 W (max 112.0, limit 250 W), SM 2,851 MHz mean, memory 13,801 MHz, 60 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (51.98 MH/s, fingerprint matched); read ceiling 52.3 MH/s, the hash at 0.99 of it. Repeats the 7 October row (52.01 then, 51.82 now) at 8 W more on this host; second to the 5070 Ti in MH per watt among consumer cards (GDDR7 lifts Blackwell here, not SM count); 12 GB holds the class v4 set with 10 GB spare. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "110.8 (class v3)" }, { "generator": "v2", @@ -970,13 +1020,15 @@ "mh_per_w": 0.312, "driver_os": "NVIDIA driver 580.178.04, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 9 (a rented card, result cb2-rtx-4070-super-12--8582-result.json; 09:36Z)", - "note": "rented single-card host, 0.13 h; power.draw mean 100.2 W (max 103.9) under a 110 W host cap (stock 220 W), SM 2,764 MHz mean, memory 10,251 MHz, 53 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.29 MH/s, fingerprint matched); read ceiling 34.4 MH/s, the hash at 0.91 of it. Lands on the 4070 Ti's number (31.26 on 7 October) at 7 W less (the same 192-bit 21 Gbps bus); the host's 110 W cap cost no rate, a tune-down reading in itself: an owner can cap a 4070 Super at half its stock power and lose nothing on class v4", + "note": "rented single-card host, 0.13 h; power.draw mean 100.2 W (max 103.9) under a 110 W host cap (stock 220 W), SM 2,764 MHz mean, memory 10,251 MHz, 53 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.29 MH/s, fingerprint matched); read ceiling 34.4 MH/s, the hash at 0.91 of it. Lands on the 4070 Ti's number (31.26 on 7 October) at 7 W less (the same 192-bit 21 Gbps bus); the host's 110 W cap cost no rate, a tune-down reading in itself: an owner can cap a 4070 Super at half its stock power and lose nothing on class v4. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "100.2 (class v3)" }, { "generator": "v2", @@ -992,13 +1044,15 @@ "mh_per_w": 0.316, "driver_os": "NVIDIA driver 595.91.07, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 10 (a rented card, result cb2-rtx-3060-ti-8-gb-8771-result.json; 09:37Z)", - "note": "rented single-card host, 0.12 h; power.draw mean 100.4 W (max 104.7, limit 180 W), SM 1,560 MHz mean, memory 5,001 MHz on this host (below the 7,000 MHz GDDR6 stock), 61 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.75 MH/s, fingerprint matched); read ceiling 31.8 MH/s, the hash at 1.00 of it. Four percent under the 7 October 3060 Ti row (33.1 on another host), consistent with this host's lower memory clock; the best MH per watt of the Ampere cards", + "note": "rented single-card host, 0.12 h; power.draw mean 100.4 W (max 104.7, limit 180 W), SM 1,560 MHz mean, memory 5,001 MHz on this host (below the 7,000 MHz GDDR6 stock), 61 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.75 MH/s, fingerprint matched); read ceiling 31.8 MH/s, the hash at 1.00 of it. Four percent under the 7 October 3060 Ti row (33.1 on another host), consistent with this host's lower memory clock; the best MH per watt of the Ampere cards. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "100.4 (class v3)" }, { "generator": "v2", @@ -1014,13 +1068,15 @@ "mh_per_w": 0.116, "driver_os": "NVIDIA driver 580.159.03, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 11 (a rented card, result cb2-rtx-2080-ti-9077-result.json; 09:40Z)", - "note": "rented single-card host, 0.13 h; power.draw mean 224.6 W (max 234.7, limit 300 W; the driver reports no power average on Turing), SM 1,914 MHz mean, memory 6,802 MHz, 72 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (26.04 MH/s, fingerprint matched); read ceiling 29.9 MH/s, the hash at 0.87 of it. The first Turing row: the 7 October 2080 Ti pod that left no result was a host fault, not the card. The worst MH per watt measured on any card (the Titan RTX, the same silicon, read 28.3 MH/s at 226 W on 7 October): an owner mines at a loss on any real power price; 11 GB holds the set", + "note": "rented single-card host, 0.13 h; power.draw mean 224.6 W (max 234.7, limit 300 W; the driver reports no power average on Turing), SM 1,914 MHz mean, memory 6,802 MHz, 72 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (26.04 MH/s, fingerprint matched); read ceiling 29.9 MH/s, the hash at 0.87 of it. The first Turing row: the 7 October 2080 Ti pod that left no result was a host fault, not the card. The worst MH per watt measured on any card (the Titan RTX, the same silicon, read 28.3 MH/s at 226 W on 7 October): an owner mines at a loss on any real power price; 11 GB holds the set. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "224.6 (class v3)" }, { "generator": "v2", @@ -1036,13 +1092,15 @@ "mh_per_w": 0.115, "driver_os": "NVIDIA driver 575.57.08, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 12 (a rented card, result cb2-rtx-2070-super-8-g-9129-result.json; 09:41Z)", - "note": "rented single-card host, 0.15 h; power.draw mean 166.0 W (max 167.8, limit 215 W; no power average on Turing), SM 1,955 MHz mean, memory 6,802 MHz, 64 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (19.12 MH/s, fingerprint matched); read ceiling 30.1 MH/s, the hash at 0.63 of it: Turing does not overlap the independent reads the hash needs (its indep probe 2.45 G loads/s binds). The Turing per-watt floor: a 4060's rate at twice the power", + "note": "rented single-card host, 0.15 h; power.draw mean 166.0 W (max 167.8, limit 215 W; no power average on Turing), SM 1,955 MHz mean, memory 6,802 MHz, 64 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (19.12 MH/s, fingerprint matched); read ceiling 30.1 MH/s, the hash at 0.63 of it: Turing does not overlap the independent reads the hash needs (its indep probe 2.45 G loads/s binds). The Turing per-watt floor: a 4060's rate at twice the power. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "166 (class v3)" }, { "generator": "v2", @@ -1058,13 +1116,15 @@ "mh_per_w": 0.515, "driver_os": "NVIDIA driver 595.71.05, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 13 (a rented card, result cb2-a100-80-gb-j79g-result.json; 09:43Z)", - "note": "rented single-card host, 0.04 h; power.draw mean 268.6 W (max 274.1, limit 400 W; no power average on this host), SM 1,410 MHz, HBM2e 1,593 MHz, 44 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit reads 136.48 MH/s, 1.3 percent under class v4 (fingerprint matched), the first card where v5 is not level with v4; read ceiling 138.6 MH/s, the hash at 1.00 of it. Repeats the three 7 October readings (138.39, 138.74, 138.76) to 0.3 percent on a fourth host; datacentre-class per watt, a third of a 3080 Ti's MH per rented dollar", + "note": "rented single-card host, 0.04 h; power.draw mean 268.6 W (max 274.1, limit 400 W; no power average on this host), SM 1,410 MHz, HBM2e 1,593 MHz, 44 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit reads 136.48 MH/s, 1.3 percent under class v4 (fingerprint matched), the first card where v5 is not level with v4; read ceiling 138.6 MH/s, the hash at 1.00 of it. Repeats the three 7 October readings (138.39, 138.74, 138.76) to 0.3 percent on a fourth host; datacentre-class per watt, a third of a 3080 Ti's MH per rented dollar. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "268.6 (class v3)" }, { "generator": "v2", @@ -1080,13 +1140,15 @@ "mh_per_w": 0.605, "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 14 (a rented card, result cb2-h100-80-gb-rs3d-result.json; 09:43Z)", - "note": "rented single-card host, 0.03 h; power.draw mean 411.7 W (instant 434.1, the largest gap between the two fields in the sweep, 22 W; max 437.1, limit 700 W), SM 1,980 MHz, HBM3 2,619 MHz, 36 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit reads 252.64 MH/s, 1.5 percent over class v4 (fingerprint matched); read ceiling 252.4 MH/s, the hash at 0.99 of it. Repeats the 7 October rows (248.70, 248.21) on a third host; the best per-watt figure of the sweep on power.draw (0.573 on the instant field) and the worst MH per rented dollar: an owner's card on cheap power, never a renter's", + "note": "rented single-card host, 0.03 h; power.draw mean 411.7 W (instant 434.1, the largest gap between the two fields in the sweep, 22 W; max 437.1, limit 700 W), SM 1,980 MHz, HBM3 2,619 MHz, 36 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit reads 252.64 MH/s, 1.5 percent over class v4 (fingerprint matched); read ceiling 252.4 MH/s, the hash at 0.99 of it. Repeats the 7 October rows (248.70, 248.21) on a third host; the best per-watt figure of the sweep on power.draw (0.573 on the instant field) and the worst MH per rented dollar: an owner's card on cheap power, never a renter's. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "411.7 (class v3)" }, { "generator": "v2", @@ -1102,13 +1164,15 @@ "mh_per_w": 0.279, "driver_os": "NVIDIA driver 595.91.07, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 15 (a rented card, result cb2-rtx-a4000-16-gb-9473-result.json; 09:45Z)", - "note": "rented single-card host, 0.16 h; power.draw mean 112.8 W on a host that caps the card at 114 W (stock 140 W; the SW power-cap bit set the whole run), SM 1,562 MHz mean, memory 6,501 MHz, 81 C; self-test PASS (96 of 96 vector lanes); read ceiling 31.6 MH/s, the hash at 0.99 of it. Class v4 repeats the 7 October row (31.45 then, 31.41 now) under the cap. The class v5 kit's reading on this capped host is held off the page until an uncapped A4000 confirms it", + "note": "rented single-card host, 0.16 h; power.draw mean 112.8 W on a host that caps the card at 114 W (stock 140 W; the SW power-cap bit set the whole run), SM 1,562 MHz mean, memory 6,501 MHz, 81 C; self-test PASS (96 of 96 vector lanes); read ceiling 31.6 MH/s, the hash at 0.99 of it. Class v4 repeats the 7 October row (31.45 then, 31.41 now) under the cap. The class v5 kit's reading on this capped host is held off the page until an uncapped A4000 confirms it. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "112.8 (class v3)" }, { "generator": "v2", @@ -1124,13 +1188,15 @@ "mh_per_w": 0.224, "driver_os": "NVIDIA driver 595.84, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 16 (a rented card, result cb2-rtx-a6000-48-gb-9600-result.json; 09:46Z)", - "note": "rented single-card host, 0.15 h; power.draw mean 239.1 W (max 251.0, limit 300 W), SM 1,913 MHz mean, memory 7,601 MHz, 65 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (53.82 MH/s, fingerprint matched), so on an uncapped sm_86 card v5 is level with v4 and the A4000's v5 reading narrows to its power cap; read ceiling 53.8 MH/s, the hash at 1.00 of it. Repeats the 7 October row (53.66 then, 53.64 now) at 43 W less on this host; a 3090-class rate at a 3090-class draw", + "note": "rented single-card host, 0.15 h; power.draw mean 239.1 W (max 251.0, limit 300 W), SM 1,913 MHz mean, memory 7,601 MHz, 65 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (53.82 MH/s, fingerprint matched), so on an uncapped sm_86 card v5 is level with v4 and the A4000's v5 reading narrows to its power cap; read ceiling 53.8 MH/s, the hash at 1.00 of it. Repeats the 7 October row (53.66 then, 53.64 now) at 43 W less on this host; a 3090-class rate at a 3090-class draw. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "239.1 (class v3)" }, { "generator": "v2", @@ -1146,13 +1212,15 @@ "mh_per_w": 0.131, "driver_os": "NVIDIA driver 580.173.02, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, rows 17 and 22 (two rented hosts, results cb2-gtx-1080-ti-11-gb-0377-result.json and cb2-gtx-1080-ti-11-gb-1452-result.json; 09:49Z and 10:02Z)", - "note": "the mean of the first host's two clean runs (16.26 MH/s at 124.3 W, the card uncapped at a 280 W limit); a second host read 16.29 MH/s at 129.4 W under a 130 W host cap with five of five self-tests passing, so the first host's one-in-three cache self-test failure was that host's VRAM and not Pascal, and the flag is cleared. Power.draw on the first host max 125.5 W (no power average on Pascal), SM 1,936 MHz mean, memory 5,005 MHz, 59 C; the class v5 kit held the rate on the uncapped host (16.27 MH/s) and read 13.27 MH/s on the 130 W-capped one (fingerprint matched on both); read ceiling 16.3 MH/s, the hash at 1.00 of it. The first Pascal row: a 2017 card that runs the class at its own ceiling", + "note": "the mean of the first host's two clean runs (16.26 MH/s at 124.3 W, the card uncapped at a 280 W limit); a second host read 16.29 MH/s at 129.4 W under a 130 W host cap with five of five self-tests passing, so the first host's one-in-three cache self-test failure was that host's VRAM and not Pascal, and the flag is cleared. Power.draw on the first host max 125.5 W (no power average on Pascal), SM 1,936 MHz mean, memory 5,005 MHz, 59 C; the class v5 kit held the rate on the uncapped host (16.27 MH/s) and read 13.27 MH/s on the 130 W-capped one (fingerprint matched on both); read ceiling 16.3 MH/s, the hash at 1.00 of it. The first Pascal row: a 2017 card that runs the class at its own ceiling. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card; Pascal has no clock lock; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "124.3 (class v3)" }, { "generator": "v2", @@ -1168,13 +1236,14 @@ "mh_per_w": null, "driver_os": "NVIDIA driver 595.91.07, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 18 (a rented card, result cb2-rtx-4060-8-gb-0672-result.json; 09:53Z)", - "note": "watts not read: this host exposes no power sensor (every power field absent on all 244 samples, as the 7 October 4060 host); a third host is being rented for the watts, and the card's draw is about 60 to 70 W by its 115 W limit and the 4060 Ti's 79 W, an estimate until then. Rented single-card host, 0.12 h; SM 2,732 MHz mean, memory 8,251 MHz, 57 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (19.11 MH/s, fingerprint matched); read ceiling 30.2 MH/s, the hash at 0.63 of it (the 128-bit bus: the independent-read line binds, as on the 2070 Super). Repeats the 7 October row (19.09 then, 19.10 now)", + "note": "watts not read: this host exposes no power sensor (every power field absent on all 244 samples, as the 7 October 4060 host); a third host is being rented for the watts, and the card's draw is about 60 to 70 W by its 115 W limit and the 4060 Ti's 79 W, an estimate until then. Rented single-card host, 0.12 h; SM 2,732 MHz mean, memory 8,251 MHz, 57 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (19.11 MH/s, fingerprint matched); read ceiling 30.2 MH/s, the hash at 0.63 of it (the 128-bit bus: the independent-read line binds, as on the 2070 Super). Repeats the 7 October row (19.09 then, 19.10 now). Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3" }, { "generator": "v2", @@ -1190,13 +1259,15 @@ "mh_per_w": 0.117, "driver_os": "NVIDIA driver 580.159.03, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 19 (a rented card, result cb2-rtx-2060-0673-result.json; 09:55Z)", - "note": "rented single-card host, 0.15 h; power.draw mean 120.6 W (max 124.6, limit 160 W; no power average on Turing), SM 1,948 MHz mean, memory 6,801 MHz, 69 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (14.14 MH/s, fingerprint matched); read ceiling 16.1 MH/s, the hash at 0.88 of it. The floor of the list and the smallest card on it: the class v4 set fits with 4.6 GB spare; the same Turing per-watt floor as the 2070 Super and the 2080 Ti, so every Turing card earns the same per watt and only the absolute rate differs", + "note": "rented single-card host, 0.15 h; power.draw mean 120.6 W (max 124.6, limit 160 W; no power average on Turing), SM 1,948 MHz mean, memory 6,801 MHz, 69 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (14.14 MH/s, fingerprint matched); read ceiling 16.1 MH/s, the hash at 0.88 of it. The floor of the list and the smallest card on it: the class v4 set fits with 4.6 GB spare; the same Turing per-watt floor as the 2070 Super and the 2080 Ti, so every Turing card earns the same per watt and only the absolute rate differs. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "120.6 (class v3)" }, { "generator": "v2", @@ -1212,13 +1283,15 @@ "mh_per_w": 0.327, "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 20 (a rented card, result cb2-rtx-4070-ti-12-gb-0695-result.json; 09:57Z)", - "note": "rented single-card host, 0.18 h; power.draw mean 95.4 W (max 96.7, limit 285 W), SM 2,825 MHz mean, memory 10,251 MHz, 49 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.28 MH/s, fingerprint matched); read ceiling 31.7 MH/s, the hash at 0.99 of it. Repeats the 7 October rate to 0.07 percent on another provider at 12 W less; the 192-bit Ada pair (4070 Ti, 4070 Super) land on one number, 31.2 MH/s under 100 W", + "note": "rented single-card host, 0.18 h; power.draw mean 95.4 W (max 96.7, limit 285 W), SM 2,825 MHz mean, memory 10,251 MHz, 49 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (31.28 MH/s, fingerprint matched); read ceiling 31.7 MH/s, the hash at 0.99 of it. Repeats the 7 October rate to 0.07 percent on another provider at 12 W less; the 192-bit Ada pair (4070 Ti, 4070 Super) land on one number, 31.2 MH/s under 100 W. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "95.4 (class v3)" }, { "generator": "v2", @@ -1234,13 +1307,15 @@ "mh_per_w": 0.256, "driver_os": "NVIDIA driver 570.133.20, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 21 (a rented card, result cb2-l40s-48-gb-0890-result.json; 09:57Z)", - "note": "rented single-card host, 0.17 h; power.draw mean 220.7 W (max 227.2, limit 350 W), SM 2,520 MHz, memory 9,001 MHz, 45 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (56.51 MH/s, fingerprint matched); read ceiling 57.8 MH/s, the hash at 0.98 of it. Repeats the 7 October row (56.36 then, 56.41 now) at 20 W less on this host; a 384-bit GDDR6 card that lands where the 3090 Ti and the A6000 do at a datacentre price. Driver 570.133, the oldest of the sweep, ran the CUDA 12.8 worker clean: 570 is a safe floor", + "note": "rented single-card host, 0.17 h; power.draw mean 220.7 W (max 227.2, limit 350 W), SM 2,520 MHz, memory 9,001 MHz, 45 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (56.51 MH/s, fingerprint matched); read ceiling 57.8 MH/s, the hash at 0.98 of it. Repeats the 7 October row (56.36 then, 56.41 now) at 20 W less on this host; a 384-bit GDDR6 card that lands where the 3090 Ti and the A6000 do at a datacentre price. Driver 570.133, the oldest of the sweep, ran the CUDA 12.8 worker clean: 570 is a safe floor. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "220.7 (class v3)" }, { "generator": "v2", @@ -1256,13 +1331,15 @@ "mh_per_w": 0.316, "driver_os": "NVIDIA driver 580.126.09, Ubuntu 24.04 (CUDA 12.8.1 image)", "source": "model sweep 2026-10-08, row 23 (a rented card, result cb2-rtx-4080-super-16-g-2453-result.json; 10:03Z)", - "note": "rented single-card host, 0.12 h; power.draw mean 134.9 W (max 138.4, limit 320 W), SM 2,740 MHz mean, memory 11,254 MHz, 51 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (42.68 MH/s, fingerprint matched); read ceiling 43.1 MH/s, the hash at 0.99 of it. The 23 Gbps GDDR6X gives it 4.6 percent over the 4080 for 6 W more at the same MH per watt: the Ada 256-bit tier is 41 to 43 MH/s whichever card", + "note": "rented single-card host, 0.12 h; power.draw mean 134.9 W (max 138.4, limit 320 W), SM 2,740 MHz mean, memory 11,254 MHz, 51 C, no throttle reason; self-test PASS (96 of 96 vector lanes); the class v5 kit holds the rate (42.68 MH/s, fingerprint matched); read ceiling 43.1 MH/s, the hash at 0.99 of it. The 23 Gbps GDDR6X gives it 4.6 percent over the 4080 for 6 W more at the same MH per watt: the Ada 256-bit tier is 41 to 43 MH/s whichever card. Watts are class v3 (read 8 October 2026, 11:4x UK): the fleet's bench pack was the class v3 program (program_class v3, generator 3, load_class mx8-erad810f22d), so this row's watts and MH per watt are class v3 figures; the rate holds under class v4 (memory-bound: a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4). The class v4 watts rerun on the same models lands with its date, the v3 watts kept beside as the shadow premium per card", "hive": { "core_mhz": null, "mem_mhz": null, "pl_w": null, "label": "stock (rented card, no clock control on the host; no measured tune point)" - } + }, + "watts_class": "v3", + "watts_display": "134.9 (class v3)" } ] } diff --git a/site/miners.html b/site/miners.html index 02d373359..d7f8df81e 100644 --- a/site/miners.html +++ b/site/miners.html @@ -262,14 +262,14 @@ table.bench2 tr.detail .d{display:block;margin:0 0 4px}table.bench2 tr.detail b{

The table

-

Best desktop card: NVIDIA RTX 5090, 136.1 MH/s (measured, 2026-10-06). Best stock MH per wall watt: NVIDIA RTX 5070 Ti, 0.577 at 135.9 W, untuned (measured by the fleet, 2026-10-08). Best tuned: NVIDIA RTX 5090, 0.563 at 226.8 W (measured, 2026-10-06).

-

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found class v4's rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held class v4's rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it; the table's rates are class v4.

+

Best desktop card: NVIDIA RTX 5090, 136.1 MH/s (measured, 2026-10-06). Best stock MH per wall watt: NVIDIA RTX 4090, 0.285 at 183.1 W, untuned (measured by the fleet, 2026-10-07). Best tuned: NVIDIA RTX 5090, 0.563 at 226.8 W (measured, 2026-10-06).

+

The rented model sweep of 8 October 2026 (23 rows, 22 models, every row untuned) found the rate cap-proof: a card on a host power cap gives the same MH/s as an uncapped one. Class v5 held the rate on every uncapped card and read 17 to 18 percent under it on the two power-capped hosts (an A4000 at 114 W, a GTX 1080 Ti at 130 W), so a class v5 rate needs the card's power state beside it. The sweep's watts, and the 7 October fleet rows' watts, are class v3 figures: the fleet's bench pack ran the class v3 program, and class v4 draws more for the same rate (a 5070 Ti measured 78.7 MH/s at 141 W on class v3 and 78.8 at 224 W on class v4, 8 October 2026). Those rows are ranked by rate alone; the per-watt claims above come from rows measured under class v4; the class v4 watts rerun lands on each row with its date, the v3 watts kept beside it as the shadow premium per card.

One row per card on the current class: the class v4 program (the latency-shadow block over the class v3 hash), or a class v3 row re-measured with its class v4 cost on 6 October 2026 or later. MH per wall watt uses board or wall power; a row whose watts are the chip's (Apple silicon: GPU plus DRAM from IOReport) says so and is not ranked on that column. Integrated GPUs are not listed.

Cards you can buy

-
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.50.577135.9 WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.40.539145.6 WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-07620.248249.5 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-07590.221267.3 WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.90.201293.5 WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-07520.506102.8 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.80.468110.8 WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4080 Super (16 GB)class v4 not measured · 2026-10-0842.60.316134.9 WStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.20.289142.6 WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.70.317128.6 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-07390.219178.3 WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-08390.22177.1 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.10.261142.3 WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.10.256129.5 WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.70.316100.4 WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.30.41575.4 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.30.291107.3 WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.20.312100.2 WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0831.20.32795.4 WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.90.241111.6 WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-08260.116224.6 WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.10.25977.5 WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.10.25379.2 WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.10.115166 WStock
NVIDIA RTX 4060 (8 GB)class v4 not measured · 2026-10-0819.1not measurednot readStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
NVIDIA GTX 1080 Ti (11 GB)class v4 not measured · 2026-10-0816.30.131124.3 WStock
NVIDIA RTX 2060 (6 GB)class v4 not measured · 2026-10-0814.10.117120.6 WStock
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock
+
NVIDIA RTX 5090 (32 GB)class v4 133.8 MH/s, 305 W, 0.439 · 2026-10-06136.1 stock (127.7 tuned)0.563226.8 tuned (350 stock) WTuned
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0878.5○ 0.577135.9 (class v3) WStock
NVIDIA RTX 5070 Ti (16 GB)class v4 not measured · 2026-10-0778.4○ 0.539145.6 (class v3) WStock
NVIDIA RTX 5080 (16 GB)class v4 +83 W unlocked, +41 W at the best points · 2026-10-0871.4 stock (71.2 tuned)0.486146.6 tuned (253 stock) WTuned
NVIDIA RTX 3090 Ti (24 GB)class v4 not measured · 2026-10-0762○ 0.248249.5 (class v3) WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0759○ 0.221267.3 (class v3) WStock
NVIDIA RTX 3080 Ti (12 GB)class v4 not measured · 2026-10-0858.9○ 0.201293.5 (class v3) WStock
NVIDIA RTX 4090 (24 GB)class v4 not measured · 2026-10-0752.30.285183.1 WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0752○ 0.506102.8 (class v3) WStock
NVIDIA RTX 5070 (12 GB)class v4 not measured · 2026-10-0851.8○ 0.468110.8 (class v3) WStock
NVIDIA RTX 3090 (24 GB)class v4 not measured · 2026-10-0750not measurednot readStock
NVIDIA RTX 3080 (10 GB)class v4 not measured · 2026-10-0743.7not measurednot readStock
NVIDIA RTX 4080 Super (16 GB)class v4 not measured · 2026-10-0842.6○ 0.316134.9 (class v3) WStock
NVIDIA RTX 4070 Ti Super (16 GB)class v4 not measured · 2026-10-0841.2○ 0.289142.6 (class v3) WStock
NVIDIA RTX 4080 (16 GB)class v4 not measured · 2026-10-0840.7○ 0.317128.6 (class v3) WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-0739○ 0.219178.3 (class v3) WStock
NVIDIA RTX 3070 Ti (8 GB)class v4 not measured · 2026-10-0839○ 0.22177.1 (class v3) WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0837.1○ 0.261142.3 (class v3) WStock
NVIDIA RTX 3070 (8 GB)class v4 not measured · 2026-10-0733.7not measurednot readStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0733.1○ 0.256129.5 (class v3) WStock
NVIDIA RTX 3060 Ti (8 GB)class v4 not measured · 2026-10-0831.7○ 0.316100.4 (class v3) WStock
NVIDIA RTX 5060 (8 GB)class v4 not measured · 2026-10-0731.3○ 0.41575.4 (class v3) WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0731.3○ 0.291107.3 (class v3) WStock
NVIDIA RTX 4070 Super (12 GB)class v4 not measured · 2026-10-0831.2○ 0.312100.2 (class v3) WStock
NVIDIA RTX 4070 Ti (12 GB)class v4 not measured · 2026-10-0831.2○ 0.32795.4 (class v3) WStock
NVIDIA RTX 4070 (12 GB)class v4 +30 W · 2026-10-06310.38979.5 WTuned
NVIDIA RTX 5060 Ti (16 GB)class v4 0.1 percent rate · 2026-10-0730.90.269114.8 WStock
Apple M5 Max (40 GPU cores, Metal)class v4 +16 W · 2026-10-0627○ 1.2921 W chip, not wallNo lever
NVIDIA RTX 3060 (12 GB)class v4 not measured · 2026-10-0726.9○ 0.241111.6 (class v3) WStock
NVIDIA RTX 2080 Ti (11 GB)class v4 not measured · 2026-10-0826○ 0.116224.6 (class v3) WStock
NVIDIA RTX 4060 Ti (8 GB)class v4 not measured · 2026-10-0720.1○ 0.25977.5 (class v3) WStock
NVIDIA RTX 4060 Ti (16 GB)class v4 not measured · 2026-10-0820.1○ 0.25379.2 (class v3) WStock
NVIDIA RTX 2070 Super (8 GB)class v4 not measured · 2026-10-0819.1○ 0.115166 (class v3) WStock
NVIDIA RTX 4060 (8 GB)class v4 not measured · 2026-10-0819.1not measurednot readStock
AMD Radeon RX 9070 XT (16 GB)class v4 +2 percent rate · 2026-10-0818.9 (18.8 to 19.2 on the G1 ladder)0.093202 stock WNo lever
NVIDIA GTX 1080 Ti (11 GB)class v4 not measured · 2026-10-0816.3○ 0.131124.3 (class v3) WStock
NVIDIA RTX 2060 (6 GB)class v4 not measured · 2026-10-0814.1○ 0.117120.6 (class v3) WStock
Intel Arc B580 (12 GB)class v4 0.1 percent rate · 2026-10-0711not measurednot readStock

38 cards. Details opens a card's Hive flight-sheet values, its full class v4 cost, miner, source and note.

Datacentre (13 cards, rented for the measurement; about three times the rented dollars per hash of a desktop card) -
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.40.487855.6 WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-073130.723432.9 WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.70.645385.6 WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.30.605411.7 WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-071550.517299.6 WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.40.52266.3 WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.10.515268.6 WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.50.452288.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0856.40.256220.7 WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.40.234240.7 WStock
NVIDIA RTX A6000 (48 GB)class v4 not measured · 2026-10-0853.60.224239.1 WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.40.279112.8 WStock
+
NVIDIA B200 (180 GB)class v4 not measured · 2026-10-07416.4○ 0.487855.6 (class v3) WStock
NVIDIA H200 SXM (141 GB)class v4 not measured · 2026-10-07313○ 0.723432.9 (class v3) WStock
NVIDIA H100 SXM (80 GB)class v4 not measured · 2026-10-07248.7○ 0.645385.6 (class v3) WStock
NVIDIA H100 SXM (80 GB HBM3)class v4 not measured · 2026-10-08248.3○ 0.605411.7 (class v3) WStock
NVIDIA A100 PCIe (80 GB)class v4 not measured · 2026-10-07155○ 0.517299.6 (class v3) WStock
NVIDIA A100 SXM (80 GB)class v4 not measured · 2026-10-07138.4○ 0.52266.3 (class v3) WStock
NVIDIA A100 SXM4 (80 GB)class v4 not measured · 2026-10-08138.1○ 0.515268.6 (class v3) WStock
NVIDIA RTX PRO 6000 Blackwell (96 GB)class v4 not measured · 2026-10-07130.5○ 0.452288.7 (class v3) WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0856.4○ 0.256220.7 (class v3) WStock
NVIDIA L40S (48 GB)class v4 not measured · 2026-10-0756.4○ 0.234240.7 (class v3) WStock
NVIDIA RTX A6000 (48 GB)class v4 not measured · 2026-10-0853.6○ 0.224239.1 (class v3) WStock
NVIDIA RTX A5000 (24 GB)class v4 not measured · 2026-10-0747.6not measurednot readStock
NVIDIA RTX A4000 (16 GB)class v4 not measured · 2026-10-0831.4○ 0.279112.8 (class v3) WStock
Earlier classes (6 rows: the genesis program, the hourly program, class v3 before the shadow; not comparable with the table above)

The bench numbers of 3 and 4 October 2026. A higher MH/s here is a different hash, not a faster card.

From caa0071efc2015e5db2a5d9bad7df8d57fa828d6 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:23:29 +0000 Subject: [PATCH 15/19] Site: the Devnet 3 swap page at /swap (MetaMask and any injected wallet by chain id 4463 and the public RPC; quotes, swaps, the faucets, add liquidity; every swap's block with its proof status from /api/live; Devnet 3, test tokens, no value on every view), Swap in the Mine panel and the sheet of every page Co-Authored-By: Claude Fable 5.1 --- site/404.html | 2 + site/address.html | 2 + site/app.html | 2 + site/bench.html | 2 + site/block.html | 2 + site/claims.html | 2 + site/dev-fee.html | 2 + site/download.html | 2 + site/evidence.html | 2 + site/explorer.html | 2 + site/faucet.html | 2 + site/income.html | 2 + site/index.html | 2 + site/journey.html | 2 + site/ledger.html | 6 +- site/litepaper.html | 2 + site/live.html | 2 + site/metamask.html | 2 + site/miner.html | 2 + site/miners.html | 2 + site/partials/nav.html | 2 + site/provenance.html | 2 + site/randomx.html | 2 + site/scenes.html | 2 + site/swap.html | 531 ++++++++++++++++++++++++++++++++++++ site/wallet.html | 2 + tools/ci/site-nav-check.mjs | 2 +- 27 files changed, 584 insertions(+), 3 deletions(-) create mode 100644 site/swap.html diff --git a/site/404.html b/site/404.html index 4020c5c81..f422fa0cf 100644 --- a/site/404.html +++ b/site/404.html @@ -78,6 +78,7 @@ main{flex:1} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -112,6 +113,7 @@ main{flex:1} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/address.html b/site/address.html index 1b21eab43..c3139fc3d 100644 --- a/site/address.html +++ b/site/address.html @@ -126,6 +126,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -160,6 +161,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/app.html b/site/app.html index 143257d2b..70fccdca9 100644 --- a/site/app.html +++ b/site/app.html @@ -100,6 +100,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -134,6 +135,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/bench.html b/site/bench.html index 3559f6c31..3935a6a13 100644 --- a/site/bench.html +++ b/site/bench.html @@ -105,6 +105,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -139,6 +140,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/block.html b/site/block.html index 1b1d4e62e..e5418208c 100644 --- a/site/block.html +++ b/site/block.html @@ -126,6 +126,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -160,6 +161,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/claims.html b/site/claims.html index f86f4de02..3daf9494c 100644 --- a/site/claims.html +++ b/site/claims.html @@ -104,6 +104,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -138,6 +139,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/dev-fee.html b/site/dev-fee.html index dc749c60b..bee097174 100644 --- a/site/dev-fee.html +++ b/site/dev-fee.html @@ -104,6 +104,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -138,6 +139,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/download.html b/site/download.html index 7c6872ad9..46ea6d8e2 100644 --- a/site/download.html +++ b/site/download.html @@ -94,6 +94,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -128,6 +129,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/evidence.html b/site/evidence.html index 391c2479c..380b3bcbd 100644 --- a/site/evidence.html +++ b/site/evidence.html @@ -115,6 +115,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -149,6 +150,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/explorer.html b/site/explorer.html index 4dcc62e34..1a68453bf 100644 --- a/site/explorer.html +++ b/site/explorer.html @@ -125,6 +125,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -159,6 +160,7 @@ main{padding-bottom:100px}.card{background:var(--row);border:1px solid var(--lin GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/faucet.html b/site/faucet.html index d571e217e..d5ab1511b 100644 --- a/site/faucet.html +++ b/site/faucet.html @@ -107,6 +107,7 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:14px;overflo GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -141,6 +142,7 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:14px;overflo GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/income.html b/site/income.html index 5f73c7dbf..b1feee37f 100644 --- a/site/income.html +++ b/site/income.html @@ -64,6 +64,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -98,6 +99,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/index.html b/site/index.html index 716624fbb..50c5489e7 100644 --- a/site/index.html +++ b/site/index.html @@ -96,6 +96,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -130,6 +131,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/journey.html b/site/journey.html index a613273f6..51ddee055 100644 --- a/site/journey.html +++ b/site/journey.html @@ -104,6 +104,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -138,6 +139,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/ledger.html b/site/ledger.html index f896139f2..ff9fdf9aa 100644 --- a/site/ledger.html +++ b/site/ledger.html @@ -89,7 +89,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c
@@ -108,6 +108,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -142,6 +143,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
@@ -154,7 +156,7 @@ blockquote{margin:10px 0;padding:10px 14px;border-left:3px solid var(--line-2);c
Learn
LitepaperThe design, as published. Income per tierIGN a day per card at three network sizes, and the electricity. - LedgerEvery criticism, answered or conceded. + LedgerEvery criticism, answered or conceded. What Igneum does not claimThe limits, stated first. Igneum vs RandomXWhat was kept and what was rebuilt for GPUs. Built on the shouldersEvery borrowed part, credited. diff --git a/site/litepaper.html b/site/litepaper.html index defe6ddf3..48f6e4c5f 100644 --- a/site/litepaper.html +++ b/site/litepaper.html @@ -161,6 +161,7 @@ body.all .pager{display:none} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -195,6 +196,7 @@ body.all .pager{display:none} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/live.html b/site/live.html index 12c52735c..88e815262 100644 --- a/site/live.html +++ b/site/live.html @@ -245,6 +245,7 @@ details.tablebar summary{display:flex;align-items:center} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -279,6 +280,7 @@ details.tablebar summary{display:flex;align-items:center} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/metamask.html b/site/metamask.html index 8f2ef399a..16865c270 100644 --- a/site/metamask.html +++ b/site/metamask.html @@ -106,6 +106,7 @@ ol{margin:0 0 14px;padding-left:22px}li{margin-bottom:6px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -140,6 +141,7 @@ ol{margin:0 0 14px;padding-left:22px}li{margin-bottom:6px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/miner.html b/site/miner.html index 1a34df3a7..5cad52743 100644 --- a/site/miner.html +++ b/site/miner.html @@ -101,6 +101,7 @@ pre b{color:var(--molten-text);font-weight:500} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -135,6 +136,7 @@ pre b{color:var(--molten-text);font-weight:500} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/miners.html b/site/miners.html index 8abcfc7f6..856f3c9f6 100644 --- a/site/miners.html +++ b/site/miners.html @@ -105,6 +105,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -139,6 +140,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/partials/nav.html b/site/partials/nav.html index 58fd4456a..b5aa57afc 100644 --- a/site/partials/nav.html +++ b/site/partials/nav.html @@ -28,6 +28,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -62,6 +63,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/provenance.html b/site/provenance.html index c303deb48..0459eb93e 100644 --- a/site/provenance.html +++ b/site/provenance.html @@ -105,6 +105,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -139,6 +140,7 @@ table{min-width:560px} GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/randomx.html b/site/randomx.html index ebc61f0a5..9d9444f56 100644 --- a/site/randomx.html +++ b/site/randomx.html @@ -104,6 +104,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -138,6 +139,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/scenes.html b/site/scenes.html index fd8804a07..b944d070e 100644 --- a/site/scenes.html +++ b/site/scenes.html @@ -82,6 +82,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -116,6 +117,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/site/swap.html b/site/swap.html new file mode 100644 index 000000000..196dc10fb --- /dev/null +++ b/site/swap.html @@ -0,0 +1,531 @@ + + + + + +Igneum swap on Devnet 3. Test tokens, no value + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
+
+
Igneum/Swap
+
Devnet 3 swap
+

Swap test tokens on Devnet 3.

+

A constant-product exchange on the Igneum zkEVM: wrapped IGN and two faucet tokens in three pools, 0.3 percent to the pool. Every swap is a Devnet 3 block, and the block's proof status is read live below.

+

Devnet 3, test tokens, no value. The chain may reset and these pools go with it. Nothing here is an offer, a price or a market.

+
+
+
+

Swap

+
+ + No wallet connected. MetaMask or the Igneum Wallet, pointed at chain id 4463. +
+
+ +
+ + +
+ + +
+ + +
+

Enter an amount for a quote. Slippage limit 1 percent, deadline 10 minutes.

+ +
+
+

Your swaps this session

+

Each row is a Devnet 3 block. Proof status comes from the public feed (/api/live): planned, proving, verified or paid per shard, and whether a finality checkpoint has locked the block.

+ + + +
SwapTransactionBlockProof status
No swaps yet this session.
+
+ +
+
+ + + + + + + + diff --git a/site/wallet.html b/site/wallet.html index dc32d079b..a4e7906e0 100644 --- a/site/wallet.html +++ b/site/wallet.html @@ -114,6 +114,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
@@ -148,6 +149,7 @@ GPU bench tableMeasured rates, card by card. The dev feeThe optional one, in full view, off with one flag. Testnet faucetTest coin for an address, once a day. + SwapTest tokens on Devnet 3; every swap is a proven block. Add to MetaMaskIgneum as a network in your wallet.
Network
diff --git a/tools/ci/site-nav-check.mjs b/tools/ci/site-nav-check.mjs index 05acee2e6..dfac14e78 100644 --- a/tools/ci/site-nav-check.mjs +++ b/tools/ci/site-nav-check.mjs @@ -13,7 +13,7 @@ import { tmpdir } from 'node:os'; import { fileURLToPath } from 'node:url'; const here = dirname(fileURLToPath(import.meta.url)); const GROUPS = { - mine: ['/miner', '/download', '/app', '/wallet', '/miners', '/dev-fee', '/faucet', '/metamask'], + mine: ['/miner', '/download', '/app', '/wallet', '/miners', '/dev-fee', '/faucet', '/swap', '/metamask'], network: ['/live', '/explorer', '/journey', '/bench', '/evidence'], learn: ['/litepaper', '/income', '/ledger', '/claims', '/randomx', '/provenance'], }; From 6821b6862e14f85e9ebd4dbed798a42d00cc9821 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:24:10 +0000 Subject: [PATCH 16/19] Counter ASIC 3.0 status: main's word on the two reds (the detach in 0.3.25 if green by 12:15), the day-key scan to 2^28 PASS, the sweep reruns read as none, the attack-pass branch merged and pushed, the family gate's first coverage (16 eras; the index-bit bias firing at address bit R on 7 of 16 drawn eras, a class v6 design row: load_index's form, not a floor; the bucket bound in sigma), the ota test literal fixed (f1a8102a, the detach rides 0.3.25) Co-Authored-By: Claude Fable 5.1 --- docs/plans/counter-asic-3-status.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md index 602a7891c..1bfc3cdb9 100644 --- a/docs/plans/counter-asic-3-status.md +++ b/docs/plans/counter-asic-3-status.md @@ -465,7 +465,7 @@ Reading: the class v4 premium is 145.3 W at the unlocked clock (not the 80 W of | 1,200 | 133.80 | 305.1 | 0.439 | 129.54 | 215.7 | 0.601 | 1,192 | | 1,100 | 122.43 | 287.3 | 0.426 | 118.70 | 209.4 | 0.567 | 1,087 | -The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. THE 0.3.25 PAIRS ON build-1 (the build-server lane, /srv/artefacts/0325-7bd2940f/): the seed pair at 10:34:44 BST (igneumd c7fc542b, igneum-miner 4494ecc4, glibc 2.34), the Windows pair at 10:36:13 (igneumd.exe 5d1dea23, igneum-miner.exe eee7bdfa), the hive package igneum-hive-0.3.25-7bd2940f.tar.gz at 10:37:49 (sha d977797f..., the three kit zips, smoked in ubuntu:20.04); the kit re-cut from 92f004f1 (sha 676240f6, 424,540 B) staged in both folders, the PC 1 host from it bc8d4f79 (in host.sha256 at the shipper's 24680e1d), the 0.3.25 Windows payload from 24680e1d cutting. Every pair of the cut exists by 10:38; the pin's gate set and the crossing are the only waits. FOUR NEW LANES ON THE FOUNDER'S ORDER (11:00 BST, "build all this today to close this gap"), mirrored by the coordinator as the shipper's clocks are: the explorer (a5ef1d5801084005b; explorer.igneum.network by 16:00), the canonical DEX and the Sepolia certificate verifier (a74a8267813d6ea34; the AMM by 14:00, the swap UI by 17:00, the verifier by 20:00), the builder pages, faucet and grants (adb29da59baf27898; /build and /grants by 15:00, the faucet by 16:00), three reference apps that only work on a proven chain (a2060899d2a27d31c; /light by 16:00, /receipt by 18:00, the Sepolia oracle demo by 21:00); the build-server lane stands up rpc.devnet.igneum.network by 12:00; they do not touch the 0.3.25 cut, the crossing or the fleet, sharing the boxes' lease pools (class measure) and the master-only deploy; a lane silent past 25 minutes gets the word from the coordinator and then main. THE FAST-TIME GATE ON THE 0.3.25 PAIR: SUMMARY PASS (cross-0325-39f127a1) at 09:54:40Z (10:54 BST) on the pair 39f127a1 (the node code and object byte for byte e0644958's; igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 09:42:25Z to 09:54:40Z, every check green (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8 at rung 1 at 9,985 bps, the stale node refused, the restart step resynced in 8 s, four sinks equal, 0 PoW rejections); the ceiling's two new fields absent from the 60x file so the ceiling stayed at never there (the node lane's note); to the shipper the same minute; the pin line names e0644958 and its gates. THE FOUNDER'S WORD AT 11:0x BST ("can we add in any more layers? class rotating? things that would render an ASIC useless as soon as it dropped"): the class v6 design opens today as a rotating family, the research lane and the hash lane under the coordinator, the design doc docs/design/class-v6-rotating-family.md by 18:00 BST with the chip-model rows beside each layer (what it does to k and capex for a fixed-function chip and to the per-joule edge for a GPU-like chip; what it costs every GPU tier, Apple included): (1) per-era draws of the class parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement), drawn from chain state like the program; (2) the state-derived dataset's size tracking chain-state growth with a floor, so fixed-memory silicon ages out; (3) scheduled family epochs by height (every 180 days by default) with no release; (4) the (c''') acceptance floor and the F8-form uniformity test generalised to each era's parameter draw, redraw on failure, so layers 1 and 3 need no per-era cryptanalysis. Per layer: the gate it needs (the family analysed as a family: the attack board's shape over the testnet period), the known-failed test, an honest line on what a fully general chip still gets. No consensus code this week; the document, the numbers and the gate plan. Per tier for the founder tonight: what each layer does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE ARC RE-READ IN ITS CHAIN (the hash lane, 10:57 UK): no clear came from the shipper, so the default ran at 10:50: the rotate-fold kit's fetch (sha 65b47211) published to PC 2 at 10:51:41, the run (run-ca3-pc2-v5-intel-bench-20261008, the v5 lane's script c0d398a1) in the publish chain behind another lane's publish-jobs.sh sign --deploy from the build-server worktree (the publisher serialises); the fingerprint line by 11:15 if the publisher frees inside ten minutes, else the blocking process named by 11:10. Queued on PC 1 behind the same publisher: run-ca3-pc1-v5lock-5090-20261008 (class v5 against v4 at unlocked, 1,300 and 1,200 MHz, the v5 kit's CUDA packs), its rows by 12:30; the AMD grid after it from about 11:25. The class-key work on ember-tiers-25 started; the v6 cost rows by 16:00 taken. THE TIERS' CLASS-FLIP STATE, THE UI HALF (the UI lane, 10:57 BST): tiers-class-25 at d949e274 on the mirror, off release-0.3.25's tip 24680e1d (the shipper having merged tiers-25 d3d0704a into release-0.3.25 at 111dae69), the crate gate GREEN 303 + 35 + 8 on build-1, the full pre-push GREEN 60, the UI tests 74 green known-failed first (the v4 tiers stayed on the buttons after the flip on d3d0704a); after the flip the table reads "re-measuring for class v5" on every button with the start minute or "queued (within ten minutes of the crossing)", the v4 watts never current, the strip's sentence naming the crossing; the knee note under the table when knee_loss_pct is over 2 percent; the captures tiers-flip-dark.png and -light.png; the fields tiers_class, program_class, tiers_remeasure_at, knee_loss_pct (the shape sent to the hash lane at 10:4x; the engine sha by 12:30); the default: the display rides 0.3.25 inert if the engine half is late and lights up on 0.3.26. THE FOUNDER'S WORD AT 11:1x BST: class v6 is DECLARED with the four layers as its spine (per-era parameter draws, the dataset tracking chain state, scheduled family epochs by height, the acceptance floor generalised to parameters), and deep past-and-future research opens now under the coordinator with serious resources ("see if anything can be optimised, added or invented"; reading public research is in-house, nothing paid or asked of anyone outside): four research lanes today, (A) history (every ASIC-resistant proof-of-work and how it fell or held: Ethash and the E3 and Linzhi chips, ProgPoW's review, RandomX and its chip analyses, Cuckoo, Equihash and the Z9, Argon2 and Scrypt and the Litecoin chips, KawPow, Autolykos, Octopus, kHeavyHash's chips; the exact mechanism each chip used and what the design missed, each mapped to Igneum's layers with "does v6 close it" as a sentence and a number), (B) the hardware future five years out (PIM and processing-near-memory, HBM3e and HBM4, LPDDR6, 3D DRAM, CXL memory pools, wafer-scale, chiplets, FPGA with HBM; for each the chip-model k band against a state-sized dataset and dependent random reads, and the one layer that would blunt it), (C) invention (layers beyond the four, each a paragraph, a known-failed test and a chip-model row: data-dependent program graphs, latency-bound dependent reads tied to the shard proof, randomised memory topology per era, VRAM-size ratchets, proof-carrying hashes sampled by the pool, time-locked parameter commitments, and what the lane invents; rejecting what costs GPUs more than chips), (D) the family gate (how a parameter family is cryptanalysed as a family: sampling bounds, coverage, the F8-form and (c''') tests over the parameter space, the attack board's shape over the testnet period, so layers 1, 3 and 4 can be automatic with a proof of what was tested). Resources: all four boxes under lease class measure, PC 1 by job for card rows, the rented fleet for one-shot measurements inside the ceiling. Deliverables: a first synthesis in docs/design/class-v6-rotating-family.md by 20:00 BST (the four layers priced, every finding from A to D with its number, a ranked list of what v6 adds beyond the four, the honest line on what a fully general chip still gets), the full report by 09:00 tomorrow, one line to main per lane as each lands; per tier at 20:00: what v6 does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE 0.3.25 APP TIP AND PIN CANDIDATE (the shipper, 10:58 BST): the app tip 9b93e649 (push gate GREEN; the crate unchanged from e0d4425f; the node-source pin to e0644958 and host.sha256 bc8d4f79); the pin candidate the node lane's ceiling cut e0644958 (digest 1b37cb9d, every gate green 10:53, the fast-time SUMMARY PASS 10:54, the floor at DAA 82,800 about 16:53 BST, a publish up to 14:53 without a second cut); the tiers' class-key halves: the UI lane's tiers-class-25 d949e274 green and inert alone, merged with the hash lane's engine sha the moment it lands (12:30), gated as a pair on the release tip, riding only if green by the 13:00 pin; the 0.3.24 Windows take 2 failed at a new place (Inno stopped the app and copied nothing); the update-return lane owns the fix on release-0.3.25 by 12:30, the default the 0.3.25 Windows entry waiting for a clean take 3 while Mac and HiveOS move at the minute. THE FOUR CLASS V6 RESEARCH LANES SPAWNED (the coordinator, 11:0x BST, each with its worktree, its box resources under lease class measure, its clocks and the rules): lane A history (a603a938582c43ab5; the first cut docs/analysis/class-v6/history.md by 15:00), lane B the hardware future (a4f73e2a6f2d1b757; hardware-future.md by 16:00), lane C invention (a5dfe95ee8c47cd0f; invention.md by 17:00), lane D the family gate (a07a99a3788566af2; family-gate.md by 17:00); each feeds the research lane's synthesis docs/design/class-v6-rotating-family.md by 20:00 (its outline by 13:00; the hash lane's per-tier rows by 16:00); the full reports by 09:00 tomorrow; the coordinator's lane mirror carries their clocks. A HELD PUSH AND ITS CAUSE (11:00 BST): the hash lane's push of ca3-v4-amend was refused at 10:58 by the gh-account hook reading the founder's gh (his personal login active again; nothing switched by any lane); the cause is the branch's own hook, which predates the per-process fix (34b0884d): the hook runs the branch's tools/ci, so every branch older than 09:58 must merge the mirror's master before its next push, under which the check reads Igneum's own gh directory and skips under the suspension marker; the rule to every lane. Live: the Arc re-read on PC 2 (published 10:59:47) and the v5lock job on PC 1 (published 10:53, about 12 minutes). THE CLASS V6 OUTLINE ON THE MIRROR (the research lane, docs/design/class-v6-rotating-family.md on counter-asic-4, the commit after fb61ed4b, pushed 10:5x UTC, two hours ahead of 13:00): section 0 the founder's table (per layer, what it does to a fixed-function chip and to a GPU-like chip on its release day, and the 5090, 5070 Ti and M5 Max columns, measured where the night's rows exist, the 5070 Ti scaled until the hash lane's row); the honest frame on top: the four layers render a FIXED-FUNCTION chip useless on the first era its wired value leaves (one tape-out lives one era) and move nothing for the stored-dataset chip with a programmable core except the core's size and the N5 project it forces; that chip keeps 3.6x at zero premium and 2.1x at k = 1 on a 5090 at its knee. The layer table (sections 1 and 2) names the bands each draw takes and the measured rows that set them: the mixer in {4, 8, 16} (x16 open), the op-mix weights within B = 4 with shuffle and mulhi capped (shfl 55.8 pJ per op), the read width in {1, 4} words (w64 excluded by the 5 October rows), the block shape 64 to 256 (never 1,024), N left to the ladder's signal (an unconditional draw retires the Apple tier at 200,000). Open numbers asked of the hash lane with defaults at 16:00: the 5070 Ti row (the rented 5070 scaled), the x16 mixer's verifier and build (the chip model's estimate), two re-weighted shadow packs for the op-mix band (the microbench arithmetic). Layers 2 to 4 and the gate plan are skeletons with their sources named, filling by 18:00 with the four research lanes' cuts, the synthesis by 20:00. THE FOUNDER'S WORD AT 11:2x BST ("all builders are idle, load them up"): build-1 to build-4 filled now and kept above 80 percent all day under the lease pool, class measure behind the release gates, in this order of value: (1) the class v6 family gate's sampling runs for lane D (the F8-form census and the (c''') floor over the parameter bands: the mixer {4, 8, 16}, the op-mix weights within B = 4 with shuffle and mulhi capped, the read width {1, 4}, the block 64 to 256; thousands of drawn eras, the uniformity and bucket tests on each, so the family document carries measured coverage tonight); (2) the attack families at scale on the 0.3.25 pin candidate's igneum-pow (F8 to 256 seeds, F9 and F1 to 10^6 on the frozen 1c420786, the day-key scan to 2^28) as the record's strengthening lines; (3) the invention lane's candidate layers measured as packs as fast as it writes them; (4) the full suite matrix of the 0.3.25 pin on every box as the pre-pin check; (5) the Windows and hive cross builds and the sweep's reruns; the lease tool's pre-emption giving release-class work the cores when the pin's gates need them; one line to main at 12:00 with the load on each box and what runs there, then hourly only if a box drops idle. THE DRIFT CLASS SETTLED (the node lane, from the fleet's fresh-walk node, a shared-devnet node synced from an empty datadir to 191,441 chain blocks at 10:03:45Z): hub-1 and five standing boxes number the fresh chain exactly; seven boxes carry numbering inherited from an exec snapshot taken on a chain that later re-walked (+2: p1-4090, p1-a5000, pool-1, build-1's node1; +3: p2-3090-2; +4: p2-3090-4; +5: p1-5090 and p2-3090-3), and a restart on the kept datadir does not re-walk (p1-5090 at 09:22Z stayed +5); the cost: a prover on drifted numbering signs statements the hub vetoes, so the seven earn nothing from proving until they re-walk, the drift refusal stopping the waste. The node lane's word to the fleet: p1-5090 first, both snapshot files moved aside so the executor re-walks from the DAG, the re-walk timed and read against the fresh node, then the other six in series, hub-1 untouched, build-1's node1 after the 12:40Z move; if the re-walk reads over two hours the six wait for the node-side fix on the next node line (the loader re-numbering a resumed range against the DAG before serving). THE 0.3.25 PIN CANDIDATE CONFIRMED (the node lane): e0644958 on both mirrors (keygen, the re-announce, the ceiling switch at 82,800 in the Devnet 3 object, digest 1b37cb9d, the 0.3.24 pin refused both ways), every gate green at 09:53:01Z, the fast-time SUMMARY PASS at 09:54:40Z on the same object; the publish ceiling DAA 75,600 (14:53 BST); waiting only on the 68,400 crossing reading clean (about 12:54; the node lane's line the moment the DAA passes it); the shipper names the pin at 13:00; the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15. LANE C'S FIRST PACK (the invention lane, 11:0x BST by the Mac's clock; its own line read "12:1x", a clock to correct): build-1 takes the igneum-pow build from counter-asic-4 at 5984ffab, then the per-load shadow in its sound form (mx8+shl6912x1: 16 sub-blocks of 432, one pass, the form 20.2a named and never drew) as the first candidate: the acceptance census over 64 seeds and 16 drawn eras against the 16x27 form and the class v4 shape, the pack export, the F8 read at 2^24 and the verifier bench on a leased core, the first read by 13:30; build-2 next for the second candidate (warp-uniform data-dependent block selection); the candidates with no pack form (the VDF commitment, the VRAM ratchet, the pool-sampled witness, the state-tied reads) stay modelled and the 17:00 cut says so; the worktree igneum-wt-v6-invention on class-v6-invention. THE WINDOWS INSTALLER CLASS AND THE 0.3.25 TIP (the shipper, 11:08 BST): the app tip 139c147a (9b93e649 plus install-detach-25 52a34111, packaging/windows and tools/ci only, the crate unchanged; push gate GREEN); the 0.3.24 take 2 class: an installer started under the app's job runner is a child of the engine, and the engine's kill_tree on quit ended it between PrepareToInstall and the copy; the fix re-launches the installer as a one-shot scheduled task outside the job's tree; the 0.3.24 Windows entry skipped; the rule-14 take on PC 2 is the 0.3.25 installer over the running 0.3.21 app, queued ahead of the Arc re-read; the pin candidate e0644958, the DMG 501ba293 staged, the 13:00 pin and the 13:40 provisional minute standing. THE ATTACK FAMILIES AT SCALE (the attack-pass lane, cores held at 11:08 BST, every run under lease pool class measure): box 2 (88 cores): F8 seeds p66 to p257 (192 new, 256 with the gate's p2 to p65) at 2^24 on class v5 at the freeze 1c420786 (the gated binary 0f5c98dc, pairing e5a4ac5978462156; the leaves re-run on 8f481459 if the kit pairing flips), the window-model control, by-site, as three thirds of 64 seeds; box 4 (80 cores held, 16 asked): F9 to 10^6 on 1c420786 (seeds 100,000 to 999,999 in six chunks of 150,000 at 8 threads, four running), F1 to 10^6 class v5 programs on 1c420786 (one census at 40 threads with the progress line and flushed partials, re-drawing the record's first 10^5 on the way as a reproduction check), the F4 day-key scan to 2^28 on 8ca66afa's redraw rule at 8 threads; nothing on build-1 or build-3 (lane D's); the projections: F4 about 2 to 3 hours, F8's 192 seeds about 9 hours, F9's 900,000 and F1's 10^6 about 30 hours each, so the 17:00 default is partials for those two with the lane (d) rows carrying counts so far; any pre-emption by release-class work reported. THE RPC AND THE 0.3.25 PAIRS ON THE PIN CANDIDATE (the build-server lane, 11:0x BST): rpc.devnet.igneum.network up since 11:08 BST (the first of the founder's builder clocks, 52 minutes ahead); the 0.3.25 pairs on e0644958 running on build-1 since 11:06 (the app tip 139c147a), the Windows and hive crosses on build-2, build-3 and build-4 as reproducibility rows at class release by about 12:40; the sweep reruns' list not held by the lane, the default at 12:30: last night's sweep logs on build-1 read for rows that ended without a result line and those rerun at class measure. THE SWEEP RERUNS' LIST (the fleet lane to the build-server lane, 11:1x BST): the fleet ran nothing under the build boxes' lease pool last night (every fleet bench a rented GPU one-shot), so the rows the lease kills cut short are the hash and class lanes' and the build-server lane's default read on build-1 is the right one; the fleet's own rows without a result (A10, A40, A100 40 GB, H100 NVL, H100 PCIe, MI250, RTX 3050, RX 7800 XT, 7900 XT, 7900 XTX, 6900 XT) are provider gaps needing a GPU host, rerun the moment a provider lists one. THE CLASS-FLIP TIERS, BOTH HALVES (the UI lane, 11:13 BST by the Mac's clock, 1 h 47 min inside the 13:00 pin): tiers-class-25 at 081b3ba7 (the display d949e274 plus the hash lane's ember-tiers-25 0a838072, on release-0.3.25's 9b93e649; the field names matched exactly): the crate gate GREEN 305 + 35 + 8 on build-1, the pre-push GREEN 60, the UI tests 78 green known-failed first, the push gate GREEN; with the shipper. A RED ON THE RELEASE TIP, for the shipper and the update-return lane: release-0.3.25's 139c147a is red on one crate test (ota::return_tests::no_relaunch_while_an_installer_runs_and_a_relaunch_when_it_clears, 304 of 305): install-detach's 0c588b09 reshaped the installer's clear step into a multi-line block while the test asserts the one-line literal at app/igneum-app/src/ota.rs:1348; 9b93e649 passes; the fix is the test's literal on the install-detach line; the UI lane built on 9b93e649 so its tip is green alone. A RED FROM THE PIN MATRIX (the CI steward, 11:13 UK): the core suite fails on the pair (the node e0644958 with the app tree 9b93e649): config::params::tests::fast_time_60x_file_is_the_devnet_at_60x panics "override-60x.json lacks the field base_unit_decimals"; the field was added by 0e4ec18a on ca3-v4-node yesterday at 21:45 UK and reached neither master, release-0.3.25 nor the app tip while the node line's test demands it; so every box reads red on core, and the fix is one line on release-0.3.25 (the cherry-pick of 0e4ec18a, or "base_unit_decimals": 8 in infra/fast-time/override-60x.json); sent to the shipper; green so far pow and app on build-1 and build-3; the two new boxes' toolchains read the same as build-1 (Ubuntu 24.04.5, glibc 2.39, the pinned rustc, sccache and lease, no nvcc). The founder's fourth load item paid in its first ten minutes: a red no single-box gate had read. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | +The knee by main's rule (more than 1 percent lost against unlocked): 1,300 MHz on both classes (the rate within 1.5 percent of unlocked down to it; v3 falls 5.1 percent at 1,200, v4 10.5 percent at 1,100); the best MH per watt one step past it: v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W, 168.6 W recovered for 2.2 percent of rate), v3 at 1,300 (134.62, 223.3 W, 0.603, 106.6 W for 1.4 percent). The v4 premium 143.8 W unlocked, 81.8 W at the best points; the v4 rate 0.25 percent over v3 unlocked and 0.61 percent under at the best points; the residual at the floor is the shadow's ALU work, not the clock. Per tier: a 5090 owner on class v4 locked at 1,200 to 1,300 MHz draws 305 to 313 W instead of 474 for 1.5 to 2.2 percent less rate, MH per watt up 49 to 52 percent; the Ember knob (0.3.24, the hash lane on the engine side, the UI lane's drawing) carries these as its reference rows. A FAULT FOUND AND FIXED: the steps 1,000 down to 300 and the closing reset got no answer from the Power Helper and the card sat at the 1,100 lock for about five minutes after the job (118 to 122 MH/s live); the installed app's own Ember tune on the 5080 wrote the same cmd.txt with higher sequence numbers while the script wrote lower ones, and the helper skips any sequence at or under the last run; the restore job run-ca3-pc1-clocks-restore-20261007 (exit 0 at 20:45:58Z) put the 5090 back at 2,865 MHz; the fix 45f9497f on the mirror (the sequence base from helper.log and cmd.txt, re-based after a timeout, an unanswered lock stops the grid, the task restarted before every reset); the rule for the knob: it takes its sequences from the engine's counter and no script shares the file with a running tune. The driver's floor below 1,100 is unmeasured. THE PC 1 QUEUE after the shipper's 0.3.23 host job (main, 21:5x UK): the 5080 full grid with the fix; the research lane's SM-sparse kernel job (the hash on a fraction of the SMs, several chains per thread, the rest clock-gated; the research lane hands the kernel to the hash lane); the third 5090 pass from 1,100 down to the driver's floor at the tail; then the 9070 XT G1 and ladder, the v5 AMD bench, item 6 on AMD, the 5080 and 9070 XT tunes, the L2 cache-policy hot table; each exit line to the shipper and the coordinator; the honest site sentence (the premium at the knee and the floor it buys, labelled measured, the Ember knob named as how a user gets there) once the 5080 reads. THE DERIVATION FINDING FIXED (the hash lane, 15008aca and 0f45c8be on the mirror): one byte recipe (generator::IdRecipe) builds the id and the printed text; program.json states the generator 4 suffix and the rung form; spec 1.4.6 corrected (class v5 = generator 5, no suffix); tests/derivation.rs re-derives all 18 pinned packs from their own text (the plain text gives 8aa9f185d63f269e for the devnet v4 pack, the known-failed case); 38 packs' program.json re-exported with ids, kernels and fingerprints byte-identical; the full igneum-pow suite green on box 2. CLASS V5 FROZEN: class-v5 1c420786 on both box mirrors at 21:53 UK (the (c''') floor with its number; section 14 with seven of seven live hot sets refused at 0.9821 to 0.9919, seed 170 at 0.9880 the seventh, and the three mild residuals at 0.9992 to 0.9997 named at about 1.0004x; the pinned pack unchanged; the flip-stale harness PASS on the matched binaries at 21:03 UK; the AP-F4-1 first form and the AP-F1-1 shadow rule, the latter's measured trigger 11 permille maximum over 6,000 first draws against the 30 bound, 0 redraws; the igneum-pow suite green on box 2: 73 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7; the gate GREEN at 58 checks). The kits lane: the 0.3.24 kit is packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 (state.igsd1 included), fingerprint 82b19cbde8557ea5 on Metal, Apple OpenCL and a CUDA 4090; AMD on PC 1's queue, Intel deferred; the shipper has the line. The attack-pass lane runs F8 at 2^24, F9 at 10^5 and F1 on 1c420786 under class v5. The v5 lane's next commit on the freeze: AP-F4-1 in the agreed form (cost at most 205 against the median 226, w32 without the position-32 digit, k >= 1 and all-ROT-equal rejected, the known-failed day 29,337 = 2050-04-28) and the verified last resort (part (a) repaired by re-sourcing stale loads, then the whole rule over a 256-candidate scan, known-failed first on adv-accept-3's adv3/steer/2); both move the stream only on days and seeds the chain never reaches. THE FOURTH EXCEPTION ON THE RESTART STEP (the fast-time lane's held-miner run on the third pair 63524e28, 20:4xZ): the IBD catch-up's body sync anchored on the node's own sink and moved only on a whole chunk's successful join, so with the honest headers arriving as one chunk failing on its v5 tail it fetched nothing and the executor never reached the seed block; the relay hold-off and the mining hold from the earlier fixes read green on that run. FIXED by the node lane at f0c56f50 (the refused chunk split by consensus's own record, the anchor moved to the highest validated header, the honest v4 prefix through the seed block, only the unvalidated headers deferred; kaspa-p2p-flows 38). PAIR 4 = v5-object-0323 c8f9b383, re-archived from the frozen 1c420786 (generator.rs and accept.rs moved since ab6f980b, memhard.rs not), building on build-1 at gate priority since 20:54:32Z with the line's gates beside it; the restart step's PASS must come from pair 4; the object commit lands the minute it does, with dn3-g1's DAA at the cut plus 7,200 rounded up to the 3,600 boundary and its UTC clock named; the testnet lane told to pair its re-cut with 1c420786. The crossing clock is not yet a reading: about 22:15Z (23:15 BST) at the earliest if every line reads green on its first pass. The site audit lane: no other "12 days" form served; its row-17 edit keeps main's outside-check clause and adds the 5090 efficiency numbers. THE CHIP TEXTS, THE X9 WORDING RETIRED (main's order from the counter-asic-4 research file d7721ebe, 22:0x UK): the withdrawn Antminer X9's claimed ratio ("a third of a CPU's energy per RandomX hash") is against a CPU core (about 100 pJ per instruction, Horowitz and Dally, claimed), not a GPU lane (6.5 to 10.4 pJ measured), so a chip three times better than a CPU is worse than a GPU lane per op and the X9 is not a pessimistic chip core against us. The served texts (the home line, the litepaper's lead, chip table, ladder sentence and chip bullet, /claims through it, the miner line, evidence row 17) now give the floor and the premium as measured numbers at the 5090's knee: the chip at 2.1x per joule with a core as good as a GPU lane (k = 1) and 3.4x with one three times better (k about 0.33), no core below about 1.8 pJ per op in the model's range, the shadow's premium 81.8 W at the best points (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W against class v3 at 1,300 MHz 134.62 at 223.3 W, 7 October 2026), Ember Tune's core-clock knob named as how a user gets there; the ledger text check's pins X35 and X36 moved with the wording; no "3.9x" remains on any served page. One number stated against main's wording: main's line read "2.9x with one three times better", which in the research file is the figure for the RE-WEIGHTED op mix (row 3, held by the coordinator until the SM-sparse read); today's mix at a core three times better reads 3.4x in the same file, so the served text carries 3.4x and the 2.9x waits for the re-weight to ship. THE RESEARCH FILE's TWO ORDERS: (1) the texts as above; (2) one zero-code measurement at the PC 1 tail after the third 5090 pass: the 5 October hot-table packs (packs-ca2-hot, 32 and 64 MiB) with the worker's `--variant ldcs` (dataset loads streaming, evict-first; the hot loads plain and L2-resident) against base on the 5090, the rate ratio g and the watts (the 5 October rows without the hint g 0.84 to 0.87); the one class where a chip's cost per op (a 64 MiB SRAM read, 0.2 to 0.5 nJ approximate) may exceed the GPU's (an L2 hit, 0.1 to 0.3 nJ); Metal has no such hint. The shadow stays at rung 0; the op-mix re-weight waits for the SM-sparse read (the research lane's worker variants sp170/85/43/21/11-w32, one block of 32 warps per SM, run through the hash lane's efficiency script in its ca4 mode at 4f3a064e; the no-prompt and sequence rules hold by the same code). THE 0.3.24 PAIRING RULED (the shipper, 22:1x UK): the v5 object commit pairs with the frozen class-v5 1c420786 as it stands (the gates and the attack-pass lines run on it); the post-freeze fix 8ca66afa is 0.3.25's pairing. 0.3.25's FIRST ROW: class-v5 8ca66afa (both mirrors, 22:10 UK, on 1c420786): (1) AP-F4-1 in the agreed form (decc7c17): the day's draw rejected when cost A = 64 + sum(w32(MUL_i) - 1) is at most 205 against the median 226, w32 over bit positions 0 to 31 (the position-32 carry digit dropped), any MUL with w32 at most 3 rejected (k >= 1), the eight ROT all equal rejected, a rejected block redrawn whole from the continuing stream; known-failed first on chain day 29,337 (2050-04-28): the sub-version 3 block of that day read cost 203, rejected at 205 and redrawn under class v5. (2) Class v5's verified last resort: the rewrite, then repair_stale_loads (a stale load re-sourced to the lowest register written since its last load, to a fixpoint), then the whole rule over a 256-candidate scan from the cap; the unchecked fallback past the scan under 1e-300; known-failed first on adv-accept-3's adv3/steer/2 (the sub-version 3 rewrite fails part (a) at instruction 47 reading r3; the repair restores (a) moving only load sources; class v5's last resort passes at attempt 256, id 9b29c9481f6941d4; steer 11, 33, 56, 58 and 77 pass too); sub-version 3's path untouched. The stream moves only on days and seeds the chain never reaches: the pinned v5 packs byte-identical, the fingerprint 82b19cbde8557ea5 and the epoch-0 id e5a4ac5978462156 unchanged; the igneum-pow suite green on box 2 (74 unit, packs 20, derive 7, mixer 4, recheck 2, scratch 7), the gate GREEN at 58 checks. The harness's class-walk case (v4 floor 0, v3 never) read FAIL on the unfixed fork 546fe4b5 (the known-failed shape, 22:08 UK) and runs on pair 4. THE IN-HOUSE PASS, THE EIGHTH HOT SET (adv-accept, 22:06 BST, the wider sweep over 88,051 accepted programs): seed 122960 (id 4be7393ab6c84802, the lowest 256-unit ratio at 0.9885) reads live at 2^24 X_f +0.111 percent, X/f 1.11, 1.54x the window model, with the heaviest single item measured tonight (0x81ad88 at 475,616 reads, 0.022 percent of all reads, 16x 100767's hottest) from an all-ones source at instruction 4 (writer shfl at 3); site 12's saturated-source share 0.353 percent, a third of (c')'s limit; the other four lowest 256-unit proxies clean live, so the 256-unit proxy is noise at its own extreme and the 2^20 ratio is the selector; the tally 8 hot sets in 30 tail seeds against 0 in 20 random; the price unchanged (0.34 percent of reads on 1 MB, 1.002x); its minimum-site ratio at 2^20 against the 0.995 floor OWED (ordered first), deciding whether the freeze record reads eight of eight refused or names the first hot set the floor misses. THE 5080 AT STOCK (run-ca3-pc1-v4-eff-5080-20261007-b, exit 0 at 21:03:02Z, the card alone, 60 s, both fingerprints matched): class v4 71.43 MH/s at 255.1 W (0.280 MH/W, sm 2,958, mem 14,801 MHz); class v3 71.30 at 170.7 W (0.418); the v4 premium 84.4 W (49 percent over v3's draw), the rate 0.18 percent over v3; against the fleet's rented 5080 (71.16 MH/s at 143.4 W on class v4, driver 580) the rate agrees to 0.4 percent and the watts do not (255 against 143), a question to the fleet lane (its sampler, a cap on the rented card, the memory clock) before either row enters the public table; the lock grid did not run in -b (a PowerShell function defined below its first call left the script without the helper path; nothing set, nothing to restore), republished as -c at 21:07:10Z with the full grid (unlocked to 300 MHz, about 58 minutes). The site audit lane's row 17 and litepaper paragraph carry the 1,400 MHz rows labelled measured, with the best-points clause asked beside the 88 W at 1,400. THE 0.3.24 OBJECT COMMIT AND PIN: v5-object-0323 774f16c9 (21:26:35Z, both mirrors; the fork 432ea3d6 + f0c56f50 + 9ad1d9c6 + 294e3670 + the pool lane's 95ae3e50), paired with the frozen igneum-pow 1c420786: program_class_v5_activation_daa 28,800 (the Devnet 3 seed node at virtual DAA 16,208 at 21:22:24Z; the publish minute 22:30Z = DAA 20,264; plus 7,200 = 27,464; the next 3,600 boundary 28,800, epoch 8), byte 6 counted exactly, the window 86,400; the crossing on Devnet 3 by height about 00:52Z on 8 October (01:52 BST) at 1.0 DAA/s; the constant holds while the publish DAA stays at or under 21,600 (22:52:16Z), past which the node lane re-reads dn3-g1 and re-cuts to 32,400; chain id 4463 below the floor and 4464 from it; the three heights stay, the pool split never. Its gates: core 155 of 155, miner 28 of 28, pow 19 of 19, p2p-flows 38 of 38, exec 46 of 46, consensus 126 of 126 on the gate-priority rerun at 21:44:24Z (the earlier one red at 205 ms on the latency bound under a box load of 127, the known load class); the canary set on build-1 (21:29:38Z to 21:31:18Z): the digest moves to 4a284b1d on igneum-devnet-3 as the v5 arm requires, "this node stamps object version 6 into its headers (block version 1538)", the override file refused, two empty nodes handshake on 4a284b1d, the shared-devnet node refused on network mismatch, a 0.3.23 node refused on the digest both ways; every Devnet 3 node restarts inside one minute at the fleet's named clock on pre-placed binaries. release-0.3.24-node OPEN at 774f16c9 on both mirrors (21:45:19Z, the shipper's word), artefact /srv/artefacts/0324-774f16c9/node-lane (igneumd ed36f246...); the testnet staging 47b9b229 on the pin all green (consensus 134, core 175, exec 47, miner 28, p2p-flows 38, pow 19, digest b2e856ed). THE FAST-TIME GATE CLOSED: SUMMARY PASS (cross-c8f9b383-2) at 21:36:35Z on the matched pair c8f9b383 (igneumd f1b5b32c..., igneum-pow 1c420786), every check green, none skipped: class v4 sub-version 3 from genesis at rung 0; rung 1 by signal from epoch 6 at 21:29:39Z; class v5 by signal at byte 6 counted exactly from epoch 8 (DAA 480) at rung 1 at 21:31:33Z on 4 of 4 nodes, 9,985 bps, before the floor; the second rung at epoch 12 the rule's earliest allowed; 11 of 11 program ids equal to the CPU verifier's; 0 PoW rejections on the honest nodes; the stale node 69 of 69 refused; the restart step: n2 stopped at DAA 455, restarted on its own datadir at DAA 500 at 21:31:56Z, no lock fault, no IBD refusal, "class v5 catch-up done: 19 deferred headers validated after 6 s", nothing of its own accepted during the catch-up and 75 after, at n0's sink 12.1 s after its start; four sinks equal at 660; the digest-compat PASS from 20:08:30Z stands; records on v5-fasttime 4419e8d3. The three earlier pairs (959b57c9, 63524e28, 432ea3d6) each failed the restart step on a node defect fixed in the next (the IBD refusal, the catch-up's anchor at the node's own sink, the node mining while its catch-up waited). THE FLOOR READS EIGHT OF EIGHT (adv-accept, 22:41 BST): seed 122960 (the deepest live hot set) reads minimum site 12 at 0.9824 at the acceptance's 2^20 sample (live 0.9822), REFUSED by (c''') at 0.995 (its site 12 puts 1.31 percent of its reads on word indices read 8 or more times, the largest repeated-index share measured; 100767's site 6: 0.17); every live hot set by X_f at or above f found in the tail of 88,051 accepted programs is refused (minimum sites 0.9821 to 0.9919) against 0 hot sets in 20 random programs; the floor misses the three mild concentrations at 0.9992 to 0.9997 (Devnet 3's first program among them), about 1.0004x; the v5 design's section 14 and the ledger's AP-F8-1 carry the line. THE 0.3.24 CUT waits on the attack-pass verdicts on 1c420786 alone (F8's two halves on build-2 since 21:17:41Z, about 22:20 to 22:35Z; F9 at 10^5 and F1 on build-1); the lease pool now pre-empts adv holders at any size for a v5 or release waiter after 120 s (lease ce30e357). PC 1 EXCEPTION: the Power Helper task dies within seconds of each start since 21:08:34Z (six starts, zero commands, the task Running while no helper process exists; the last good command the 20:45:52Z rgc, its idle exit clean at 21:05:52Z); the suspect the shipper's 0.3.23 host job at 20:51Z replacing the install folder's exe under the registered task, the second a panic in the helper's start path; a read-only diagnostic plus a 20 s unelevated probe placed; the locked grids (the 5080 full grid, the third 5090 pass), the SM-sparse job and the tunes wait on the helper; the lock-free jobs run (the 9070 XT G1 and ladder from 21:27:41Z, then the family run and the v5 AMD bench); nothing raises a prompt to get round it. THE 5080 AT STOCK (two runs agreeing, -b and -c): class v4 71.42 MH/s at 254.5 W (0.281 MH/W, sm 2,960, mem 14,801), class v3 71.30 at 170.8 W (0.418), the premium 84 W; against the fleet's rented 5080 (71.16 MH/s at 145.4 W busy mean, cap 350 W not binding, 1 Hz power.draw instantaneous on Linux driver 580, bench batches with host gaps) the rate agrees to 0.4 percent and the watts do not (110 W apart, the sampler field on Blackwell under two drivers or the load shape); the public table carries the method per row and takes neither as the card's figure until both power fields are sampled on both sides (the fleet's re-measure, PC 1's next NVIDIA pass). THE CA4 SECOND PASS (bca23f96, sections 15 to 19): the tensor-tile k column (2.1x at k = 1, 1.6x at k = 1.5, the k 0.3 column removed for a tensor shadow; a design candidate needing a SIMD byte-dot verifier) and the capex column (the f = 1 GDDR7 chip USD 2.8 per MH/s, at most 4.3 with the hot table, the shadow core and an interposer; capex-dominated 7x; the break-even cap moving only through the project cost) carried into chip-model-v3 as section 5.11. THE PUBLIC TEXTS (main's two orders, 22:3x UK): the served sentence "the one outside check is staged and waits on its escrow and the publish word" read as an escrowed prize to a reader and is replaced everywhere it is served (evidence row 17, the litepaper and /claims through it, the public text file) by "no outside review has run yet", the in-house pass sentence kept; the forbidden-strings gate gains the phrase class ("outside check", "waits on its escrow", "staged and waits", "the publish word"; the bare words stay allowed, since the proving pool's escrow and a staged build are ordinary). THE /miners DESIGN PASS is on the mirror's ca3-coord at e88edae4 with the full gate GREEN (the overlap check clean at 390 to 1600 px after two fixes: the phone grid gives every cell its own area; the desktop row is six columns with the class v4 cost and the date as the muted second line under the card name, the card layout below 1,100 px, the wrapper scrolling as a safety); the 1440 and 390 dark captures go to main for the word on the look; nothing deploys from the branch before it. The in-house pass: four lanes complete (adv-cache, adv-accept-2, adv-cache-3, adv-mixer; adv-mixer's Q1 BOUND on the commutation probe at 0 in 1,454,080,000 over 1,024 days, its SAT row a solver-reach bound at the one-hour cap); adv-mixer-2 one row from complete; adv-accept, adv-accept-3, adv-cache-2 and adv-mixer-3 sweeping to 00:00 BST. F8 ON CLASS V5: PASS (the attack-pass lane, 22:03Z; the frozen igneum-pow class-v5 1c420786, binary sha256 0f5c98dc41a1b3aa...; the pairing bit for bit on 66 validation lines, the library drawing Devnet 3's epoch-0 program as e5a4ac5978462156; 64 seeds p2 to p65 at 2^24 nonces each, chain path, the v5 dataset from v5-dn3-epoch0's state.igsd1 on day 20,733, window-model control, build-2 under lease pool class v5 as two halves of 32, ended 21:58:43Z and 22:03:21Z): 61 of 64 under 1.2x of the window model (0.9919x to 1.144x, p75 1.0024x); 3 over, all inside the named four-seed residue and none new: p10 1.5047x (hottest item 0x4018f5 at 346 reads of 2^31, no predicted source), p8 1.3787x (419 reads), p4 1.2166x (363 reads); p34 reads 0.9997x under the (c''') floor; every strong seed of sub-versions 1 and 2 at 0.9997x to 1.0001x (p23 1.0000, p19 0.9997, p15 0.9998, p18 1.0001, p56 1.0000); seed for seed the ratios equal sub-version 3's within 0.001 except where the floor moved a draw: the state leaves change the words, not the read addresses. F9 (10^5 exhaustion) and F1 (10^5 redundancy) on 1c420786 and F4's 2^24 on 8ca66afa hold or wait in build-1's pool as strengthening lines. THE 0.3.24 NODE PIN MOVED on the shipper's word to 47b9b229 (the object 774f16c9 plus the testnet re-cut 34892a36) after the Devnet 3 canary set read clean on its own binary (21:59:04Z to 22:00:43Z: digest 4a284b1d, byte 6, the override refused, shutdown 725 ms, the handshake, the shared-devnet dialler and a 2720d8d2 node refused); release-0.3.24-node at 47b9b229 on both mirrors (22:01:05Z), igneumd 6bc18ac2..., pairing 1c420786; the build-server lane builds the pairs and the hive from it; the Devnet 3 digest 4a284b1d, the testnet b2e856ed; the floor 28,800 and its slip rule, the dn3-g1 re-read armed for 22:30Z. THE AMD HALF OF G1 PAID (run-ca3-pc1-v4-sub3-amd-g1-20261007, exit 0 at 21:46:14Z, the RX 9070 XT alone): 14 of 14 fingerprints equal to the Mac's Metal and Apple OpenCL and to the 5090's (the control, the seven sub-version 3 packs, the five ladder packs), self-test PASS on all; the ladder rows flat within 2.3 percent from 930 to 330,700 ops per hash (18.8 to 19.2 MH/s; the installed worker's control cross-check 18.96), the card latency-bound on the whole ladder; the watts row owed (the ADLX sampler read 0 samples in the per-pack windows). THE HELPER FAULT READ: not the shipper's; the task's exe is the install folder's 0.3.20 (mtime 12:24:42Z, sha256 0443ae17..., untouched by the host jobs); the helper's code path runs (an unelevated probe answered a dev line in 4 s); the scheduler refuses the ELEVATED instance from a non-interactive start (Last Result 0x800710E0, the task's logon mode interactive only); at 21:41:32Z the 0.3.20 engine's own tune took its legacy "task not registered" branch (the old sweep.rs helper.ps1 written, cmd.txt truncated), the prompt path, so whether a prompt stood on the desk is for the founder's screen in the morning; the class (the engine's registered() check and its fallback, the scheduler's logon mode) is the update-return lane's for 0.3.24; the locked PC 1 jobs stay parked. THE CA4 PROTOTYPES (the research lane, counter-asic-4 6404f62b): two experimental classes behind the pack, no consensus change: +shlx (the shadow's 256 instructions and 27 passes split into 16 sub-blocks of 16, each run after its load) and +mm (R int8 mma u8 tiles per iteration after the shadow; CUDA native PTX, the shuffle reference on Metal and OpenCL; the verifier scalar plus AVX2, SIMD pinned equal to scalar on 64 seeds); the suite green (64 + 7 + 4 + 19 + 2 + 7), the pinned packs byte-identical; packs exported with every OVERALL PASS (mx8_sh256x27 control, mx8_shl256x27, mm128, mm512, mm1430 at 11,440 tiles per hash); their card rows on PC 1 behind the helper; by construction neither lowers the premium (the per-load placement moves the chip's capex, the tile block its k floor). THE LEDGER CLOSE landed the chip rows on the mirror's master at b94a77ad (22:56 BST): X35 and X36 restated, AP-F8-1 with the eight-of-eight sentence, X37 new (the class v4 premium: measured, levers in flight). THE RECORD LANDED (23:24 BST): the regroup 2336a3c5, the outside-check rewrite and chip model 5.11 (6c19c790) and the status 015cc839 picked onto ca3-coord-record from the mirror's master and merged as ddfaf7a7 through the gate (GREEN, 7 checks in 30 s on f252b514); the first pick hit the audit lane's best-points clause in the litepaper, claims and evidence pages and the resolution keeps master's text with only the escrow sentence replaced by "No outside review has run yet." (main: the right sentence); the design pass stays on ca3-coord for its own landing on main's word after the captures. ADV-ACCEPT-3 CLOSED (the v5 lane, 23:12 UK): 8ca66afa closes its class as stated (the 9.0 percent of rewritten 256th-attempt programs the rule refuses are repaired for part (a) and re-drawn under the 256-candidate scan; the known-failed test on adv3/steer/2, five more steer rows passing); ledger row AP-F8-3 written (sub-version 3's last resort recorded unreachable and unverified, class v5's verified) at class-v5 7f58af97 with the v5-kits branch merged (the OpenCL, NVRTC and Metal hosts with the leaves upload, the kit scripts); the kit zip rebuilt from the merged tip, /srv/artefacts/packs/packs-ca3-v5-20261007T221001Z.zip sha256 4aaf9b9edfad0e466f6b6b59051250afad6a8e0a340728ec068bec48113c0fc9, the packs and the fingerprint 82b19cbde8557ea5 unchanged; Metal, Apple OpenCL and CUDA agree; AMD and Intel fingerprints owed. A GAP: tools/ledger-page.mjs renders only [A-Z]\d+ ids, so no AP-* row (AP-F8-1 to AP-F8-4) reaches /ledger; the site audit lane widens the regex tonight as its own commit with a known-failed case. THE SPEC SPLIT: the site audit lane holds 1.4.3, 1.4.6 and 1.13 (the acceptance-rule rewrite on spec-accept-23) and builds tools/ci/spec-constants-check.mjs, a constants table in the spec parsed against the crate's pub consts (known-failed first) with the class v4 test vectors stated in 1.4.6, since the attack-pass lane has no read-back test and writes none; the hash lane sent it the file and line of every constant from 017e7037 (= master's igneum-pow byte for byte, cf7d6ccb) plus ACCEPT_TAG, the window cap literal in distinct_ratio_pass and the full Devnet 3 genesis hex, no wrong values, one text quirk: the (c) reject prints "limit 163" while MAX_SATURATED is 164 (the first refused count); main's ruling: the spec words the constant, the message string is corrected on the post-freeze line, never in the frozen 1c420786. The v5 lane's 1.4.7 and 1.8.6 are on both mirrors at class-v5 73daadc2 (23:23 UK; full gate GREEN 58 checks at 066c9cbb): class v5's load class, generator 5 and the id, (c''') with the 0.995 floor and the census, the verified last resort, AP-F4-1 and AP-F1-1, the activation object byte 6 and the seven-window 95 percent signal, the test vectors (the three pinned packs, seed 100767, day 29,337, adv3/steer/2), 1.4.7.6 the constants table in the audit lane's shape (Constant, Value, Where); the state leaves (IGSD1 stream, leaf derivation, keyed sample, the leaf line before M_0, the per-epoch refresh and the witness, the measured cost). THE ERA-DRAW MECHANISM (the crypto lane's adv-cache-2, 6e34ebe3, 23:1x to 23:3x BST; report-chained-cache-2.md section 2.3, the 61-program table: 2 real, 27 drawn-era with epoch and era hex, attempt, id, R, site and ratio, 32 devnet-era controls): the mild residual class has its mechanism; a product's biased low bits (P(bit 0) = 1/4, measured exactly) survive the odd stride multiplier and the stride rotation places them at address bits R and up, inside the 28-bit item index unless R is 28 or more; the devnet era draws R = 29 and cuts them off, so 2 of 32 devnet-era programs carry a site over 1.04x while 13 of 27 drawn-era programs (R 3 to 22) do, 8 over 1.2x, worst era-drawn-28 site 15 at 1.7451x and era-drawn-25 site 11 at 1.3571x; under the 2 GiB genesis dataset (D = 29) R = 29 would show it too; the devnet's cleanliness is an era-draw accident, the chain prevalence is the drawn-era figure. The price to a partial-store chip stays under 0.1 percent of a hash's reads per site, so no chip number moves. Disposition: the class v5 (c''') census was already across drawn eras (each of the 4,600 f8 seeds carries its own era bytes), so the 2.435 percent and the eight of eight stand; the pointed reading runs on box 2 (the v5 lane, about 20 minutes from 23:3x): the 2^20 floor read on the 27 drawn-era programs plus era-fixed-20 and four devnet controls, reporting how many of the eight over 1.2x and the band 1.04x to 1.2x the 0.995 floor refuses; the value-level question (biased product bits feeding an address, independent of the distinctness ratio) and the era draw's R range go to the CA4 file as a named requirement with this reading as its evidence, and the research lane's per-load census gains a drawn-era split; nothing in class v4 or v5 moves without main's word. THE ATTEMPTS CENSUS on the frozen sub-version 3 rule (adv-accept row 90, 23:24 BST, 10,000 seeds): 21,119 rejected candidates, by first failing part (a') unfresh 83.3 percent, (a) stale 11.7, (b) no injecting write 3.1, (c'') low-entropy site 1.1, (c) constant bit 0.4, (c) saturated 0.3, (c') 0.1, the distinct-address floor 0.04, lane-constant and bias 0; per-candidate rejection 0.6787, flat at 67.5 to 68.7 percent over attempts 0 to 3 (independent draws); accepted-attempt mean 2.112, max 24; 0 exhaustions; P(256 consecutive rejections) 8e-44 per seed, so the last-resort draw is unreachable by chance and the attempt index is no lever for a seed-steering attacker; accepted programs' distinct-item mean 127.95 of 128, minimum 123.67; spec 1.4.6's 5.14 percent (the class v3 census) is stale against it, the audit lane rewrites; the second 10,000 queued on build-1. Also PASS: the line census at 2^35 + 3 x 2^33 and the 16,384-day weak-day scan. THE PC 1 QUEUE TONIGHT (the hash lane): run-ca3-pc1-amd-family-20261007-e exit 0 at 22:09:25Z (the 9070 XT alone, gfx1201, driver 3683.0, 32 CUs, three runs every row exact against the alu chain; step costs as a ratio to alu 741 G steps per second: rotr 1.05, shflx 0.89 (bperm native), shl 0.92, shr 0.99, bfe 1.03 native and 0.83 C sequence, andn 0.93, perm 1.21 emulated (perm_amd refused), popc 0.85, clz 0.83, sel 0.72, shfla 0.77 (bperm), dot4 0.75 native (dot4_khr refused), mm8 1.20 (gfx12 path, unverified); the khr and intel shuffle builds refused as on 6 October); the shipper's 0.3.24 host slot holds PC 1; on its "slot closed": fetch-ca3-v5-kit-20261007 (the 4aaf9b9e zip), then run-ca3-pc1-v5-amd-bench-20261007 (the v5 lane's script, the 9070 XT by name, beside the miners, about 3 minutes), lock-free and non-elevated, quiet. The Intel fingerprint: main first routed it to PC 1, the hash lane's device lists (the 22:09Z --list, the kit README) show no Arc on PC 1, and main's second word places the Arc B580 as PC 2's eGPU (tonight's PC 2 crash was an Intel driver install over that card while it mined); the job (tools/class-v5/pc1-intel-v5-bench.ps1 at a4b08245) moves to PC 2 by job after the shipper's 0.3.23 take 3 smoke and the update-return lane's scheduler proof have reported on that box, never concurrent with an install or a build there, the same lock-free class; a fingerprint that differs from 82b19cbde8557ea5 holds that card's v5 kit out of 0.3.24 and the crossing time is stated on its page row. PC 2 carries the RTX 5080 since about 15:00Z (tonight's stock row is that card). THE HASH LANE'S LANDING (the derivation fix, the no-prompt rule, the PC 1 job scripts, the Ember core-clock knob 74585c91: the ladder below 45 percent in 100 MHz steps to a 20 percent floor, the stop rule at the knee or on a faulted row, lock_result and the card's lock_* fields, 18 Ember tests and the app crate's 158 green on box 2, the 1 percent tolerance landing the 5090 at 1,854 MHz on tonight's rows and 1.5 percent at 1,300, the tolerance the manifest's; ledger row AP-F8-4) went RED once on the pre-public scrub (the founder's name in a ledger row and two script comments), fixed, the mirror's master merged in again, the gate rerunning from 23:2x; the merge commit follows. THE FLOOR'S FULL TALLY (adv-accept gap-deep4, 23:25 BST): the four deepest remaining 256-unit seeds all read under 0.995 at the acceptance sample (148927 at 0.9814, 150347 at 0.9896, 34501 at 0.9929, 29307 at 0.9912); the first three clean live (0.9998x to 1.0028x), 29307 at 1.29x on one item from a non-saturated source, no hot set by X_f. Over everything the lane read at 2^20: 8 of 8 live hot sets refused; 6 clean-live programs refused (false refusals) and 1 clean passed among the 9 deepest 256-unit seeds; 3 mild residuals missed at about 1.0004x. The lane's reading of why both sides exist: (c'') counts repeated word indices on the stand-in, which the live set usually spreads thin rather than concentrating, so a low ratio is not a hot set; that is the 2.4 percent clean rejection the floor pays, and a true hot set needs the value-level source test to be caught without it (the CA4 requirement). THE SPEC REWRITE committed on spec-accept-23 (the audit lane, 23:3x UK): 1.4.3 and 1.4.6.1 to 1.4.6.6 to the shipped rule at 017e7037, the shadow block in 1.7, the ninth era draw in 1.13.1, ledger AP-F8-5 (the stale spec text) with the public ledger regenerated, the two tables in the check's shape (Constants of the shipped rule: Constant, Value, Where, 17 rows; Pinned program ids: Seed, Attempt, Id, Note, 6 rows with Devnet 3's full genesis hash and the three must-differ ids); the full gate running; it merges the mirror's master after the hash lane's landing so the check and the text arrive together. THE PER-LOAD FIX (the research lane, counter-asic-4 2f718001, pushed 22:24Z; the fixed pack mx8_shl256x27_v2 22:29Z, attempt 3, id bd64b207a30413fb, the first export 854050a4293f0615 kept as the known-failed record): known-failed first at 22:16Z (tests/ca4_trace.rs on build-2): the first export derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism from the 64-seed census (29 of 64 seeds failing, up to 620 duplicate lanes a seed, sources collapsed to 1 to 17 distinct values in 32 lanes): a lossy base writer (mulhi, mul, or) followed by 27 passes of the 16-instruction map collapses the register before the next load, so the static last-writer rule catches only part of it. The fix in two layers: the static redraw (a sub-block writer of the next load's source drawn from the injecting families when it is mul, mulhi or or) and the dynamic acceptance test stepping the per-load sub-blocks in the order the class executes (accept.rs alu_step inside run_unit) with a new rejection DuplicateLanes (any load reading one address in two lanes of a unit), a rejected candidate redrawing the attempt. After, 22:23Z: 12,287 of 12,288 and 0 duplicate lanes on the genesis seed; the census (64 seeds x 2 units on a second dataset, 16,384 load rows) 1 duplicate pair in all (seed ca4-census/49 site 3, the chance floor of a 2^24 index space, about 0.5 pairs expected; the class v4 shape's own trace shows 2 of 12,288 from the same floor); the suite 64 + 2 + 7 + 4 + 19 + 2 + 7 passed on build-2. Owed: the Metal fingerprint (the Mac, one at a time under the measure lock), the F8-form uniformity on the fixed export through the attack-pass harness, the drawn-era split of the census (R 3 to 22 against 28 to 31) and the biased-low-bits requirement row from adv-cache-2, the PC 1 card row on both exports. Nothing in class v4 or v5 moves. THE "LIMIT 163" FIX (the hash lane): the one-line fix on a post-freeze branch off the mirror's master, pow-reject-text-24 at 79c5c07d (pre-push GREEN): the (c) saturated reject text prints its limit as MAX_SATURATED - 1 and names 164 as the first refused count, with the test the_saturated_reject_text_prints_its_limit_from_the_constant reading the printed limit back (green on box 2); the frozen 1c420786 line untouched; it lands with 0.3.25's line. The derivation fix's landing: the second gate run RED on the public-ledger check (AP-F8-4's last paragraph must start with one of the six status words), the row now closing "Status: Fixed (7 October 2026, night)" and docs/ledger-public.md regenerated; the third gate run from 23:3x UK. PC 2's Intel job prepared as run-ca3-pc2-v5-intel-bench-20261007 (the kit fetch to PC 2 first) behind the shipper's "PC 2 clear"; the CA4 packs job on PC 1 runs both per-load exports (dir and id on every row). THE FLOOR RE-CUT (main's ruling, the shipper 23:3x UK): the 28,800 floor lost to the clock (the pairs, the hive kits, the fleet's fetches and the ten minutes after the last FETCHED cannot land before 23:52 BST, past the 22:52:16Z slip point), so the node lane re-cuts program_class_v5_activation_daa to 32,400 (epoch 9) on release-0.3.24-node, the same object otherwise (pairing 1c420786, chain id 4464 from the floor, the testnet re-cut inside); the new pin and its gates about 25 minutes from 23:3x; the crossing on Devnet 3 by height then about 01:52Z on 8 October (02:52 BST) at 1.0 DAA/s; the move minute after F9 and F1 PASS and the last FETCHED. THE ERA READING ON THE FLOOR (the v5 lane, box 2, 23:3x BST, igneum-pow at 73daadc2, the 2^20 acceptance sample): 0 of 29 of adv-cache-2's programs are refused by the 0.995 floor at their listed attempt, and the class v5 draw lands on the same attempt as class v4 for all 29; the six over 1.2x read minimum sites 0.9965 to 0.9997 (era-drawn-15's 1.51x site 14 at 0.9965 the lowest), the 1.04x to 1.2x band 0.9986 to 0.9998, the clean ones 0.9999 to 1.0000, the devnet-era controls 0.9996 to 0.9999. So the floor's statistic does not reach adv-cache-2's class: the distinct-index count at 2^20 reads concentration on FEW items (adv-accept's hot sets put 3 percent of a site's reads on 512 word indices, moving the collision count by thousands), not a diffuse excess over the top 0.1 percent of items (era-drawn-15's 1.51x is about +0.08 percent of the site's reads spread over 16,384 items, a few hundred collisions, inside the clean spread). Two classes, two instruments: the floor closes the few-item hot sets (8 of 8); the era-stride diffuse class needs the per-site item-share test at live scale or a draw rule on R and the shadow block's last write (the next class's row); its chip value is bounded by its own diffuseness (a 1 MiB hot table of the top 0.1 percent of items serves about 1.0024x at the worst site read so far, under the AP-F8-1 bound by an order). The v5 design's section 14 gains this paragraph with the 61-row log (era-drawn-25 to -28 and the 32 controls running; era-drawn-28 at 1.75x the one to watch) and its bound sentence corrected (the "top-0.1-percent share under about 1.3x" form, never served, lived in section 14 only); a ledger row for the miss asked. Nothing in the freeze moves. MAIN'S ROW WORDING for Devnet 3: a 0.3.23 node that has not updated falls off at the digest move minute (the fleet's named minute, about 00:52 BST at the latest), not at the 02:52 crossing; the row reads "update before or the node stops following Devnet 3; class v5 begins at DAA 32,400, about 02:52 BST". F4 ON CLASS V5 PASS (the attack-pass lane, 8ca66afa, build-1 under class adv, 379 s, ended 22:3x UTC; the agreed w32 convention, median 226, 2^24 chain days from 20,729): M1 0 of 2^24 days over 1.1x, the minimum cost 206 (day 27,016, 1.097x), so the bound holds with no margin and no day over the line, mean 225.79, sd 6.07 (the pre-rule census 5.69e-4 over, min 203); M2 0 days with k >= 2; day 29,337 redrawn under the rule (203 to 228), day 20,729 at 219 unchanged; AP-F4-1 FIXED-AND-PASSED; F9 and F1 under class release on build-1, lines within the hour. THE CA4 FILE (the research lane, 22:3x UTC, sections 20.2a and 20.2b): the drawn-era split of the per-load census: 16 eras over the fixed class, 2 units each, R under 28: 12 eras, 3,072 rows, 0 duplicate pairs; R 28 and up: 4 eras, 1,024 rows, 0 pairs; every era accepted at attempt 3; the adv-cache-2 reading written as a named requirement (value-level bit-bias of the index at a product-sourced site, judged across drawn eras split by R, owed for every CA4 class and the same item as class v5's acceptance; the per-load dynamic rule covers distinctness, not bias). Metal fingerprints (22:30 UTC, M5 Max under the measure lock): the fixed per-load pack ee5d7c71180e5ea7, vectors 3 of 3, 26.88 MH/s against the control's 27.01 (the placement costs Apple nothing); the tile packs bit-exact against the Rust verifier on the Metal reference path (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1); the Apple cost is the finding: 1,024 tiles per hash take 35 percent of the M5 Max's rate, 4,096 take 78 percent, so a tile shadow at the ALU shadow's premium would take the Apple tier out unless Metal gains an integer matrix path; the tile class moves from rank 3 to beside rank 5 until that path is measured. Main's rule: no served number mentions the per-load fix before its F8-form uniformity and drawn-era split (the split now read; the uniformity owed). THE PUBLIC SENTENCE ON THE FLOOR (main's wording, 23:3x UK): "eight of eight hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test", the same words on ledger row AP-F8-1 (landing from ca3-coord-record 6d09d96e with the two-instrument reading and the AP-F8-6 pointer), on AP-F8-6 and in the v5 design's section 14 (the v5 lane, class-v5 54e52b8a at 23:36 BST carrying AP-F8-6, F4's PASS in the attack row and its clock corrections: build-2 prints CEST, every page time re-read to BST); no served page carries a hot-set sentence tonight, so the sentence reaches readers through the ledger once the AP-* regex fix lands. F4's no-margin hold (the minimum accepted cost 206 against the 205 bound at day 27,016) is a record sentence, not a served number. ADV-MIXER-2 CLOSED (the crypto lane, 2a632579 on build/adv-mixer-2, 23:37 BST; 0.31 box-hours, 0 pod-hours): the redraw rule (continue the stream and redraw all 40 draws when the LUT cost A is 205 or less, or a 2-adder MUL, or all ROT equal) over 2^24 and 2^28 days leaves 0 days over 1.1x; 6.0e-4 of days redrawn once, 3e-7 twice, never three times; the mean cost unchanged; verdict BOUND for every chip, GPU and the verifier (gain 1.0 every day at 9,360 ops per item), FINDING on the per-day FPGA LUT-area reading only (2^-10.8 of days over 1.1x, worst 28 April 2050 at 1.113x), closed by the redraw rule or by the spec's O-1.10 day derivation; five lanes closed (adv-cache, adv-accept-2, adv-cache-3, adv-mixer, adv-mixer-2), four to the 00:00 reading (adv-accept, adv-accept-3, adv-cache-2, adv-mixer-3). THE HASH LANE'S BRANCH ON MASTER: da2fc101 at 23:37 BST (ca3-v4-amend a7ff10a2; the full gate GREEN, 69 checks in 351 s): the derivation fix with AP-F8-4 and the regenerated public ledger, the no-prompt rule (publish-jobs.sh refuses --elevated; playbook-quit-check rule 3), the PC 1 and PC 2 job scripts, the Ember core-clock knob for 0.3.24 (ember.rs, state.rs, engine.rs; 18 Ember and 158 app tests green on box 2), the ca3-v4-uniform parallel census; igneum-pow against 017e7037 differs in generator.rs (the recipe refactor, every id and pin unchanged), emit.rs (the one print) and tests/derivation.rs only; the shipper's tip for 0.3.24's engine work is this master. THE 0.3.24 NODE PIN RE-CUT (the node lane, every gate green at 22:39:31Z): c9e385eb on release-0.3.24-node (47b9b229 with Devnet 3's class v5 floor at 32,400, epoch 9, the same object otherwise; pairing 1c420786): build 22:34Z rc 0 (igneumd 7a841b20..., /srv/artefacts/0324-c9e385eb/node-lane), consensus 134 at gate priority, core 175, exec 47, miner 28, p2p-flows 38, pow 19; the Devnet 3 canary set with the new digest d0d6a4754f3bfc4a173aeaddbab0e151583047283932b70cbb8e27878c115e91 (byte 6, override refused, handshake, the shared-devnet dialler and a 2720d8d2 node refused); the testnet canary on b2e856ed unchanged. The floor from the 22:30:17Z read (DAA 20,268, 1.0 DAA/s): about 01:52:29Z on 8 October (02:52 BST), holding for a move minute up to a publish at DAA 25,200 (23:52:29Z, 00:52 BST). The fast-time SUMMARY on c9e385eb asked; the fleet lane asked whether its hub or any reader depends on build-1's three old-object Devnet 3 nodes (the seed on 27632, the observer node, node1), whether they join the move or retire, and which 0.3.24 node the DAA is read from after it; the crossing read at 32,400 and the TESTNET_PARAMS v5-at-0 re-cut follow on that node. THE FAST-TIME GATE ON THE RE-CUT: SUMMARY PASS (cross-0324-c9e385eb) at 22:49:32Z (23:49 BST) on the shipped 0.3.24 re-cut c9e385eb (igneumd 7a841b20..., igneum-miner 1e209b9e..., igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 22:36:25Z to 22:49:32Z, every check green: rung 1 by signal at epoch 6 (22:42:54Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (22:44:54Z, 4 of 4, 9,985 bps), 11 of 11 ids equal to the CPU verifier's, the stale node 86 of 86 refused with 0 accepted after the first refresh, the restart step across the boundary on a kept datadir resynced in 28.1 s with the catch-up done after 10 s and 0 of its own blocks during it, four sinks equal at 660, honest nodes 0 PoW rejections; record on v5-fasttime 76276be6, docs/design/class-v5-harness/fasttime/cross-0324-c9e385eb.json. The 0.3.24 move's gates left (the shipper's correction of this record): not F9 and F1's full 10^5 PASS (landing about 00:40 BST, too close to the 00:52 ceiling) but an F9/F1 interim line from the attack-pass lane read inside the five minutes before the minute showing 0 exhausted, 0 panics and 0 redundancy failures over everything drawn so far (16,003 seeds at 23:35 BST, max attempt 25), any non-zero holding the move, the full 10^5 the record line after; the minute named by the fleet on the last FETCHED plus ten once the build-server lane's c9e385eb pairs land. THE 61-ROW ERA READING (the v5 lane, box 2, 23:4x to 23:5x BST, docs/design/class-v5-harness/v5-listed-adv-cache-2-full.log): 0 of 61 refused by the 0.995 floor at the table attempts (the two real programs, 27 drawn-era, 32 devnet-era controls), every class v5 draw on the class v4 attempt; era-drawn-28 (id 5e9eb01efbbf653e, attempt 6, R 15, the worst of adv-cache-2's census at 1.7451x) reads its biased site 15 at 0.9969, over the floor by 0.0019; era-drawn-25 (1.3571x, R 21) site 11 at 0.9994; the eight over 1.2x span 0.9965 to 0.9997 while the eight few-item hot sets sat 0.003 to 0.013 under the line. Main's sentence opens AP-F8-6 and section 14 verbatim with the two-instrument reading under it. THE CLASS V5 ATTEMPTS CENSUS for 1.4.7 (1,000 f8 seeds through the chain draw, v5-attempts-census-1000.log, the crypto lane's form): 3,219 candidates, 2,219 rejected, per-candidate rejection 0.6893 (sub-version 3: 0.68), accepted attempt mean 2.219, 0 exhaustions, P(256 consecutive) 4.4e-42; first failing part (a') 83.4 percent of rejections, (a) 10.7, (b) 3.0, (c'') 1.2, (c''') 1.0 (0.7 percent of candidates, one in 140: the floor's own share, 0.045 on the attempt mean), (c) 0.7 together, (c') none; the 5.14 percent of class v3 that 1.4.6 quotes is the audit lane's to replace. Both on class-v5 at 3b1dffd6 with main's sentence (891dd008), the mirror's master merged (e0471019: AP-F8-1's update and AP-F8-4 taken, the program-id recipe form with the state tag, no conflict), the design page's pre-public scrub (the founder's name six times, gone), M35's status word and the regenerated public ledger; the push waits on the full gate and the pinned-packs test on the merged tree (the proof that e5a4ac5978462156 and the other ids still derive under master's recipe form). THE 00:00 BST READINGS (the crypto lane; the verified roll-up of all nine lanes in section 13 of in-house-pass.md on crypto-engage, every branch tip read from the mirror and igneum-pow identical to 017e7037 on each). adv-accept, tip a7c49399 (about 5.5 box-hours, 0 pod-hours): 182,646 distinct accepted programs drawn (18 percent of the 10^6); eight pass every part of the frozen rule and flag the live hot-set test at 2^24 (X at 0.1 percent +0.102 to +0.221, 1.54x to 2.24x), all in the lowest 34 stand-in-ratio seeds against 0 in 20 random; each about 1 MB of items holding 0.26 to 0.41 percent of reads, 1.002x at the largest; the mechanism a near-saturated source at one site mapped by the era stride to one fixed item (plus two lesser shapes); the exemplar reads the same under the class v5 dataset. Against the class v5 floor: 8 of 8 refused; 3 mild residuals missed (adv-cache-2's rotation class, a load_index question not a floor question); 6 clean programs refused among the 9 deepest (the 2.4 percent). Q2 BOUND (54 programs plus 17 reads, 0 disagreements). Row 90: 0.6787 per candidate, (a') 83.3 percent, 0 exhaustions, P 8e-44. Partial named: 18 percent of seeds, 54 live rows, row 90 at half; a longer pass adds rows of the same shapes, not a different answer, unless a seed reads a hot set over 1 percent of reads, which 182,646 draws did not produce. THE PER-LOAD CLASS CLOSED (the research lane, for main; clock readings UTC): the per-load shadow fix held for distinctness and then met the value-level requirement from adv-cache-2, and the construction did not survive it; the per-load 16 x 27 class is dead as a chain class. 22:44 the attempt verdicts on four seeds (igneum-genesis 0 of 32 accepted); 22:47 the 64-seed census under the full rule (duplicate lanes at a load row plus the one-count of every index bit per site over the 64 units, 6-sigma band): 22 of 1,621 candidates accepted (1.4 percent), 42 of 64 seeds exhaust the chain's 32 attempts (an epoch without a program); the first failing test per candidate: biased index bit 775, duplicate lanes 643, the base rule 110, (b) 43, (a) 28; candidate 0 of the class carries index bit 0 set in 40 of 1,024 addresses (z 29.5); 22:52 the suite green (64 + 5 + 7 + 4 + 19 + 2 + 7); the acceptance rule with BiasedIndexBit for this class and the tests pushed as the record, the file's 20.2a closed. The structural reason: 27 passes of a 16-instruction map right before a load is an iterated small function and collapses or biases the load's address register before any base instruction re-randomises it; the class v4 shape has 64 base instructions and 16 loads between its block and every load. Both exports were accepted only because the rule did not model the placement; their PC 1 rows stay as an energy reading of the placement, labelled unsound. Rank 4 and the USD 200 M capex row rest on a construction not shown to exist (chip model 5.11's clause marked so in this landing); the sound form is one pass of a 432-instruction sub-block per load (a program segment, not an iterated map), a new class to draw, accept and measure, not tonight's. Replicated by a second instrument: the class v4 shape on this pre-amendment generator carries the adv-cache-2 product bit at address bit R exactly in 14 of 17 drawn eras (one-count 250 or 780 of 1,024, z 15 to 19), 0 duplicate pairs across the eras. What stands from the two prototypes: the tile block (bit-exact on the Metal reference, the AVX2 verifier at 0.047 us per tile, the Apple emulation cost 35 to 78 percent) awaiting its 5090 rows; the per-load placement closed. THE SPEC REWRITE ON MASTER (the site audit lane, 8b834634 at 23:56 BST; gate GREEN on 64e2a91b, 71 checks; the igneum-pow suite green on the box for that commit with derivation.rs and spec_readback.rs): spec 01 sections 1.4.3 and 1.4.6.1 to 1.4.6.6 rewritten to 017e7037 with the 20-row constants table (ACCEPT_TAG, the window-cap literal, MAX_SATURATED as the first refused count with the 163 message noted) and the 6-row pinned-ids table with Devnet 3's full genesis hex; the shadow block in 1.7; the ninth era draw in 1.13.1; tools/ci/spec-constants-check.mjs in the gate (known-failed first, every Constant | Value | Where table, pending rows skipped while absent); igneum-pow/tests/spec_readback.rs (ids derived through the crate, each class v4 row drawn to its attempt); ledger AP-F8-5 after AP-F8-4; the ledger-page fix (both heading forms, the pass as its own section, known-failed self-test in the gate; AP-F8-1, AP-F8-4 and AP-F8-5 render on /ledger); the fud-ledger's two prize clauses and "paid independent cryptanalysis" removed at the source so the regenerated page carries neither (commit 90424d5a, merge 64e2a91b). A HARDWARE FACT IN DISPUTE, for main: tonight's 5080 efficiency rows came from PC 1 jobs (run-ca3-pc1-v4-eff-5080-20261007-b and -c), the audit lane's record reads the RTX 5080 and the Arc B580 on PC 1, the hash lane's 22:09Z device list on PC 1 shows the 5090, the 9070 XT and the 4070 only, and main places the 5080 and the B580 on PC 2; identity-check.sh's "PC 2" substitution text names cards and is left card-free until the PC 2 job's own --list settles which cards sit where. THE IDENTITY CHECK'S PC 2 TEXT (the CI steward, 00:05 UK on 8 October): tools/ci/identity-check.sh rewrites "PC 2" card-free as "the second Windows rig" (commit 40f2be54, merge 0d2cf334, gate GREEN 71 checks, identity grep 0 hits over 306 export files and 52 served pages); line 69's PC 1 list untouched; the reason recorded in a bash comment above the perl call. THE 32,400 FLOOR LOST (the node lane, 00:0x UK on 8 October): dn3-g1's chain read DAA 25,126 at 23:52:03Z and 25,169 at 23:52:38Z, so the publish DAA passed 25,200 at about 23:53:09Z with no 0.3.24 move made (build-1's three Devnet 3 nodes last restarted about 21:31Z on the 0.3.23 move; the old seed holds 38 peers on ba75bf6f; no move minute was named). The next boundary is 36,000 (epoch 10), about 02:52Z on 8 October (03:52 BST) at 1.0 DAA/s, holding for a publish up to DAA 28,800 (about 00:53Z, 01:53 BST). Two routes put to the shipper and main: the same re-cut script on release-0.3.24-node (program_class_v5_activation_daa 36,000, nothing else, the same gate set, about 20 minutes to the pin line), or the fleet names its minute first and the floor is cut from it in one go (publish DAA plus 7,200 to the next 3,600) instead of a fourth chase; the pin c9e385eb stands meanwhile. THE FLOOR RE-CUT FROM A NAMED MINUTE (the shipper, 00:1x BST on 8 October, under the slip rule main set with the object commit): the floor re-cuts once more to 39,600 (epoch 11, about 04:52 BST) from a move minute the shipper named: 02:00 BST on 8 October, or the fleet's last FETCHED plus ten if later but before 02:53 BST (DAA 32,400, the ceiling); the node lane's pin line in about 20 minutes with the new Devnet 3 digest; the F9/F1 interim read at 01:55 BST; the publish minute equals the move minute (the apps' entries at or after it); the fast-time SUMMARY PASS reruns on the new pin as part of its gate set; the cause of the lost floor named: the c9e385eb pairs and the two PC jobs unreported for forty minutes, so the fleet had nothing to point its move file at. "slot closed" on PC 1 still waits on the host job's exit. THE 0.3.24 NODE PIN AT 39,600 (the node lane): dfbd1e10 on release-0.3.24-node (both mirrors, 23:54:13Z) = c9e385eb with program_class_v5_activation_daa 39,600 (epoch 11), nothing else; pairing igneum-pow 1c420786; every gate green at 00:01:52Z (build 23:56Z rc 0 at gate priority, igneumd 4870ccf2..., igneum-miner aa8c2978..., /srv/artefacts/0324-dfbd1e10/node-lane; pow 19, consensus 134, p2p-flows 38, exec 47, core 175, miner 28); the Devnet 3 canary set (23:56:33Z to 23:58:13Z): digest b1ba78229b069dc395fa666638a686a66615eb760d251798adfa6a654a415f82 on igneum-devnet-3 from ba75bf6f, object version 6 stamped (block version 1538), the override file refused, shutdown 2,015 ms, two empty nodes handshaking on it, the shared-devnet dialler rejected, a 2720d8d2 node refused on the digest both ways; the testnet canary b2e856ed unchanged (byte 7, a live old-object testnet node refused). The cut's read: dn3-g1 at DAA 25,169 at 23:52:38Z (1.0 DAA/s), the publish DAA at the named minute 01:00Z about 29,211, plus 7,200 = 36,411, the boundary 39,600 about 03:53:09Z on 8 October (04:53 BST), holding for a publish up to DAA 32,400 (about 01:53:09Z, 02:53 BST). The one gate running: the fast-time pair on dfbd1e10 (about 13 minutes from its start). c9e385eb is void as a pin; the F9/F1 interim read armed at 00:55Z. THE TWO PC QUEUES AT 01:03 BST (the hash lane): PC 1's "slot closed" has not come (the shipper's 0.3.24 host job, the build-server lane's, took the slot at 22:13Z for an expected two to three minutes; nothing reported in 110 minutes); nothing of the hash lane's has run on PC 1 since 22:09:25Z; the v5 kit fetch and the 9070 XT v5 bench are prepared and unpublished (tools/ca3-v4-amend/pc1-publish-20261007.sh, steps v5-kit and v5-amd), so no 9070 XT class v5 fingerprint exists yet; the lock protocol holds unless main says the lock-free pair goes ahead of the silent host job. PC 2's "clear" has not come either (the 0.3.23 take 3 smoke and the scheduler proof unreported by either lane); the Intel job is prepared and unpublished. THE HARDWARE FACT, read from tonight's PC 1 lines: nvidia-smi on PC 1 lists GPU 0 RTX 5090 (bus 01:00.0) and GPU 1 RTX 5080 (bus 0D:00.0); its OpenCL list carries the RX 9070 XT (gfx1201) and the integrated gfx1036 and no Intel platform; so the 5080 is on PC 1 (the audit lane's record right, the 22:09Z device-list summary short by one card) and the Arc B580 is not, which agrees with main's word that it is PC 2's eGPU; the kits row, the bench notes and identity-check's card-free PC 2 text stand on that. The locked PC 1 jobs stay parked (the 5080 full grid, the third 5090 pass, SM-sparse, the microbench and packs knee states, the two Ember tunes, the hot-table ldcs rows); the lock-free CA4 rows queue after the v5 bench on the same "slot closed". THE 00:00 BST READINGS, THE OTHER THREE (read by the crypto lane from each branch's report on the mirror at 01:03 BST; the roll-up section 13 of in-house-pass.md at crypto-engage c84ba51b with adv-accept's reading at 1b4e07ff; all nine branch tips read back from the mirror and igneum-pow IDENTICAL to 017e7037 on every one: adv-mixer d2ba3134, adv-mixer-2 2a632579, adv-mixer-3 4ebe2455, adv-cache 555c3e42, adv-cache-2 9384ee09, adv-cache-3 9452c0bf, adv-accept a7c49399, adv-accept-2 92168536, adv-accept-3 0c150e3c). adv-accept-3 (exhaustion or steering of the draw), tip 0c150e3c, every sweep ended 23:05 BST, about 3.3 box-hours, 0 pod-hours: Q1 exhaustion BOUND (per attempt accept 0.323, reject 0.677 ((a') 0.568, (a) 0.079, (b) 0.022, dynamic parts about 0.009), geometric histogram, P(exhaust) 0.677^256 = 4.6e-44, 0 of 16,337 seeds at the cap); Q1b the last resort FINDING (correctness; the mirror fired at cap 256 byte-identically; of 3,000 last-resort programs the real rule rejects 271, 9.0 percent: 251 by (a), 14 by (b), 6 by (c) distinct sum; handed out unchecked; unreachable; closed in class v5 by 8ca66afa, AP-F8-3); Q2 steering BOUND (45 of 48 planted rows fired, the real rule rejects every effective plant by (a'); 975 seeds at the first part, min ratio 0.998, 18 of 18 chain re-draws equal); Q2b the price of a seed property at 1 in 10^6 tries is a shadow block with 38 multiplies of 256 against a mean 74, about 2 to 3 percent of the f = 1 chip's energy per hash, the load critical path worth nothing at the memory activate ceiling; Q2c the 256-unit ratio is noise as a selector; Q3 program id FINDING (documentation: the "sub/" || 3_le16 suffix omitted from program.json and spec 1.4.6; a text-derived implementation computes 30956569d8f3d8d7 for Devnet 3 against the pack's fce15bf61030be57; 0 collisions over 10^7 pairs; fixed as AP-F8-4 at da2fc101); Q4 determinism DONE (the (c'') f64 compare never disagrees with the integer rule on any of the 2^20 + 1 values, margins 0.32 to 0.44 counts; a second interpretation agrees on 5,748 of 5,748 verdicts of 1,792 seeds); Q5 the era lever BOUND (400 eras, no stride under NAF weight 7, all 31 rotations, 354 distinct interleaves; epoch 0's accepted attempt is 3 under every era, so the era moves the address map, not the program). Partial named: the steering sweep at 975 of a planned 10^5 seeds (about 8 box-hours more at 32 cores). adv-cache-2 (the hot-set attack), tip 9384ee09 at 23:52 BST, about 2.2 box-hours by wall times threads over 96 (the boxes at load 400 to 600 for the first two hours), 0 pod-hours; two shards still queued at 00:00 (lines-2e30-s2c, warps-devnet-2e25-v2), named partial: Q1 the line index PASS (pooled 16 days; segments max +4.84 sigma against a control's +4.24, lines +5.61 against +5.35, chi2/dof 0.99937, top 0.1 and 1 percent of lines 1.0003x and 1.0002x of control; the 2^35 + 3 x 2^33 census all PASS; 0 mirror mismatches); Q2a the real programs PASS on the hot-set test (devnet at 2^26 1.0002x; Devnet 3 at 2^26 items 1.0071x, lines 1.0000x) with the FINDING at Devnet 3 site 0; Q2b all 64 programs done, every one clear on the hot-set test (items 0.9993x to 1.0075x of the windowed control) but the site class as recorded above (13 of 27 drawn-era over 1.04x, 8 over 1.2x, worst 1.7451x; the v5 floor refuses 0 of 61; AP-F8-6); Q3(1) steering by t PASS (worst cell 3.95 sigma in 2 x 2,112 cells); Q3(2) the weak-day scan PASS over 16,384 days (2^30 derivations in 707 s; worst per-day max bucket +8.13 sigma against the control's +7.78; the plant fired at +1,090); Q3(3) the window layer: the exact distribution matches the 4,096-program census to four digits (top quarter mean 0.3382, top half 0.5811), with a FINDING against the chip model's table: the f = 0.25 and f = 0.5 partial-store rows overstate the recompute share by up to 1.8x at f = 0.5, the full-store (f = 1) verdict unchanged (a correction owed in chip-model-v3's partial-store rows; no served number rests on f under 1); Q4 the prices: the only measured excess over f is the window layer's and the line reference multiplicity (a hottest-lines half store hits 57.8 percent instead of 50 at a higher miss cost than the stride). adv-mixer-3 (the statistical distinguisher and round margin), tip 4ebe2455 at 00:41 BST, still RUNNING at 01:03 (Q3 and Q4 at k = 8 on day 20729, queue 07 in the pool, the SAT ladder at k = 3 timed out; the total box-hours the lane's to give): Q1 the exhaustive round-0 line-index census over all 2^32 t PASS to k = 8 on days 20729 and 20733 and at k = 2, 3, 4, 8 on 20730 (z within 1.5); Q2 single-bit avalanche FINDING at k = 1 (354 and 266 holes, 130,000 cells beyond 6 sigma, the known one-application diffusion), PASS from k = 2 at 2^24 (0 holes, worst z under 5.3 through k = 8); Q2b the t-bit avalanche the same shape; Q3 differential multiplicity over 576 low-weight differences FINDING at k = 1 (695 and 537 deterministic output bits), PASS k = 2 through 7, k = 8 running; Q4 and Q4b linear correlations PASS from k = 1 (worst c 0.00046 to 0.00062, z under 5.1); Q5 rotational-XOR PASS from k = 1; Q6 SAT: k = 1 SATISFIABLE in 137 s (t = 0x49880000 verified through the real code), k = 2 and 3 TIMEOUT at the one-hour cap. The round margin as it stands: no statistic survives 2 of the 8 applications between reads; a chip gets nothing from the k = 1 findings because every read sits behind 8. The lanes' own lines go into section 13.1 as they arrive. THE LANES' OWN 00:00 LINES (adv-accept-3 and adv-mixer-3, 01:0x BST, in section 13.1 of in-house-pass.md): adv-accept-3's P(exhaust) refined to 1.0e-43 per epoch seed from 62,240 full-rule candidates plus 3.0e6 static candidates; Q2 steering BOUND over 19,975 full-rule and 1e6 static seeds, no property buying over about 1.03x at 1 in 1e6 tries; a second documentary FINDING: an implementation written from the spec text (not the code) at 017e7037's spec differs on 264 of 400 epoch programs, the same text-against-code gap as the id suffix (the audit lane's rewrite 8b834634 with spec_readback.rs is the fix; the proof that it closes this is a re-run of the text-derived implementation against the rewritten text, asked); a plant note: the floor-0.97 known-failed variant did not fire because the (c'') ratios are bimodal (accepted 0.989 to 0.999, rejected 0.814 to 0.966), replaced by a single-pass (a) variant that did; 3.3 box-hours, nothing running. adv-mixer-3: about 3.0 wall-hours of sweep plus 4 single-core CaDiCaL hours; the round margin stated as 6 of 8 applications between reads and 70 of 72 per item on every measured statistic, the k = 1 effects one mechanism (the lowest-set-bit trail through one application, dead once both addends carry a difference), nothing saving one application against 9,360 ops per item; still running at 4 cores on build-1 (2^27 and 2^28 avalanche rows, finish about 03:00 BST) and the day-20733 SAT ladder on build-2 (about 03:45 BST); not attempted: multi-bit linear masks and a MILP trail bound. adv-cache-2's own line still owed. ADV-CACHE-2'S OWN LINE (01:05 BST, tip 3f50d6c4; section 13 of in-house-pass.md now carries every lane's reading in its own words plus the verified roll-up): the drawn-era prevalence read on the SAME 32 base programs is 2 of 32 under the devnet era against 16 of 32 under drawn eras (8 over 1.2x, worst 1.75x), the mechanism carried by rotl(x times M, R) into the item index unless R is 29 or 30 (2 of 31 rotations), with a sub-class of warp-uniform sources once in 16,000 warps; the window layer's price restated: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the chip model's partial-store rows overstate the recompute share by up to 2.3x on these programs, the f = 1 verdict unchanged (the correction to chip-model-v3's partial-store rows is the coordinator's next commit); partial named (the drawn-era windows census of 4,096 and one line shard in the pool); the longer-pass line: the biased-site rate per era in closed form (the R in {29, 30} rate 2 in 31) and a 2^28 read of the worst site. Nothing of the pass stands between the pool and a higher-class job except two pre-emptable shards on box 2. THE SPEC-TEXT RE-DERIVATION ORDERED (01:06 BST): adv-accept-3 re-derives its 400 epoch programs from the rewritten spec text alone at master 8b834634 (1.4.3 to 1.4.6 grown from 79 to 198 lines with the constants and pinned-ids tables), lease pool 16 --min 8 class adv, row Q4c in its report; the expected reading 0 of 400, any non-zero naming the diverging sentence to the audit lane; the proof that AP-F8-5 closed the text-against-code gap. THE CHIP MODEL'S PARTIAL-STORE ROWS carry a second correction (section 5, 8 October 2026) from adv-cache-2's window-layer reading: a chip holding the hottest f of items serves 0.4219, 0.7188 and 0.8907 of reads at f = 0.25, 0.5 and 0.75, so the uniform-store rows overstate the recompute share by up to 2.3x; the f = 1 row, the SRAM column and the full-store verdict unchanged, no served number on f under 1. THE CLASS V5 PACKS TEST ON THE MERGED TREE (the v5 lane, 01:0x UK): the job ran on box 2 the minute two adv-accept holders ended (65 cores; no lease fault, plain starvation before); 19 passed, 1 FAILED: v5_pack_is_the_v4_program_over_the_state_leaves (tests/packs.rs:977), the byte-for-byte compare of every pinned pack file with the crate's export. The ids are EQUAL (v4-genesis exports a217c7f698880830 as pinned; the state tag rides in master's recipe form unchanged); what differs is the program_id_derivation TEXT in program.json, which master's export (the hash lane's AP-F8-4 read-back form) now writes as "... || attempt_le32 || 'sub/' || sub_version_le16" for generator 4 while the pinned packs carry the pre-suffix wording. Disposition: the three pinned packs re-exported from the merged crate (text only; the ids, kernel texts, leaves and the fingerprint 82b19cbde8557ea5 must come out byte-identical, proved by the same test); the CLI rebuilding on build-1 from 51aa5bc4, the export from the box's IGSD1 streams, the packs test and the full suite on box 2 at 16 cores, then the push; readiness about 01:35 UK. The 0.3.24 kit zip (packs-ca3-v5-20261007T221001Z.zip) carries the old derivation text in its program.json files: a text field only, no id, kernel or fingerprint change, so the kit stands for 0.3.24 and the shipper is told; the re-exported packs go in the next kit. THE ONE 0.3.24 KIT, NAMED for the shipper (01:1x BST): packs-ca3-v5-20261007T183921Z.zip, sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e, byte-identical to the frozen 1c420786 the pin pairs with; the fleet keeps placing it. The 23:11 zip packs-ca3-v5-20261007T221001Z.zip (sha256 4aaf9b9e..., /srv/artefacts/packs/ on build-1, from class-v5 7f58af97) carries the same packs, ids, kernels, leaves, fingerprint and derivation text and differs only in the merged kit host code and scripts beside the packs; it is the bench lanes' kit for the fingerprint jobs. The coordinator's earlier line naming 4aaf9b9e as the 0.3.24 kit was wrong and is corrected here. THE SPEC-TEXT READ-BACK RUNNING (adv-accept-3's Q4c, 01:11 BST on build-2, lease pool 16 --min 8 class adv): the same 400 epoch seeds re-derived from the spec text at master 8b834634 alone (1.3, 1.4.2, 1.4.3, 1.4.6, 1.6, 1.7, 1.13.1; a fresh text interpretation), compared field for field with the chain draw; the count about 01:21. One sentence already named divergent before the count: 1.4.6 part (c) cites dataset_elem(idx, S[0], S[1]) "of verify.rs" without stating its six operations, so part (c) cannot be computed from the text alone and the derivation takes that one function from the crate; the audit lane is to state the closed form's six operations in the text or the constants table, else 1.4.6 stays code-dependent on that line. A NINTH LIVE HOT SET (adv-accept, 01:12 BST): seed 228763 (id 2c4be0f6dc44c423, stand-in 0.9820) at 2^24 (X at 0.1 percent +0.118, 1.82x the window model), its single hottest item 0xe2cc96 at 1,218,380 reads, 0.057 percent of ALL reads, the largest single item of the pass (40x 100767's), from a NON-saturated source r0 at site 9 (the sel register), saturated-source share 0.000: the third shape at scale, a value-level concentration neither (c') nor a saturation test can see by construction; its 2^20 ratio against the 0.995 floor lands in minutes and decides whether the floor's instrument reaches it (if missed, the exemplar for the next class's non-saturated case). 638990 reads 1.51x beyond the gate with one item at 0.027 percent (r0, no saturation), no hot set; 623492 clean. Tally: 9 hot sets in 37 tail seeds against 0 in 20 random, 269,250 programs drawn; the price unchanged at 1.002x (0.27 percent of reads on 1 MB; one item 64 bytes). The public sentence's "eight of eight" moves to "nine of nine" or gains the first miss when the ratio reads. THE FLOOR REACHES THE NON-SATURATED SHAPE (adv-accept gap-tail3, 01:13 BST): seed 228763 reads minimum site 9 at 0.9809 at the 2^20 sample, the lowest of the pass, REFUSED; 638990 site 2 at 0.9872, REFUSED; 623492 (clean live) site 0 at 0.9922, REFUSED, a seventh false refusal. Final tally over everything the lane read at 2^20: 9 of 9 live hot sets refused (0.9809 to 0.9919), both single-item programs refused, 7 clean-live programs refused and 1 passed among the 12 deepest 256-unit seeds, 3 mild residuals missed at about 1.0004x. The reading: the distinct-index ratio reads any few-item concentration whatever its source, saturated or not, and misses only the diffuse era-stride excess; the class v5 floor closes the hot-set class entire at the 2.4 percent clean-rejection cost; the next class's value-level test is for the diffuse class alone. The public sentence reads "nine of nine hot sets refused" from here (the v5 lane's follow-up cfce57ea rides its push; AP-F8-1 on master updates with the next record commit). THE FAST-TIME GATE ON dfbd1e10: SUMMARY PASS (cross-0324-dfbd1e10) at 00:12:57Z on 8 October (01:13 BST), the shipped re-cut's binaries (igneumd 4870ccf2..., igneum-miner aa8c2978..., igneum-pow 1c420786), build-1 under lease pool class v5, 23:58:56Z to 00:12:57Z, every check green: rung 1 by signal at epoch 6 (00:05:37Z), class v5 by signal at byte 6 from epoch 8 at rung 1 (00:07:46Z, 4 of 4, 9,985 bps), 12 of 12 ids equal to the CPU verifier's, the stale node 95 of 95 refused, the restart step across the boundary on a kept datadir resynced in 36.2 s with the catch-up done after 11 s (4 IsInIBD refusals of its own miner during it, 0 of its blocks accepted), four sinks equal at 661, 0 PoW rejections on the honest nodes; record on v5-fasttime 0a09eb78, docs/design/class-v5-harness/fasttime/cross-0324-dfbd1e10.json. Every gate on the pin is green; the move waits on the pairs on the dl host, the last FETCHED plus ten, and the F9/F1 interim read. THE RE-EXPORT READ (the v5 lane, box 1 with the merged crate ac285733): the three pinned packs' only difference was program.json's program_id_derivation text (generator 4 now "|| 'sub/' || sub_version_le16", generator 5 the class recipe "igneum-program-rw/ ..."); ids, kernel texts, leaves.bin, vectors and the fingerprint 82b19cbde8557ea5 byte-identical; the pinned packs carry the merged text; the full gate and the full igneum-pow suite with the packs test and spec_readback running on that tree, the push and commit string about 01:45 UK; main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2 at class-v5 b5d6368d (nothing with eight of eight reached the mirror). AP-F8-1's two eight-of-eight lines on master move to nine in this record commit. THE MOVE'S SOURCE (the shipper's ruling at 01:05 BST, corrected to this record at 01:1x): the 02:00 BST move does not wait on the build-server lane's pairs; that lane is dark (nothing published since 23:05 BST, nothing answered since 00:17), so the fleet moves EVERY Devnet 3 node from the node lane's dfbd1e10 pair at /srv/artefacts/0324-dfbd1e10/node-lane on build-1 (igneumd 4870ccf2, igneum-miner aa8c2978, the pair every gate ran on, native glibc 2.39 on every fleet box), the way dn3-g1 and g2 moved at 22:30; the fleet's puller fetches from build-1, not the dl host. The move waits on the fleet publishing the dfbd1e10 move file and naming the minute (asked 01:05) and the F9/F1 interim at 01:55. The hive and the Windows pairs are the dark lane's loss for tonight unless main gives the shipper the word to build them (asked 01:06); the Mac entry publishes at the minute regardless; if the fleet has not published the move file by 01:40 BST, main and the coordinator hear it with the clock. THE PC 1 LOCK VOID, THE V5 AMD BENCH PUBLISHED (the hash lane, 01:1x BST): the shipper's 0.3.24 host job was never published to the jobs file, so the slot was void (the shipper's "slot void" at 01:05 BST); the v5 kit fetch landed on both PCs at 00:12:06Z (919,273 bytes, sha256 ok); run-ca3-pc1-v5-amd-bench-20261007 published 00:14:19Z (the 9070 XT by name, about 3 minutes, lock-free), its start line printing app_version, so the 0.3.20 or 0.3.23 reading of PC 1's app comes with the fingerprint; PC 2's Arc job needs only the shipper's "PC 2 clear". PC 1's app had NOT taken the 0.3.23 kit as of the last reads (every job log through 21:46Z app_version 0.3.20; the install folder's exe igneum-app 0.3.20, mtime 12:24:42Z, sha256 0443ae17...). The update-return lane (a22d765a2e0355a9f) last spoke at 23:0x BST: the helper workaround for 0.3.20 scripts (truncate cmd.txt, restart the task, wait for helper.alive, then write; or four leading " dev " padding lines), the locked jobs held as they are, power-helper-24 b9a72b9b merged into release-0.3.24 (daa7427b: a silent change becomes a logged line, the helper writes its exit reason), install-close-23 4ad6c199 for 0.3.23's take 3, the re-probe job when PC 1's app has taken the 0.3.23 kit; nothing since. THE SPEC-TEXT READ-BACK PASS (adv-accept-3 Q4c, 01:11 to 01:15 BST on build-2 at 16 cores; report section 6.5 on build/adv-accept-3, log 983-textderive-8b834634.tsv, pushed): the spec text at master 8b834634, implemented fresh without the crate's generator or rule, reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences (every instruction, the chosen attempt, the id, the rejection sequence); the 264-of-400 divergence against the text at 017e7037 is closed, so AP-F8-5 reads fixed on a measurement. The one remaining gap: 1.4.6.4 names dataset_elem "of verify.rs" without its six operations, so parts (c), (c') and (c'') still take that function from the crate; the audit lane's one-sentence closed form (asked 01:1x) closes it, and the read-back re-runs on the new text. THE AMD CLASS V5 FINGERPRINT (PC 1's RX 9070 XT, gfx1201, beside the miners, lock-free): 82b19cbde8557ea5 at 01:16:14 BST, equal to the kit e6c088bb's on Metal, Apple OpenCL and CUDA, self-test PASS, the v4-genesis control 892b6d55a7ddcfcb PASS; the 0.3.24 kit stands on four platforms; Intel waits on PC 2 (held until main's word, since the 0.3.23 take 3 never ran there); PC 1's queue continues with the CA4 unlocked rows. The kits row reads: Metal, Apple OpenCL, CUDA, AMD equal; Intel not measured tonight. THE UPDATE-RETURN LANE'S THREE READINGS (01:17 BST, from the live manifest and the intake): (1) 0.3.23 take 3 (install-close-23 4ad6c199) never reported; the live manifest igneum-app-latest.json reads 0.3.23 published 20:37:44Z with platforms = {mac} only, NO Windows entry, so neither PC has anything to take through its update path; PC 2's app run is still take 1's relaunch from 21:08:43Z (997 uploads, last 00:16Z); (2) PC 1 will not take 0.3.23 unattended tonight for want of a Windows entry; its run win-ae432dc7-20261007-160110 (0.3.20) never restarted (2,376 uploads, last 00:16Z), mining 18.96 MH/s on the 9070 XT; when a Windows entry is published the 0.3.20 engine's OTA takes it with no hand; the 21:41:32Z helper.ps1 write was not the prompt path (0.3.20 writes that file unconditionally), so no screen is owed in the morning for it; (3) the re-probe job (relay/playbooks/pc1-helper-reprobe.ps1 on power-helper-24 69f3c733) waits only on PC 1's exe becoming 0.3.21 or later; the 0.3.20 workaround is cleared to run tonight as a lock-free job so the locked grids go ahead: per grid job, before the first command, empty sweep\cmd.txt, Stop-ScheduledTask and Start-ScheduledTask 'Igneum Power Helper', wait until helper.alive is within 4 s, then write the lines with climbing sequences (in 0.3.20 the skip is the line count at the helper's start, fixed for its life); the helper idle-exits 20 minutes after its last command and the next start must begin over an empty file again; never pad after a command. The coordinator's order to the hash lane on it: the locked grids proceed in the earlier order (the 5080 full grid, the third 5090 pass to the driver's floor, SM-sparse, the two Ember tunes, the hot-table ldcs rows), each with its restore step, under the no-prompt rule; a Start-ScheduledTask that reads the 0x800710E0 refusal again stops the job and reports, nothing escalates. THE LOCKED GRIDS UNDER THE WORKAROUND (the hash lane, 01:2x BST; commit 24f9858e on the mirror): the three lock scripts carry the cleared sequence (empty sweep\cmd.txt, Stop- then Start-ScheduledTask 'Igneum Power Helper', helper.alive within 4 s with a 60 s cap, then dev + command with climbing sequences; repeated before any write when helper.alive is older than 10 s; a refused start 0x800710E0 or no heartbeat stops the lock path with the text on RESULT lines, nothing escalates; no padding; each grid job ends with rgc through the same sequence and the applications clock read back). PC 1's app_version on the v5 bench's start line: 0.3.20 (no Windows 0.3.23 published, nothing to take). The CA4 SM-sparse job run-ca4-pc1-ca4sparse-5090-20261007 runs since 00:20:40Z on the earlier padded script (unlocked rows first, then its 1,300 knee attempt; about 25 to 50 minutes); then in order on the shipper's acks: the 5080 full grid as run-ca3-pc1-v4-eff-5080-20261007-d, the third 5090 pass (1,100 MHz down), the microbench and the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the 5080 grid's knee and best points to the site audit lane for row 17 as read. THE ATTEMPTS CENSUS COMPLETE (adv-accept row 90, 01:34 BST, 20,000 seeds, closing the partial named at 00:00): 42,711 rejected candidates; (a') 83.5 percent, (a) 11.6, (b) 3.0, (c'') 1.1 (459 candidates), constant bit 0.4, saturated 0.3, (c') 0.05, distinct 0.04, lane-constant and bias 0; per-candidate rejection 0.681, flat across attempts 0 to 3 (the halves agree to a tenth of a percent); accepted-attempt mean 2.136, max 28; 0 exhaustions; P(256 consecutive rejections) 2e-43 per seed. The number for spec 1.4.6: under sub-version 3 the per-candidate rejection is 68.1 percent and the expected attempt 2.1. adv-accept's shards run on in the pool's gaps under the mechanical yield; the box-hours cross 8 later tonight. CLASS-V5 LANDED ON BOTH MIRRORS (the v5 lane, 091a0758 at 01:40 UK): the full pre-push gate GREEN at 71 checks (stamp on 48d38493, the last code change); the igneum-pow suite on box 2 (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs byte-identical to the merged crate's export, so e5a4ac5978462156, 7c54302b487340a1, a217c7f698880830 and 82b19cbde8557ea5 hold under master's recipe form, recheck 2, scratch 7, spec_readback 2); spec-constants 28 rows agreeing; identity grep 0 hits. Carried since 61588347: main's sentence at nine of nine on AP-F8-6, section 14 and spec 1.4.7.2; the 61-row era reading and the class v5 attempts census on the spec, the page and the ledger; AP-F8-3; spec 1.4.7 and 1.8.6 with the constants and id tables; the AMD fingerprint row; the kits branch and master merged; two corrections the proofs found: master's program_id_derivation text lacked the class v5 rung-0 arm (the v5 packs' text named the class recipe while the id was the plain form; the arm added to the TEXT, re-exported, ids unchanged; a post-freeze change on the class-v5 line, so 0.3.25's pairing, never 1c420786's), and the public-export scrub (the founder's name six times on the page, the zone name in three files; gone). Incoming to the page: the Arc fingerprint, F9 and F1. THE MOVE FILE NOT PUBLISHED (the shipper, 01:41 BST): build-1's /fleet/move.json still names commit 2720d8d2 with the 22:30 BST minute; no FETCHED count, no named minute; the fleet lane (ac055d60427caab99) has answered nothing since its 22:4x report (asks at 01:05, 01:16 and 01:41; its task output last written 22:21 BST, its last action a hand read of dn3-g1's proven share), the second dark lane beside the build-server lane (last written 22:36 BST). So 02:00 BST cannot hold; the 02:53 BST ceiling (DAA 32,400) stands only if a signed move file lands at once and the 34 pullers fetch inside forty minutes; main has the clock line with the two options (wake or replace the fleet lane; or a fourth re-cut from a morning minute, the Mac entry standing down with it). The publish record's shape stands: the Mac entry at the minute (staged, DMG 1aa301cc, both folders, armed); the hive and the Windows pairs on main's word; the pairing 1c420786, 091a0758 0.3.25's. Every other gate on dfbd1e10 green and recorded. THE RUNG-0 ARM CONFIRMED (the v5 lane, 01:4x UK): 987e90e8 touches only Program::program_id_derivation, the text in program.json; Program::program_id untouched (the v5 rung-0 plain-form branch since the freeze); the crate at 091a0758 and 1b5684ec (master 35602b30 merged, pushed 01:41 UK) derives every pinned id byte for byte (packs 20 on box 2 comparing all three pinned packs' files including program_id and leaves.bin; spec_readback 2); the shipper told 091a0758 and 1b5684ec are 0.3.25's pairing, 0.3.24 on 1c420786. THE SM-SPARSE JOB (run-ca4-pc1-ca4sparse-5090-20261007, exit 0 at 00:41:53Z, 1,171 s, the 5090 alone, every fingerprint matched, the Power Helper answering every command on the padded write, the card left unlocked at 2,855 MHz): the SM-sparse reading does NOT exist; the research lane's worker ran its base kernel on every variant row (its race line "race 0 ms variant base" on all 48 rows, no NVRTC compile text), so --bench never honoured --variant sp-w32; the sparse rows equal base in rate and drift in watts with the card's heat only; the rerun waits on the research lane's exe honouring the flag. What stands: a repeat of the efficiency pass at two states, 32 s rows, the card alone: v4 unlocked 137.07 MH/s at 465.5 W (0.294 MH/W), at 1,300 MHz 134.26 at 309.9 W (0.433; 155.6 W back for 2.05 percent of rate); v3 unlocked 136.71 at 331.6 W (0.412), at 1,300 134.03 at 219.4 W (0.611; 112.2 W back for 1.97 percent); the v4 premium 133.9 W unlocked, 90.5 W at the knee; the three power fields agree within 0.2 W on every row (power.draw = instant = average on driver 617.14), which settles the field question on PC 1's side and leaves the 5080's 110 W gap to the fleet's rented card's sampler. Next on the shipper's ack: the 5080 full grid (-d) through the cleared helper sequence, the third 5090 pass, the microbench, the seven packs, the two tunes, the hot-table ldcs rows (kit and job at 292fcc75). THE --variant FAULT FIXED (the research lane, counter-asic-4, UTC clocks on 8 October): 00:44 the fix (a --bench with --variant runs the pinned race and installs the named kernel; the RESULT line carries variant=, sparse_blocks=, block_warps=; a served kernel other than the requested one prints variant_not_installed); 00:46 the known-failed test on build-1 against the real class v4 pack, no card (base: race off, 524,288 blocks of 32, "variant base"; sp43-w32: race on, 43 sparse blocks of 32 warps, the rewritten kernel with the nonces argument and the unit function, 43 blocks of 1,024; PASS; before the fix both read the base shape); 00:46 the Windows exe igneum-worker-cuda-ca4sparse3.exe sha256 0ba97edcd5c46a302a7ff5ddd1bbb1e493ca15f64d0757820ed645972df3bb56, mingw exit 0; the commit after 2d0013d1; the hash lane has the sha, the test's lines and the rerun's job shape (the same 48 rows, the race line per row); the op-mix re-weight stays behind the SM-sparse reading, the served 3.4x standing; the clean efficiency repeat in the file's 20.3a (6.6 pJ per counted op). THE SPEC'S LAST CRATE-DEPENDENT SENTENCE CLOSED (the site audit lane, master 56eebc0d at 01:49 BST, gate GREEN on c2c92eab, 71 checks; spec_readback now 3 tests): 1.4.6.4 states dataset_elem in full (the eight operations, the three constants, 32-bit wrapping) with two pinned vectors (dataset_elem(0x00000fed, 0x9E3779B9, 0x7F4A7C15) = 0x5c7dabd2; dataset_elem(0x0fffffff, 0, 0) = 0x7662c1ec) that spec_readback.rs reads from the text and checks against the crate, so part (c) computes from the text alone (34845c47); 1.4.6.5 names the class v2 figures as class v2's and carries the shipped rule's own census sentence (20,000 seeds, 68.1 percent rejected per candidate, the per-part shares, mean attempt 2.1, max 28, 0 exhaustions, 2e-43). The text-derived re-run on this text is the proof it is sufficient end to end (asked of adv-accept-3). THE MOVE FILE STAGED (the shipper, 01:5x BST): id mdfbd-1, commit dfbd1e10, want_digest b1ba7822, both pair slots on build-1's served tarball dfbd1e10-node-lane.tgz (e59ed0e6), at_epoch 0, signed with the fleet key on the Mac and verified against the fleet's public key in the puller's namespace; the read-back on placing it: the served file's id by curl and the first FETCHED on the relay intake; the 34 pullers fetch inside their one-minute timers (27 MB from build-1), the last FETCHED about five minutes after the file, the earliest minute ten after that. THE REAL LATEST-PUBLISH CLOCK: the file alone halts every miner on the restart, because each box's pack gate PAIR_MINER_SHA16 lacks aa8c2978 and the puller does not carry the file's miner sha into the restart environment; so route (A) also needs one ssh line on each of the 34 boxes before the minute with the fleet's tooling (the fleet lane's, or the shipper's on main's word). Absent main's word by 02:15 BST the shipper stands the Mac entry down under the ceiling rule (no app alone on b1ba7822) and 0.3.24 becomes a morning minute with a fourth re-cut. ROUTE (A) STAGED TO ONE COMMAND (the shipper, 01:5x BST): the gate script r0324/move/pair-gate-aa8c2978.py in its scratch (dry run by default, apply on the literal argument, the fleet's own Box helper and label list, nothing restarted); the dry run read 33 of 35 boxes, every one carrying the old gate list with fb147dd1 last and aa8c2978 absent, no env-last override; unreachable dn3-relay and p2-4090-1b (dead Vast proxies; they fall off at the move and rejoin by the pull); the apply about 90 s for the 33 with each gate read back and counted. THE F9/F1 INTERIM (the attack-pass lane, read at 01:5x BST): 71,292 seeds, 0 exhausted, 0 panics, max attempt 30; F1 0 failures at 2 h 23 min; the move's gate reads clear. On main's (A): apply 02:00, the file placed 02:02, the last FETCHED about 02:05, the minute 02:15 BST; main has the clock. Nothing applies before the word. THE SPEC TEXT SUFFICIENT END TO END (adv-accept-3 Q4d, 01:52 to 01:57 BST on build-2 at 16 cores; report section 6.6 on build/adv-accept-3, log 984-textderive-56eebc0d.tsv, every row equal to its Q4c row): the spec text at master 56eebc0d, implemented with nothing from the crate (text.rs: 0 igneum_pow imports; dataset_elem from 1.4.6.4, its two pinned vectors checked at start), reproduces the same 400 class v4 epoch programs as the code with 0 of 400 differences on every field; no sentence of the generator or acceptance sections needs the crate; the documentary finding (AP-F8-4, AP-F8-5) closed in full on two measurements; the lane at its end, 3.35 box-hours in all. F9 AND F1 AT 00:59Z (class v5 at 1c420786, pairing e5a4ac5978462156, build-1): F9 73,691 of 100,000 chain-shaped seeds written, 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; the attempt histogram 23,119 / 15,981 / 10,805 / 7,547 / 5,181 / 3,415 / 2,439 / 1,653 / 1,119 / 744 / 529 / 389 / 258 / 152 / 113 / 72 / 54 / 40 / 31 / 14 / 15 / 5 / 2 / 6 / 2 / 4 / 1 at 26 / 1 at 30, r about 0.69; the 10^5 about 01:30Z (02:30 BST). F1: the 10^5 redundancy census at 2 h 28 min under its lease with no end marker (18 minutes on an idle box; under tonight's load no minute named); its panic path live and empty, 0 failures the honest reading. Both land as record lines, then the board's close per item on sub-version 3 and class v5. AP-F8-5 ON TWO MEASUREMENTS (the site audit lane, commit 116e6055, master 5c77a7ac at 02:05 BST, gate GREEN 71 checks): the row carries Q4c (8b834634, 0 of 400 with one crate function, section 6.5, log 983) and Q4d (56eebc0d, 0 of 400 with no crate import, section 6.6, log 984); the public ledger and the ledger page regenerated; nothing open in the spec or the ledger on the audit lane's side. THE 5080 FULL GRID (run-ca3-pc1-v4-eff-5080-20261007-d, exit 0 at 01:54:00Z, 4,118 s; PC 1's dock card alone, driver 617.14, app 0.3.20, mem 14,801 MHz throughout; every lock through the cleared helper sequence, every command answered first time, every fingerprint matched, clocks reset and read back): the knee as a reading: the rate holds within 0.3 percent of unlocked down to 1,000 MHz on both classes (v4 71.19 of 71.41 MH/s; v3 71.11 of 71.28) and falls 5.2 percent at 900 MHz on v4 (67.66), where the 75-minute budget ended the grid (v3's 900 and below not taken; the drift check skipped); so the 5080's knee sits between 1,000 and 900 MHz, a third of its 2,963 MHz boost, lower than the 5090's 1,300 (84 SMs at 2,960 MHz have more compute headroom per unit of its 960 GB/s than the 5090's 170 SMs per unit of 1,792 GB/s; the memory wait hides the shadow down to a lower clock). Best MH per watt within the 1 percent rate tolerance: v4 at 1,100 MHz, 71.20 MH/s at 146.6 W (0.486 MH/W; 106.5 W recovered for 0.29 percent of rate); v3 at 1,000 MHz, 71.11 at 103.7 W (0.686; 66.0 W for 0.25 percent). The v4 premium 83.4 W unlocked (253.1 against 169.7), 41 W at the best points (146.6 against 105.6 at 1,100). Per tier: a 5080 owner on class v4 locked near 1,100 MHz draws 147 W instead of 253 for 0.3 percent less rate (MH/W up 72 percent) and the shadow's residual cost is 41 W. Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 71.41/253.1/0.282 ; 71.28/169.7/0.420 (sm 2,963/2,977); 2850 71.41/229.5/0.311 ; 71.29/155.3/0.459; 2700 71.41/209.6/0.341 ; 71.29/149.2/0.478; 2550 71.41/193.0/0.370 ; 71.29/134.4/0.531; 2400 71.41/176.0/0.406 ; 71.29/128.7/0.554; 2250 71.41/165.6/0.431 ; 71.28/117.3/0.608; 2100 71.38/157.7/0.453 ; 71.28/112.3/0.635; 1950 71.37/154.0/0.463 ; 71.26/111.6/0.639; 1800 71.35/150.3/0.475 ; 71.24/113.4/0.628; 1650 71.33/151.4/0.471 ; 71.22/109.7/0.649; 1500 71.30/149.2/0.478 ; 71.19/110.6/0.644; 1400 71.27/149.6/0.476 ; 71.16/107.0/0.665; 1300 71.24/147.9/0.482 ; 71.14/107.7/0.661; 1200 71.20/149.4/0.477 ; 71.12/104.4/0.681; 1100 71.20/146.6/0.486 ; 71.11/105.6/0.673; 1000 71.19/149.0/0.478 ; 71.11/103.7/0.686; 900 67.66/137.8/0.491 ; not taken. Throttle reason 0x400 (the power governor) on every row, never the clock lock, so the draw floor of about 147 W (v4) and 104 W (v3) from 1,500 MHz down is the memory system plus idle, not the SMs: the clock lever is spent by 1,500 MHz on this card. The three power fields agree within 0.2 W on every row. The site audit lane has the knee and best points for row 17; the bench table's 5080 row takes "71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", class v4 cost "+83 W unlocked, +41 W at the best points", hive core 1100 (the mem clock unchanged) once the fleet's rented-5080 sampler question is closed. Next on the shipper's ack: the third 5090 pass, the SM-sparse rerun on the fixed exe, the microbench, the packs, the two tunes, the hot table. THE NIGHT'S MOVE OUTCOME (the shipper, 02:58 BST): main's word on (A), (A') or (B) did not come (asked 01:41, 01:50, 01:53, 01:56 by the shipper and 01:42, 01:52, 01:5x, 02:00 by the coordinator); the gate script not applied (the dry run's 33 of 35 the only read); the move file not placed (build-1's /fleet/move.json serves m2720-1, 2720d8d2, the 22:30 minute, by curl at 02:56); no move minute; the stand-down under the ceiling rule holds from 02:15 (the shipper's stand-down line at 02:15 was not sent, its miss, the state unchanged); the Mac entry standing, not published (staged on DMG 1aa301cc in both folders, the live manifest at 0.3.23). A FINDING: build-1's Devnet 3 seed (the process on 26631 with JSON RPC 27632, the node lane's DAA reader) is DOWN (no such process; node1-dn3 26671 and the observer 26651 run on 2720d8d2; the node lane's 0.3.24 reader on 28690 runs but answers no DAA by the envelope tried), so the node lane's DAA reads since 25,169 at 00:52:38 BST may have stopped with it; at 1.0 DAA/s the DAA passed 32,400 at about 02:53 BST, the 39,600 floor is lost, and the fourth re-cut is from a morning minute main names (before 12:50 BST, or the three heights move with the floor). Every Devnet 3 node is on the 0.3.23 pin 2720d8d2, digest ba75bf6f (the 22:30 move; dn3-j1 behind its proxy unverified since); nothing of 0.3.24 is on any box or in any manifest. The night's 0.3.24: every gate green on dfbd1e10, the kit on four platforms, the move unmade for want of one word and two dark lanes. THE ATTACK-PASS BOARD'S CLOSE (lane (d), 01:58Z on 8 October; record docs/analysis/attack-pass-2026-10.md on the mirror's attack-pass; box-hours approximate: build-2 about 7 h, build-1 about 9 h plus about 6 h of F6 batches and F2 solvers earlier in the day). F9 so far: 89,301 of 100,000 chain-shaped seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 20: 20, 21: 6, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1; r about 0.69), three chunks on their cores to about 02:20Z; F1 the 10^5 redundancy census at 3 h 22 min on 17 threads, healthy, no end marker, 0 failures on its live panic path. The board: F1 shadow redundancy PASS on sub-version 3 (max 5.078 percent at honest-compiler parity; AP-F1-1 on the v5 list at 3.0 percent), running on v5; F2 mixer round margin PASS effort-bounded (no trail under weight 20 to 24 at 2 applications, 29 to 35 at 3, 39 to 47 at 4), not re-run on v5 (the mixer unchanged); F3 chained cache j+1 PASS, not re-run; F4 weak-day census PASS on v4 on the DSP-bound metric with AP-F4-1 reconciled with adv-mixer-2 (median 226, 15 days a century, worst 2050-04-28 at 1.113x), on v5 PASS at 8ca66afa (0 of 2^24 days over 1.1x on both metrics, AP-F4-1 FIXED-AND-PASSED); F5 chip-model sweep FIXED-AND-PASSED (the F2 hour skipped by decision), not re-run; F6 verifier worst case PASS (worst of 10^5 at 8.708 ms half-core; O-1.14 closed, i7-9700K 6.334 ms), not re-run; F7 era draw PASS on all three (0 of 6 re-rolls), not re-run; F8 uniformity FIXED-AND-PASSED on sub-version 3 (60 of 64 under 1.2x; AP-F8-1, 2, 3 closed), PASS on v5 (61 of 64, worst 1.50x, the residue p4, p8, p10; p34 under); F9 edges, hot set, grinding PASS on sub-version 3 (34 of 105,064 edges bounded; grinding +0.004 percent), the exhaustion count running on v5; F10 ladder signal PASS, not re-run (node rule). Findings of the pass, all in-house: AP-F1-1, AP-F4-1, AP-F5-1 (the X9), AP-F8-1, AP-F8-2, AP-F8-3; two operating hazards fixed (AP-H1 the box clean, AP-H2 the shared binary path). The open tail (p4, p8, p10, and p34 on sub-version 3) is named in the public report; no outside party holds it (the attack-pass lane's close wrote "disclosed to the firms", stale wording from before the in-house ruling; its record file is to say "named in the public report"). THE SEED'S DEATH AND THE DAA NOW (the node lane, 03:0x BST): build-1's Devnet 3 seed log /home/build/dn3seed.log ends at 01:09:05Z at DAA 29,732 mid-stream with no stop, shutdown or panic line, so it was killed abruptly (it ran under nohup from a shell, not a unit; no journal names the killer; the OOM record needs sudo the lane lacks); its datadir /home/build/dn3seed/igneum-devnet-3/datadir is intact (13 GB) and it stays down until the shipper says; the lane's reads 25,169 at 23:52:38Z and 28,906 at 00:55:09Z came from it while it lived. The DAA now from node1-dn3 on 28670: 32,659 at 01:57:50Z (the observer 32,660), both on 2720d8d2; the chain passed 32,400 at about 01:53Z, 39,600 lost. The fourth cut in one line: the script on release-0.3.24-node reads the DAA from 28670, sets the floor to the morning minute's publish DAA plus 7,200 rounded up to the next 3,600, commits, pushes both mirrors and dispatches the gate set (about 20 minutes to the pin line, then the fast-time pair about 14); the latest minute before the three heights move with the floor is about 11:50Z (12:50 BST), where the floor reaches 79,200; nothing is cut until main names the minute. A morning item for the box owner: a process on build-1 was killed at 01:09:05Z without a log line while the box carried a load of 400 to 600; the killer (OOM or a sweep's cleanup) is to be read from the journal with sudo before anything long-lived runs there again under nohup. THE BENCH LOG ENTRY (the hash lane): docs/bench-log.md "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080" (both cards' full tables, the knee per card, the best MH per watt points, the premiums at the lock, the lever's limits, the job ids and clocks, the rented-5080 watts note) on the mirror's master as merge 773b93a8 at 02:04:47Z (commit 11c698ad); the audit lane writes row 17's sentence from it. PC 1: the third 5090 pass run-ca3-pc1-v4-eff-5090-floor2-20261007 (1,100 MHz down to 300) since 01:57:03Z, about 28 minutes; then the SM-sparse rerun. ROW 17 AND THE 5080 BENCH ROW (the site audit lane, master 2c5c7f52 at 03:19 BST, gate GREEN on 6eb6fd9b, 71 checks): docs/evidence.md row 17 carries both cards' efficiency passes from the bench-log entry (the 5090's knee, best points and premium; the 5080's 71.41 MH/s at 253.1 W unlocked, 71.20 at 146.6 W at 1,100 MHz, v3 at 1,000 MHz 103.7 W, the premium 83.4 W to 41 W, the knee between 1,000 and 900 MHz, the per-tier reading, the Ember Tune lever), a what-moved table for 8 October, /evidence rebuilt (865a0a5e); site/miner-bench.json's RTX 5080 row states the team's pass as the card's figure ("71.4 stock (71.2 tuned)", "146.6 tuned (253 stock)", "+83 W unlocked, +41 W at the best points", hive core 1100 with the memory stock, driver 617.14, the bench-log entry as the source) and keeps the rented-fleet sampler reading with its 110 W gap as the open question; /miners rebuilt at 35 rows (6eb6fd9b); 0 identity hits; nothing deployed, the deploy the morning hand-off. The design pass on ca3-coord (015cc839) now sits behind this master and rebases onto it before its own landing on main's word. THE DESIGN PASS REBASED (the coordinator, 03:2x BST): ca3-coord rebased onto master 2c5c7f52 as the three site commits only (f5b7140c the design pass, 8cc4cbc6 the phone grid, 9ad3fdc9 the six-column row; the two commits already landed through the record branch skipped), site/miners.html rebuilt at each from the merged miner-bench.json so the page carries the 5080's new row ("71.4 stock (71.2 tuned)") under the design; the diff against master is build.mjs and miners.html only; pushed to the mirror (pre-push GREEN); it lands on main's word after the captures, one gate run. ADV-MIXER-3's LINE (read from its report at tip e02297ae, 03:18 BST): queue 17 finished on build-1 at 01:3x BST; Q2 single-bit avalanche at 2^27, k = 2 and 3 on day 20729: 0 holes, 0 cells beyond 6 sigma at band 0.00026, PASS (the k = 1 finding stands as the single-application diffusion); Q2b t-bit avalanche on day 20733 at 2^28: 0 cells beyond 6 sigma at band 0.00018, PASS (k = 2, 3, 4 on 20729 at 2^28 the same); Q3 at k = 8 NOT run (killed at 20:20 BST under the lease rule, not re-queued; k = 2 to 7 clean with 0 deterministic bits on both days), named partial; Q6 the day-20733 SAT ladder: k = 2 and 3 TIMEOUT at the one-hour cap, k = 4 on one build-2 core since 03:05 BST, its cap about 04:05; one pre-emption in its ledger (23:58 BST, 21 minutes of a 2^27 row lost, re-queued); box-hours about 3.0 wall-hours of sweep (build-1 1.9, build-2 1.1) plus about 4 single-core CaDiCaL hours, about 7 with the 20733 ladder. The pass's close with the per-lane table and totals at about 04:05 BST; section 13 on crypto-engage (docs only) merging the current master and going through the gate to the mirror's master so the record cites a master commit. Box 2 at 03:20: adv-accept 87 cores in four shards with three waiting, adv-mixer-3 one core; build-1 load 34, no adv lease. THE DESIGN PASS'S OVERLAP ON THE BOX (the CI steward, 03:33 BST): the 1440 and 390 dark captures of /miners from ca3-coord 9ad3fdc9 taken on build-2 under lease pool 4 (Playwright chromium 1194, the recorded feed; /srv/artefacts/captures/ca3-coord-9ad3fdc9/miners-1440-dark.png 1440 x 4280 and miners-390-dark.png 390 x 9779); the overlap sweep on the same checkout, 390 to 1600 px, light and dark: RED, 3 findings on the change itself: at 1280 px dark and 1600 px light and dark the date span in the lead cell's class v4 line is COVERED by the rate cell (4 of 5 sample points under td.big); 390 to 1024 pass. Cause: the branch's last gate ran on the Mac, which has no browser, so the sweep skipped and read GREEN; on the page the row rule's white-space:nowrap outranked the lead cell's normal by specificity, so the class v4 line ran under the rate cell from 1280 px up. FIXED at ca3-coord 2ca45001 (the lead cell's rule at the row rule's specificity, max-width 360 px, the class v4 line wrapping with overflow-wrap), rebuilt, pushed; the sweep and the captures re-run on the box before main's word. THE IN-HOUSE PASS'S PATH TO MASTER (the crypto lane, 03:2x BST): adv-accept's box-hours crossed 8 before 02:00 BST and sit near 10 (87 cores in four shards; it sweeps on under the mechanical yield, its reading unchanged); crypto-engage merged master 56eebc0d at 342b6730 (one conflict in funding.md, the pre-public scrub against the rewrite, resolved to the in-house pass with the scrub applied; the founder never named in in-house-pass.md or funding.md), the full gate running, merge-to-master on GREEN; section 13.3: master's igneum-pow moved after the freeze in four files (src/emit.rs and src/generator.rs, the derivation string and its recipe helpers, ids unchanged; tests/derivation.rs and tests/spec_readback.rs), none the hash, so the object the pass bounded is unchanged in every operation the hash performs. THE PASS IN ONE LINE (the crypto lane, 03:2x BST): eight of nine lanes closed, adv-mixer-3 on one SAT timeout (about 04:05 BST), adv-accept sweeping to its 16 box-hour line (9.2 now, the reading saturated at the 1.002x class), adv-cache-2 on one line shard; no break of class v4 sub-version 3; the acceptance's hot-set class closed by the class v5 floor (9 of 9) and its diffuse era-stride class routed to the next class; the weak-day FPGA tail reconciled and closed by a measured redraw rule; the attempts census complete; the spec text proven sufficient by two read-backs; one pod at USD 0.33 in the whole pass, none originated by the lane. THE THIRD 5090 PASS BELOW THE KNEE (run-ca3-pc1-v4-eff-5090-floor2-20261007, running at 02:34Z on its 500 MHz step; the steps lengthen as the rate falls since the batch count was sized from the unlocked rate, about 155 s at 500 against 60 at 1,100; the helper answering every command on the cleared sequence, every fingerprint matched, the 5090 alone). Rows (lock: v4 MH/s / W / MH/W ; v3): unlocked 137.09/456.7/0.300 ; 136.79/320.0/0.428 (sm 2,858/2,862); 1100 120.98/275.9/0.439 ; 117.32/198.6/0.591; 1000 110.03/254.9/0.432 ; 106.73/180.2/0.592; 900 97.43/232.7/0.419 ; 94.33/174.5/0.541; 800 86.00/216.3/0.398 ; 83.41/166.6/0.501; 700 75.98/202.7/0.375 ; 73.58/156.4/0.470; 600 65.30/178.2/0.366 ; 63.25/153.3/0.413; 500 53.03/166.5/0.319 ; v3 running. Reading: below the knee the rate falls about 10 percent per 100 MHz on both classes (compute-bound: the shadow and the base program no longer fit the memory wait) and MH per watt falls with it from 1,100 down, so the best point stays where the second pass put it (v4 at 1,200, v3 at 1,300); the driver took every lock down to 500 (the SM clock within 10 MHz), so the floor is below 500 MHz and is not where the optimum lives; the v4 premium below the knee 77 W at 1,100, 75 at 1,000, 58 at 900, 50 at 800, 46 at 700, 25 at 600 (the ALU work shrinking with the clock as the rate does). The exit line, the 400 and 300 rows, the drift check and the restore at its close; then the SM-sparse rerun on the fixed exe (each sparse row reading served= and sparse_blocks=, marked variant_row=FAILED if served as base). F9 AND F1 AT 02:34Z (class v5 at 1c420786, build-1): F9 98,945 of 100,000 seeds, 0 exhausted, 0 panics, 0 past attempt 31, max 30 (the tail 18: 39, 19: 19, 20: 24, 21: 8, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1); the last three chunks within minutes of their ends; F1 at 4 h 02 min under its lease, no end marker, 0 on its panic path. The pass record's wording fixed on the mirror's attack-pass at 9474cea8 ("named in the public report"; no "firm", "firms", "escrow", "prize", "paid review" or "Lot" line in the pass record or the ten row records; identity grep 0 hits); the section's merge to master after the two record lines, through the full gate in a detached worktree. THE SECOND SWEEP ON THE DESIGN PASS (the CI steward on 2ca45001, 03:38 BST): the desktop widths pass; RED at 390 px dark only, three findings on the lead cell (the card name and the class v4 line covered by the rate cell), the cause the new 360 px max-width on the phone grid; FIXED at ca3-coord 5158276c (the lead-cell width rule scoped to widths above 1,100 px, the phone grid's lead cell with no max-width), rebuilt, pushed; the sweep and captures re-run on it. F9 PASS ON CLASS V5 (the attack-pass lane, class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, the last chunk written 02:34:54Z): 100,000 of 100,000 seeds drawn through the chain path (era-composed class), 0 exhausted, 0 panics, 0 past attempt 31, max attempt 30; histogram 0: 31,454, 1: 21,460, 2: 14,660, 3: 10,263, 4: 7,047, 5: 4,701, 6: 3,297, 7: 2,256, 8: 1,532, 9: 1,027, 10: 702, 11: 509, 12: 365, 13: 216, 14: 153, 15: 103, 16: 80, 17: 56, 18: 39, 19: 20, 20: 24, 21: 9, 22: 6, 23: 9, 24: 3, 25: 4, 26: 2, 27: 1, 29: 1, 30: 1 (first-draw acceptance 0.3145; the mean attempt index 2.185, so 3.185 draws per seed on average; 4,862 seeds, 4.86 percent, at index 8 or above and 255, 0.255 percent, at 16 or above; the 256-attempt cap and the deterministic last resort never reached; the lane's first line read 1.993, a slip it corrected); the exhaustion gate holds for the 0.3.24 move; record docs/analysis/attack-pass/f9-grind.md and the lane (d) section on the mirror's attack-pass. F1 still running (4 h 05 min, 16 cores, 0 on its panic path, no end marker). F9's record on the mirror's attack-pass at 2bcb7e08 (the lane (d) row and f9-grind.md section (d); feature gate GREEN); F1 the one open item before the lane (d) merge to master. THE DESIGN PASS GREEN ON THE BOX (the CI steward on ca3-coord 5158276c, 03:4x BST; build-2 under lease pool 4): the overlap sweep 390 to 1600 px, light and dark, GREEN, 0 findings (the known-failed fixture fired first); the 390 px capture byte-identical to 9ad3fdc9's (the phone shape that passed before), the desktop widths carrying the wrap at 4,640 px tall; the four dark whole-page captures on build-1 under /srv/artefacts/captures/ca3-coord-5158276c/: miners-390-dark.png (sha256 9eec8f27..., 509,158 bytes), miners-1280-dark.png (1b6e636d..., 438,541), miners-1440-dark.png (87387e14..., 445,402), miners-1600-dark.png (d53973cd..., 448,545); the run log /srv/builds/bs-ci-steward/cap-out/run-5158276c.log on build-2. The branch's gate record: a full gate on the Mac skips the sweep (no browser), so the box line is the sweep's verdict for 5158276c; the branch waits on main's word on the look and lands in one gate run. THE IN-HOUSE PASS'S RECORD ON MASTER (the crypto lane): crypto-engage dab0c89f (gate GREEN, 71 checks) landed through merge-to-master.sh --remote build at 03:50 BST as master 00b8cd1b: docs/plans/cryptanalysis/in-house-pass.md section 13 (the roll-up, every lane's reading, the frozen-object note) and funding.md's in-house row and brief, scrubbed under founder-strings-check.sh. AN EXCEPTION OWNED (03:39 to 03:50 BST): the lane's first merge call used the tool's default path, which reads CI on GitHub with gh run list; GitHub is suspended and the rule says never poll it; the tool polled 21 times (each 403, nothing pushed, nothing read); the run's process outlived the task stop and the lane ended it by its pid at 03:50 BST, then used --remote build; the breach is the tool's default against the rule and the lane's for not passing the switch; no state moved on GitHub's side. The coordinator's order on it: merge-to-master.sh's default remote must refuse GitHub while the suspension stands (the CI steward, a gate-side fix with a known-failed self-test), so the rule does not rest on every lane remembering the switch. THE THIRD 5090 PASS CLOSED BY ITS CAP (run-ca3-pc1-v4-eff-5090-floor2-20261007, ended by the 45-minute cap at 02:42:05Z during the 300 MHz step, exit -1, its own finally block never ran; every row taken matched its fingerprint, the 5090 alone): the 500 row's v3 side 51.37 MH/s at 136.4 W (0.377); 400: v4 42.62/152.5/0.280, v3 41.32/131.3/0.315 (sm 390); 300 not taken; no unlocked-end drift check; the driver took every lock down to 400 (the SM clock within 10 MHz), so the floor is at or below 400 MHz. The reading: below 1,300 the rate falls about 10 percent per 100 MHz on both classes and MH per watt falls from 1,100 down (v4 0.439 at 1,100 to 0.280 at 400; v3 0.592 at 1,000 to 0.315), so the optimum stays at the second pass's points (v4 1,200 MHz, v3 1,300) and nothing below 1,100 is worth the knob's time; the v4 premium below the knee shrinks with the clock (77 W at 1,100, 46 at 700, 21 at 400). AN EXCEPTION OWNED: the 5090 sat at the 400 lock (390 MHz, 127 W mining) for four minutes until run-ca3-pc1-clocks-restore-20261008 (02:45:20 to 02:46:30Z, exit 0) started the helper over an empty cmd.txt and sent rgc ("All done"), the card reading 2,880 MHz after; the cause the batch count per step sized from the unlocked rate, so the low steps ran 2.5x longer than planned; the fix in the scripts: the budget check ends the grid with the restore inside the cap, and a probe dev line answered in helper.log counts as the helper up when its heartbeat file stays stale (the restore answered at once with helper.alive stale past 60 s). THE SM-SPARSE RERUN: fetch-ca4-sparse3-exe-20261008 landed 02:49:57Z (sha256 0ba97edc...), run-ca4-pc1-ca4sparse-5090-20261008 published 02:51:15Z on the hash lane's own order (the shipper's acks were for the void host slot); each sparse row reads served= and sparse_blocks= and is marked variant_row=FAILED if served as base; the close about 03:15Z (04:15 BST). THE STEP-BUDGET FIX ON MASTER (the hash lane, merge 9fd8b1d8 at 03:02:03Z on 8 October, commit 0b00c42e, the full gate GREEN): the efficiency pass keeps four minutes of its cap for the restore (every step and lock guarded by the deadline minus four minutes) and sizes each step's batch count from the last rate read for the pack, so a 60 s step stays 60 s as the rate falls; a probe dev line answered in helper.log counts as the helper up when the heartbeat file stays stale (all four lock scripts); the gate check tools/ci/pc1-step-budget-check.sh with the known-failed case first (under the old rule a lengthening grid ends on the cap with no restore; under the new it ends with the restore at 1,500 s of 2,700), wired into pre-push.sh and checks.txt (74 checks). The 5080 Ember tune, the 9070 XT tune pass and the hot-table ldcs rows publish behind the SM-sparse rerun, the microbench and the packs. THE SM-SPARSE RERUN FAILS THE SAME WAY, NOW NAMED (run-ca4-pc1-ca4sparse-5090-20261008, the fixed exe ca4sparse3, started 02:52:48Z): every sparse row served=base sparse_blocks=0 variant_row=FAILED, the worker's own line "RESULT variant_not_installed requested=sp43-w32 served=base race=... variants 1 base only, no race (no other variant named)", no "compile:" text, so NVRTC never saw a rewritten kernel: the variant name is parsed into the request but never added to the race's variant table in this exe; the research lane's emulation test checked resolve and rewrite, not the race list the bench builds (a test of the wrong layer; the known-failed case must be the bench's own race line reading "variants 2"). The rows are base runs; no reading. The queue goes on: the microbench at the rerun's exit (about 03:15Z), the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows; the SM-sparse question's fourth row stays with the research lane, an exe whose card-free check shows "variants 2" in its race line getting the slot within the minute. CLASS-V5'S F9 ROW PUSHED (the v5 lane, class-v5 1d5e5d23 on both mirrors at 04:04 UK): the page's F9 row (100,000 seeds, 0 exhausted, max index 30, first-draw acceptance 0.3145, mean index 2.185, F9 PASS, the record file named), F1 stated as running with 0 failures (its own commit to follow), the Intel row not measured tonight; master merged twice (2c5c7f52 gated at 5f5e0a5c, full gate GREEN 71 checks at 03:45 UK; 9fd8b1d8 auto-merged and pushed on the hook's light gate, the full gate running on 1d5e5d23); the generated ledger files and the spec-constants check clean on the tree. THE THIRD --variant FIX (the research lane, 03:04Z on build-1 under lease, with the hash lane): the cause of the 02:52Z rows: the race's push looked the pinned name up in the empty order list through the variant lookup, whose on-demand sp path answers for any list, so the sparse variant was "found" and never pushed; the order is now a pure function with membership by name; --list-race prints, with no device, the race order the worker's own option handling builds and whether the rewrite applies: with the job's exact flags "variants=2 names=base,sp43-w32" and "sparse_blocks=43 block_warps=32 rewrite=applied bytes=22261 nonces_arg=1 unit_fn=1" (before the fix variants=1); exe igneum-worker-cuda-ca4sparse4.exe sha256 84846396559004a8df61881c15ecb42fa3fc1010ad99074e0c0b53e81bb1ca3b, the commit on the mirror after 7e9d52a6; the third rerun on the hash lane's queue at the next slot; the two failed runs stay the night's SM-sparse state, the op-mix re-weight held, the served 3.4x standing. THE GITHUB GUARD ON MASTER (the CI steward, tip 2f702735 at 04:05 UK; merges 53773860 and 2f702735, full gate GREEN 71 checks each): fa7e98fe adds the tracked marker tools/ci/github-suspended (suspended-since 2026-10-07T17:02:00Z, removed by main at the cut-over) and two refusals: merge-to-master.sh refuses a GitHub remote (origin by default, or any remote whose URL carries github.com) with one line naming the switch and exit 2 before any gh or git call; the pre-push hook refuses any push to a GitHub remote the same way (the hook reads the remote URL, so a bare git push origin is refused too); known-failed first in both self-tests; the live read on the Mac: merge-to-master.sh --remote origin exits 2, nothing contacted; two follow-ups (9484b988, a08423d4) fix the tool's own --self-test under the real marker. The rule no longer rests on any lane remembering the switch. F1 READ AT 03:05:58Z (the attack-pass lane; the census process itself, not the lease wrapper): state S with 17 threads, 15 cores busy over 45 s, 2 d 19 h of CPU banked over 4 h 30 min of wall, RSS 0.8 to 1.0 GB; computing, not hung. No rows can exist before the end: the harness collects every Report in memory under thread::scope and writes census.csv in one go at the end (no progress print), named as a harness gap in the record. Why fifteenfold against the v4 reference: under class v5 every candidate draw runs the (c''') distinct-index floor over 2^20 (about 1.8 core-s per candidate under the night's load, times 3.2 draws per program, about 5.8 core-s per program before the analysis), so 10^5 programs at 15 busy cores is about 10.7 h of wall, the end about 09:00Z (10:00 BST), nearer the early side as the load fell to 21. Ruling: not killed (a kill loses 4 h 30 min with nothing on disk); the lane (d) section merges to the mirror's master now with F9 and the F1 row reading "running, 03:06Z reading, projected end about 09:00Z", F1's record line in a second merge when it writes; the harness gains a progress line before its next 10^5 run. THE IN-HOUSE ADVERSARIAL PASS CLOSED (04:08 BST on 8 October; an internal adversarial pass, not an independent review; section 14 of in-house-pass.md at crypto-engage c099e818 landing on master through --remote build; every tip read from the mirror at 04:07 with igneum-pow identical to 017e7037 on all nine). Per lane (tip; box-hours; verdict; partial): adv-mixer d2ba3134, about 0.6 plus 1.8 single-core SAT hours, the algebraic structure BOUND, none; adv-mixer-2 2a632579, 0.31, BOUND for every chip, GPU and the verifier with the FPGA LUT-area FINDING (2^-10.8 of days, 15 a century, worst 2050-04-28 at 1.113x) closed by the measured redraw rule, none; adv-mixer-3 981bfff2, about 5.0 wall-hours plus 8 single-core SAT hours, Q1 BOUND (2^32 t uniform at k = 1 to 8, both days and 8 random days), Q2 and Q2b FINDING at k = 1 only and BOUND from 2 to 8 at 2^24 to 2^28, Q3 FINDING at k = 1 and BOUND 2 to 7, Q4 and Q5 BOUND from k = 1, Q6 SAT BOUND (k = 1 in 137 s, k = 2 to 4 timeout), the round margin 70 of 72 per item, partial Q3 at k = 8 not run, multi-bit masks and a MILP bound not attempted, GPU blocked; adv-cache 555c3e42, 0.55, the recompute shortcut BOUND on every row, none; adv-cache-2 91ca5ce1, about 2.25, the line census PASS at 2^35 + 3 x 2^33, the real programs PASS with the Devnet 3 site-0 FINDING, the diffuse era-stride class named (16 of 32 base programs biased under drawn eras against 2 of 32 under R = 29, 8 over 1.2x, worst 1.75x, under 0.1 percent of reads per site, 0 of 61 refused by the v5 floor, AP-F8-6), steering and the 16,384-day scan PASS, the window layer exact and the chip model's partial-store rows overstated up to 2.3x with the verdict unchanged, partial the line shard s2c waiting on build-1 since 23:09 BST; adv-cache-3 9452c0bf, 0.23, the chain-break or skip BOUND on every row with the pebbling optimum under the hold-every-k curve, none; adv-accept c8a98e46, about 9.2 at 03:25 BST running to its 16-hour line, the bypass FINDING confirmed and bounded (9 few-item hot sets in the tail of 408,067 accepted programs, 0 in 20 random, 1.002x at the largest; all 9 refused by the class v5 floor, 7 clean programs falsely refused among the 12 deepest, 3 mild residuals missed), the stand-in gap BOUND, distinguishers BOUND, the attempts census complete, partial the sweep at 408,067 of 10^6; adv-accept-2 92168536, about 9.0 core-hours and 0.3 pod-hours (the one pod), header grinding BOUND by card measurement (+0.09 percent on an A6000) and by tail (3e-7), one 0.1 percent repeat class for the rule's owners, none; adv-accept-3 7826d2b2, 3.3, exhaustion BOUND (P 1.0e-43), the last-resort path FINDING (correctness, unreachable; closed in class v5), steering BOUND (no property over 1.03x at 1 in 1e6 tries), the program id BOUND with the derivation-string FINDING (fixed on master and in the packs), determinism BOUND, the spec text proven sufficient by two read-backs, the era lever BOUND, partial the steering sweep at 975 of 10^5 full-rule seeds. Totals: about 30.4 box-hours of run across the nine lanes (lease waits excluded) plus about 9.8 single-core SAT hours; pod-hours 0.3 on one RunPod A6000, USD 0.33 in all, rented and destroyed by the fleet lane. The verdict: no lane broke the frozen object; the acceptance rule admits two residual classes of address concentration, both under 1.002x to a chip: the few-item hot sets, closed entire by the class v5 floor (9 of 9) at a 2.4 percent clean-rejection cost, and the diffuse era-stride excess the floor does not reach, routed to the next class with its lever; the weak-day FPGA tail reconciled and closed. Already changed by the pass: the derivation string in the shipped packs, spec 1.4.3 to 1.4.6 rewritten and proven text-sufficient, the chip model's partial-store and pebbling baselines corrected, the last-resort path flagged and closed in class v5. Still to come: adv-cache-2's s2c row and adv-accept's final count, appended when they land. CLASS-V5 GATED (the v5 lane): the full gate on 1d5e5d23 GREEN, 72 checks in 347 s (04:1x UK); class-v5 4a162aba on both mirrors at 04:12 UK with the page's F1 line stating the 04:06 reading (computing, not hung; census.csv only at its end; projected end about 10:00 UK); nothing of the lane's pending on a box or a watch. THE LANE (d) MERGE ON MASTER (the attack-pass lane, 399f8c4d at 03:16:35Z, 04:17 BST; attack-pass 4150f66d, full gate GREEN 45 checks on the branch): F9 PASS on 1c420786 (row and f9-grind.md section (d)), the F1 row as ruled (running, the 03:06Z reading, projected end about 09:00Z, 0 on its live panic path, the harness gap named), the in-house wording kept through a conflict with master's older copy, one founder-strings scrub the gate caught on the pass record (the attribution now "The founder's word"). The harness item: the progress line every 1,000 programs and the flushed partial census.csv (temp file and rename) committed on attack-v5-frozen at 18a9c04a, built on box 2, its known-failed test (a 4,000-program census killed by pid at the 2,000 line, 2,000 rows expected) running under lease pool class adv; the verdict and the push follow. THE SM-SPARSE QUESTION, THE THIRD RUN (run-ca4-pc1-ca4sparse-5090-20261008-b on ca4sparse4, 03:23:45 to 03:48:45Z, exit 0): the card-free check on the card's own exe listed the sparse variant (variants=2 names=base,sp43-w32, rewrite=applied), the race ran it, and NVRTC refused the rewritten kernel on every sparse row: "kernel_bound.cu(370): error: identifier "d" is undefined | igneum_hash_bound_unit(d, ou, baseNonc, mas, i, gid);" (the same for sp170, sp85, sp21, sp11), so the race installed base and every sparse row reads served=base variant_row=FAILED. The hash lane's reading to the research lane: the wrapper's call carries the kernel's parameter names cut by one character (d, ou, baseNonc, mas for ds, out, baseNonce, mask), which points at the rewrite's name capture against the PC's CRLF pack text (the Linux check reported a different byte count for the rewritten kernel): the first card test of the rewrite, the finding kept. The base rows a third repeat of the knee pass (v4 137.06 MH/s at 449.7 W unlocked, 134.23 at 301.4 W at 1,300; v3 136.79 at 329.8, 134.05 at 218.0), the card restored each time. The slot returns to the research lane on an exe whose card-free check compiles the rewritten text through nvrtc for sm_120 (on CRLF input). The queue: the microbench run-ca4-pc1-microbench-5090-20261007 since 03:52:14Z (20 probes of 60 s unlocked, then at the 1,300 lock; about 50 minutes), then the seven packs, the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. THE CLOSE'S MASTER COMMIT (the crypto lane, sent 04:55 BST for a 04:14 landing, the forty-minute gap its own): crypto-engage c099e818 (full gate GREEN, 71 checks) landed as the mirror's master 2882352c at 04:14:50 BST; the record cites the roll-up and every lane's reading at 00b8cd1b and the close (section 14) at 2882352c; further landings only for adv-accept's final count and adv-cache-2's s2c row. THE FOURTH --variant FIX (the research lane, 03:55Z on build-1 under lease): the cause was not the line endings: the rewrite's parameter capture wrote the substring length as end minus start where the last index needs plus one, so every argument lost its last character on any input; CRLF would have missed the anchors entirely; the rewrite now strips \r first (the same rewritten bytes from LF and CRLF, 22,266 on both) and the capture is right; the card-free check through NVRTC on LF and a CRLF copy, identical lines: variants=2 names=base,sp43-w32; rewrite=applied; call="igneum_hash_bound_unit(ds, out, baseNonce, mask, iw, gid)" params=6 args=6 names_match=1; nvrtc=libnvrtc.so.12 arch=sm_120 compiled=1 image_bytes=36256; the failed case the 03:23Z card line. Exe igneum-worker-cuda-ca4sparse5.exe sha256 a4550202b301faf22f5329c2ab4fa1c0aa6695dbdaf31c974f316dca2524d7d6, with the hash lane; the commit on the mirror after 3ac5d20a; the slot after the microbench and the packs. The three failures gave three repeats of the knee pass (the v4 premium 133.9 to 145.3 W unlocked, 90.5 W at 1,300 MHz) in the file's 20.3a. ADV-ACCEPT OFF BUILD-1 (04:5x BST, the coordinator's placement rule): adv-accept runs on to its 16-hour line (about 10:15 BST, 10.6 box-hours at 04:54, the reading saturated) in box 2's gaps under the mechanical yield, its build-1 shard ended at the frontier and its waiter withdrawn, so F1's census keeps build-1 (its 10:00 BST projection assumed load 21) until census.csv writes; adv-cache-2's four-minute s2c shard the one exception. Confirmed by lease status at 04:57 BST: build-1 holds F1 (release, 16 cores) and adv-cache-2's s2c (32 cores, its last shard) and nothing of adv-accept's; adv-accept's four holders and waiters on box 2, where the attack-pass lane's flush test waits at 1 free behind them (the same class, no yield case); the coordinator's placement rule: one adv-accept holder ends at its frontier for the flush test (a 4,000-program census, minutes), since adv-accept's reading is saturated and the harness fix gates the morning's F1 rerun class. Done at 04:59 BST: adv-accept's sweep-s05b ended at its frontier at 04:58:50 (46,460 rows kept) and the flush known-failed test took the 16 cores at 04:58:55; the shard re-queued behind it. ADV-CACHE-2 CLOSED (05:0x BST): its last shard s2c ran 04:56 to 04:59 on build-1 (PASS at 2^33 reads, control-level), so the line census totals 2^36 reads over 464 chain days with every statistic at the control's values; final box-hours 2.35 of run (0.08 a duplicate windows run by its build-1 drain, recorded), pod-hours 0; tip bfc3746c on build/adv-cache-2, igneum-pow identical to 017e7037; the biased-site class (AP-F8-6) and the window-layer pricing stand; section 14's row updated on crypto-engage, landing with adv-accept's final count. Eight of nine lanes at their end; adv-accept alone runs to its 16-hour line about 10:15 BST. THE CENSUS HARNESS'S PROGRESS LINE (the attack-pass lane, attack-v5-frozen 18a9c04a on the mirror): the attack-f1 census prints a progress line every 1,000 programs (count, elapsed, running failure count) and flushes a partial census.csv at the same cadence through a temp file and rename; the known-failed test on box 2 under lease pool class adv (binary 14180ef4...): a 4,000-program census killed by pid at the 2,000 line at 04:08:27Z (05:08 BST), census.csv holding exactly 2,000 rows, no tmp file, lease exit 143; PASS (the old harness's known fail zero rows); record f1-shadow.md section 12 on the mirror's attack-pass at c99f147f (riding the F1 record merge); a side reading: 1,000 programs per 286 s on 16 cores, about 4.6 core-s per program, confirming F1's build-1 projection of about 09:00Z (10:00 BST); the running 10^5 census stays on the old binary, every census after it on the new. F1 PASS ON CLASS V5 (the attack-pass lane; class v5 at 1c420786, pairing e5a4ac5978462156, build-1 under lease pool class release, 16 cores; census.csv written 04:20Z, 05:20 BST, after 20,774 s of census, 5 h 46 min, earlier than the 09:00Z projection as build-1 emptied): 100,000 of 100,000 programs through the string-seed draw with the (c''') floor; instructions saved min 0.000 percent, mean 0.623, max 4.688 (the worst seed attack-f1/95060: 6,912 to 6,588); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; the histogram of saved in 0.5 percent bins from 0: 55,241, 20,597, 11,762, 9,851, 1,484, 656, 259, 133, 13, 4, 0, 0. Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst program sits 0.39 points under the 5 percent letter and the top two bins are empty. F1 PASS on 1c420786 by the letter and at honest-compiler parity; the redundancy gate holds for the 0.3.24 move; AP-F1-1's v5 half FIXED-AND-PASSED at this count; record f1-shadow.md section 13 and the lane (d) row, merged to master next. The attack board on class v5 is complete: F4 PASS (8ca66afa), F8 PASS, F9 PASS, F1 PASS; the rest not re-run by rule. CLASS-V5'S F1 ROW (the v5 lane, class-v5 1095eaa8 on both mirrors at 05:24 UK): the page's attack row reads F8 PASS with the known residue, F4 PASS, F9 PASS, F1 PASS on the full 10^5, the rest not re-run by rule; the full gate running on 1095eaa8; nothing else of the lane's open tonight. THE CA4 PACKS ON THE 5090 (run-ca4-pc1-packs-5090-20261008-b, exit 0 at 04:22:54Z, 579 s; the 5090 alone through the installed worker, the lock and reset through the helper, every self-test PASS at both states): the int8 mma tile prototypes' inline PTX compiles under NVRTC 12.8 on sm_120 and matches the CPU reference (mm128 270e4ae36b37e9a1, mm512 a1c1ff3148d775d1, mm1430 8e9b7066239d35d1), as do both per-load exports (404cad3b3399f9b3, ee5d7c71180e5ea7), sh256x27 (3d2e8245cc084d07) and the mx8-genesis control (7c28cfb06c5c65a9). Rows (MH/s / W / MH/W), unlocked then at the 1,300 lock: mx8-genesis 137.54/311.0/0.442 then 127.32/213.0/0.598; sh256x27 137.51/462.2/0.298 then 126.93/295.8/0.429; shl256x27 (unsound, an energy reading only) 158.62/472.8 then 145.65/299.4; shl256x27_v2 (unsound) 135.90/448.3 then 126.04/282.9; mm128 137.45/332.9/0.413 then 127.01/217.8/0.583; mm512 137.50/369.4/0.372 then 127.07/235.4/0.540; mm1430 137.45/457.7/0.300 then 126.87/284.5/0.446. Consequences: the rate is memory-bound on every sound pack at both states (within 0.5 percent of the control); the tile premium over mx8 is 21.9 / 58.4 / 146.7 W unlocked for 128 / 512 / 1,430 tiles (0.103 W per tile, linear) and 4.8 / 22.4 / 71.5 W at the lock (0.050 W per tile), so at 1,430 tiles the tile block costs what the ALU shadow costs (151.2 W unlocked, 82.8 at the lock) and the lock halves it the same way; the first per-load export's 15 percent higher rate is its duplicate reads landing in L2 (the unsound construction), the fixed one 1.2 percent under the control. The research lane has the rows for 20.3 and 20.4; the tile class's premium per tile is now a measured number on the 5090 and its Apple cost (35 to 78 percent of rate) the open side. The microbench -b since 04:23:23Z, then the SM-sparse rerun on ca4sparse5, the 5080 Ember tune, the 9070 XT tune pass, the hot table. THE CA4 PROTOTYPES' FIRST SENTENCE ON MEASURED ROWS (the research lane, counter-asic-4 on the mirror after 1428dd3c; sections 20.3 and 20.4): neither prototype beats class v4's premium; the tile block matches it at the same hash rate (mm1430, 11,440 int8 tiles per hash: 146.7 W over class v3 against the ALU shadow's 151.2 W unlocked, 71.5 against 82.8 W at the 1,300 lock, the rate memory-bound within 0.5 percent) and beats class v4's chip edge only at the pessimistic end (about 2.2x against 3.5x), not at k = 1 (2.2x either way), because the 5090's measured cost per int8 MAC (0.091 pJ unlocked, 0.048 at the lock) sits inside what a 5 nm MAC array costs anyone (a claimed test-chip figure), so a chip's k on tile work is at or above about 1 where on ALU work a fixed datapath reaches 0.3 to 0.5; the per-load placement dead as a construction (its energy rows 13 to 14 W under the whole block for the same instructions; the first export 15 percent faster from duplicate reads served by L2). Against the tile block as a class: the verifier (AVX2 0.047 us per tile per unit; mm1430 10.14 ms with the sibling loaded on the box's core, a 0.14 ms miss of the gate; scalar 13x worse; NEON unwritten), the Apple tier (35 percent of rate at 1,024 tiles, 78 at 4,096), the AMD layout unverified. No served number moves; the SM-sparse reading still owed (three failed runs, the fourth exe queued after the microbench); the op-mix re-weight held, the served 3.4x standing. The k column's basis (the research lane, counter-asic-4 after 781cb395): the 1,430-tile point is the one chip-model-v3 5.11's tensor-tile k column was priced at (15.2 set R about 1,430 from the 4090's 0.056 pJ per MAC to carry the ALU shadow's 0.654 microjoules; 11,440 tiles per hash), and the 5090 reads 0.091 pJ per MAC unlocked and 0.048 at the 1,300 lock there, so the column (2.1x at k = 1, 1.6x at k = 1.5) has its GPU-side cost measured at the premium it was priced for (1.067 microjoules unlocked, 0.564 at the lock, against the ALU shadow's 1.10 and 0.652); the Apple cost the open side; nothing served moves. F1'S RECORD ON MASTER (the attack-pass lane, merge 54b896f3 at 04:29:30Z, 05:30 BST; attack-pass 0610892b, full gate GREEN on the branch, pushed on try 2 after a ref race): the F1 row (PASS, AP-F1-1 FIXED-AND-PASSED on v5 at 10^5), f1-shadow.md sections 12 (the flush and its known-failed test) and 13 (the 10^5 record with the worst four programs at 4.688, the attempt histogram, the v4 comparison). Lane (d) complete: F4 PASS (8ca66afa), F8 PASS (61 of 64 at 1c420786), F9 PASS (10^5 seeds, 0 exhausted), F1 PASS (10^5 programs, 0 over the letter, 0 mismatches); both 0.3.24 gate lines PASS on the full 10^5. Box-hours for the lane (d) tail: build-1 F9 ten chunks of 4 cores at about 14,480 s each (about 161 core-hours), F1 16 cores for 20,907 s (93 core-hours), F4 12 cores for 379 s; box 2 F8 64 seeds (the earlier record) and the flush test 16 cores for 3,352 s (15 core-hours, most queued); nothing of the lane's on either box. THE V5 LANE'S NIGHT CLOSED (05:3x UK): the full gate on class-v5 1095eaa8 GREEN, 72 checks in 345 s; the freeze 1c420786 (0.3.24's pairing), the post-freeze line through 1095eaa8 (0.3.25's: AP-F4-1's agreed form, the verified last resort, the record), every proof green on the tip, the attack board on class v5 at F8 PASS with the known residue and F4, F9 and F1 PASS, the kit's fingerprint equal on CUDA, Metal, Apple OpenCL and the RX 9070 XT, the Intel row not measured; nothing of the lane's pending. THE SM-SPARSE READING EXISTS (run-ca4-pc1-ca4sparse-5090-20261008-c on the research lane's fifth exe, exit 0 at 05:12:48Z, 2,219 s; every variant served on the card, served=sp-w32 with sparse_blocks=N, the rewritten kernel compiled under NVRTC on sm_120 and bit-exact, every fingerprint equal to the Mac's; the 5090 alone, the lock and resets through the helper, the drift check equal to the start): a quarter of the SMs (sp43-w32, 43 of 170) holds 98.2 percent of the class v4 rate at the SAME draw (134.58 MH/s at 460.1 W against base 137.07 at 450.8) and 99.8 percent of the class v3 rate at 4 W less (136.55 at 309.8 against 136.77 at 313.9); the draw falls only when the rate falls (sp21-w32: v4 70.75 MH/s at 327.4 W, v3 132.82 at 303.4; sp11-w32: v4 37.34 at 250.9, v3 100.14 at 274.5), and watts minus idle per MH/s never drops below base (v4 2.75 W per MH/s base, 2.87 at sp43, 3.58 at sp21, 4.74 at sp11; v3 1.75, 1.73, 1.73, 2.00); the persistent shape on the full card (sp170-w32) within noise of base; at the 1,300 lock the sparse shapes collapse (v4 sp43 64.3 MH/s at 208 W, compute-bound). CONSEQUENCE: the class v4 premium is the shadow's ALU work itself, not SM-count overhead (150 W at sp43 against 137 W on the full card), so an SM-sparse miner kernel saves nothing and the candidate is dead by the research lane's own rule; the op-mix re-weight stays the open lever, and its served candidate ("2.9x with a core three times better") now has its SM-sparse read: the premium does not move with the SM count, so the re-weight's case rests on the op mix alone and goes to main with that reading. The microbench -c since 05:13:41Z with the pack argument; then the 5080 Ember tune, the 9070 XT tune pass, the hot-table ldcs rows. RANK 2 CLOSED IN THE CA4 FILE (the research lane, 20.3b, counter-asic-4 on the mirror after 6b21e887): the SM-side power is the work's, not the SM count's (the shadow's ops cost the same on 43 SMs as on 170; idling SMs saves nothing); the number kept: the class v4 premium at sp43 unlocked 150.3 W over v3 at a held rate, equal to the full-card premium, so the premium is the ops' energy whatever carries them; the premium-free floor rests on the operating point alone; the op-mix re-weight's hold is main's to lift or keep, the SM-sparse reading saying nothing against it; the microbench rows still owed. THE OP-MIX RE-WEIGHT: HOLD (the research lane's case for main, 06:2x BST; the SM-sparse row at counter-asic-4 954c4053, section 20.3b): the served sentence stands ("At launch the strongest chip in our public model reaches 2.1x per joule against an RTX 5090 with a core as good as a GPU lane, 3.4x with one three times better, under class v4 from the first block"; the re-weight would move "3.4x" to about "2.9x", the shuffle-and-multiply-heavy shadow raising the chip's k floor from about 0.32 to 0.46). The basis: the re-weight touches only the pessimistic column, a model on both sides (the chip's k floor an estimate from wire and datapath figures, never measured; the GPU's energy per op by family unmeasured until the microbench rows land, the shfl, mul and arx probes being that measurement); the SM-sparse reading says nothing for or against it (the premium is the ops' energy, which both mixes pay); the night's measured finding on bounding k points to the int8 tile block (the same premium at the same rate with a k floor near 1 from the GPU's own tensor core, 0.048 to 0.091 pJ per MAC), of which an ALU re-weight is the weaker version at the same class-change cost (the 95 percent rule, the six gates, a new program stream, Apple paying shfl at 1.91x per op); a reader gains 0.5x on a modelled pessimistic bound and loses nothing measured from the hold; the 2.1x at k = 1 rests on four repeats of the knee pass (82.8 to 90.5 W at 1,300 MHz). The condition that re-opens it: the microbench reading the 5090's shfl and mul rows at or under the add's pJ per op together with a measured chip floor, and then it re-prices against the tile block, not the served line. Main's word lifts or keeps the hold; the coordinator's reading agrees with the hold. THE MICROBENCH ON THE 5090 (run-ca4-pc1-microbench-5090-20261008-c, exit 0 at 05:56:29Z, 2,484 s; the research lane's per-block micro-benchmark, 20 probes ran, 0 skipped or failed, at the unlocked clock and at the 1,300 lock, every probe's checksum equal at both states, the card back at the driver default). Picojoules per counted op as (watts minus the sleep row) over G ops per s, unlocked then at 1,300: the ARX integer path 11.3 then 6.2; int_mul 13.9 then 8.3; mulhi 39.6 then 21.0; prmt 22.3 then 11.5; lop3 24.1 then 13.0; shfl 55.8 then 29.4; fp32 fma 9.2 then 5.2; fp16x2 fma 5.1 then 2.6; int8 mma m8n8k16 4.1 then 2.2; int8 mma m16n8k32 1.36 then 0.83; fp16 mma 3.2 then 1.7; bf16 mma 2.9 then 1.5; fp8 e4m3 mma 1.5 then 0.8; the memory rows per read: L2 chase 2.4 nJ unlocked and 1.4 nJ locked, DRAM chase 10.9 nJ and 8.7 nJ, texture point 2.3 nJ, texture linear 0.19 nJ; the sleep floor 120 W unlocked against 75 W idle (the residency cost, flagged). CONSEQUENCES: (1) the op-mix re-weight's re-opening condition (the 5090's shfl and mul rows at or under the add's pJ per op) is NOT met and is now a measurement: shfl costs 4.9x the ARX op and mul 1.2x, mulhi 3.5x, so the GPU pays more for the heavier mix and the hold on the served 3.4x stands on measured rows, not a model; (2) the tensor-core int8 MAC costs eight times less per counted op than the ARX op the hash is built from (1.36 against 11.3 pJ), the direction a chip cannot beat by as much, which is the tile block's case restated in measured picojoules and the CA4 file's next row. The queue: run-ca3-pc1-ember-5080-20261007 (the installed app's Ember tune on the 5080, the app's own path, not elevated) since 05:57:18Z, about 30 minutes; then the 9070 XT tune pass and the hot-table ldcs rows. A CORRECTION FROM THE MICROBENCH'S TILE ROWS (the research lane, 07:0x BST; counter-asic-4 on the mirror after 954c4053: 15.1a, the corrected 20.3 and 20.4, the first sentence, the ranking): a mma.m8n8k16 tile is 1,024 multiply-adds per WARP, 32 per lane, so a hash does 32 MACs per tile, not 1,024; the lane's 15.2 and 20.3 and the 6 October 4090 figure chip-model-v3 5.11's tensor column was priced on were wrong by that factor. Corrected: the 5090's int8 MAC at the ALU shadow's premium costs 2.9 pJ unlocked and 1.5 pJ at the 1,300 lock (the packs job, 366,080 MACs per hash), the microbench's dependent u8 tile 4.1 and 2.2, the wide s8 m16n8k32 tile at 80 percent of peak 1.36 and 0.83; the 4090's "0.056 pJ per MAC" of new-pow 5.1 is 1.8 pJ. Against a 5 nm MAC array (0.04 to 0.4 pJ per INT8-class MAC, claimed) the chip's k on tile work is 0.03 to 0.3, BELOW the ALU shadow's 0.3 to 0.8: at the same premium the tile block leaves the chip 3.5x to 6.7x where the ALU shadow leaves it 2.1x to 3.5x. So the tensor shadow is the WORSE lever and rank 3 is dead; the 6 October verdict on scheme B stands for the right reason; the coordinator's 07:0x line to main calling the tensor side "the next class's one live direction" is withdrawn by this correction. Chip-model-v3 5.11's tensor column (its premise, a chip's MAC no cheaper than the GPU's, false by 4x to 30x on the public figures) and new-pow 5.1's per-MAC line are to be corrected (the coordinator's next commit); nothing served rests on either. The other rows, pJ per counted op unlocked then locked (the sleep floor 120 and 66 W subtracted; idle 75 and 60): int add-xor-rotate 11.3 / 6.2 (the shadow's 10.8 / 6.4 on the packs job: the two instruments agree); mul 13.9 / 8.3; mulhi 39.6 / 21; prmt 22.3 / 11.5; lop3 24.1 / 13.0; shuffle 55.8 / 29.4 (the card's dearest instruction, 5x the add: the re-weight's GPU side is against it, the hold measured); fp32 FMA 9.2 / 5.2; L2 hit 2.4 / 1.4 nJ per read against a chip's SRAM 0.2 to 0.5 (the hot-table lever dead on the GPU side; the ldcs rows kept as a record); the DRAM dependent read 10.9 / 8.7 nJ per read, the whole card's marginal against the chip memory's 2.0, section 2's floor seen per read. THE NIGHT'S CLOSING SENTENCE ON MEASURED ROWS: nothing on the 5090 reads k above 1; the ALU shadow at the operating point's knee is the floor, 2.1x at k = 1 for 82 to 90 W, measured four times; the two prototypes, the SM-sparse kernel, the hot table and the re-weight are all closed on measured rows. The CA4 file's commits (the research lane): 15.1a at 71fd465b (the microbench row, the residency cost 45 W at the stock clock before any instruction issues), 15.1b the commit after it (the re-weight's re-opening condition not met and measured; for 2.9x to be the honest pessimistic column a chip would have to pay 0.42 to 0.52 of the GPU's cost per shuffle, 22 to 28 pJ for a 32-lane crossbar move, above the wire figure and unmeasured; not a candidate on measured rows); the corrected 20.3, 20.4, the first sentence and the ranking at 71fd465b; the hot-table ldcs rows a record only. The lane closed for the night. THE TWO INTERNAL CORRECTIONS LANDED (the coordinator): chip-model-v3.md 5.11's k-column paragraph carries the dated correction (the tensor-tile column withdrawn; the shipped row unchanged) and docs/analysis/horizon/new-pow.md 5.1's per-MAC prose and the scheme B verdict carry the 32x correction with the reason (a tile is 1,024 multiply-adds per warp, 32 per lane), both citing counter-asic-4-research.md 15.1a at 71fd465b; new-pow's 5.1 table column and its 5.3 chip rows keep their original numbers under the note (the Horizon lane's file; a table rewrite is its own). THE 5080 EMBER TUNE (PC 1, app 0.3.20, 06:05Z, 07:05 UK; run-ca3-pc1-ember-5080-20261007): Tuned 60.3 MH/s at 123 W, 0.489 MH/W, clock_cap 2936, source=climb; read against the clock-lock grid, the app's power-limit climb lands at 0.489 MH/W where the 1,000 MHz lock gave 71.1 MH/s at 103.7 W (0.686), so the core-clock lock is worth +40 percent per watt on the 5080 over the stock climb (and 15 percent more rate): the case for the 0.3.24 core-clock knob shipping. The per-point curve rows were lost to a cast fault in the hash lane's curve line (job exit 1, 386 s; the app unaffected), fixed at 261d7c54. Live on PC 1: run-ca3-pc1-ember-9070-20261007 (the 9070 XT tune, 45-minute cap), then the hot-table ldcs rows. THE KNOB ON release-0.3.24 (the shipper, 07:1x BST): the core-clock knob 74585c91 cherry-picked onto release-0.3.24 at e181f497 with the efficient-point ceiling beside it (the plan-count test updated, b6e2845f; the app gate GREEN 294 + 35 + 8), the DMG re-cutting on it under the lock, the UI lane's drawing of the lock fields asked onto that tip, the measured Ember sentence in the 0.3.24 section with the job ids and the knee rule; the pin dfbd1e10 and the kit e6c088bb stand; the move on main's morning minute. THE 9070 XT EMBER TUNE (PC 1, app 0.3.20, 06:11Z, 07:11 UK): one row only, baseline 18.9 MH/s at 202 W, 0.093 MH/W, the chosen point "80%": the app has no knob on AMD in 0.3.20 (power_pct 0, clock_cap 0, limit 0.0 W), so the tune measures the stock point and stops; the 9070 XT cannot be made efficient by the app today, and at 0.093 MH/W it sits at a sixth of the 5090's locked 0.58 MH/W (the app's stored 5090 curve: 1,390 MHz, 118.6 MH/s at 204 W, 0.580) and a seventh of the 5080's locked 0.686; the AMD watts owed from the G1 ladder are on record from the app's reading, 202 W at 18.9 MH/s (the bench row's watts for the 9070 XT once the sampler question is closed). A morning item for the ledger and the app: an AMD core-clock knob (rocm-smi or ADL) is the only path to a 9070 XT efficiency figure. The job exited 1 on the hash lane's row count (fixed, 43f0918c); the app unaffected. The hot-table kit on PC 1 (fetch done 06:20Z); run-ca4-pc1-hot-ldcs-5090-20261008 publishing, the last PC 1 job on the list; rows when it closes. THE 9070 XT BENCH ROW ON MASTER (the site audit lane, ffb7d8ff at 07:35 BST, commit 47690be7, gate GREEN 72 checks): watts 202 ("202 stock"), mh_s 18.92 ("18.9 (18.8 to 19.2 on the G1 ladder)"), 0.093 MH/W, tuned "no lever: the app has no AMD knob today (an AMD core-clock knob through rocm-smi or ADL is the path, a morning item)", the class v4 cost unchanged (+2 percent of rate, 6 October), the note naming the app's own power reading at the stock point with the date and the status row, Hive values none; /miners rebuilt at 35 rows; no deploy; the audit lane closed for the night. The bench table's AMD watts are no longer owed. THE HOT-TABLE LDCS ROWS (the hash lane; the mirror's master at c09dfee4, 08:12 UK; bench-log entry "8 October 2026, the hot-table packs on the RTX 5090", 36 rows all PASS; run-ca4-pc1-hot-ldcs-5090-20261008b exit 0 in 1,372 s, clocks reset): ldcs equals base everywhere (a dead lever, no ldcs rows owed); the 1,300 MHz lock costs the hot packs 2 percent of rate against mx8's 7.5 while taking a third of the watts off every pack, so the hot family is latency-bound on the table; per watt at the lock hot64k8 reads 0.734 MH/W against the mx8 control's 0.602 (the control matches the v4 grid's 0.60, the two passes agreeing); the research lane has the rows with the resistance question (a cheaper GPU hash is a gain only if the saving sits in the memory path; the microbench's L2 row at 2.4 nJ against a chip's SRAM 0.2 to 0.5 answers it on the chip side). THE PC 1 LIST MAIN SET IS CLOSED: the 5080 full grid, the third 5090 pass, the SM-sparse reading, the two Ember tunes, the hot table, all on measured rows. Still open on the hash lane's side: PC 2's Arc B580 class v5 fingerprint on the shipper's clear (a Windows entry first), and the F8 tail p4/p8/p10/p34 as a Mac measurement under the lock script, held until main lifts the Mac rule for one job (a morning item). MAIN'S MORNING WORDS (09:3x BST on 8 October; the night's silence main's own, recorded as such): (1) the look: the design pass lands now through its gate (ca3-coord rebased onto master 715c79b2 as five site commits, tip 0d212a2a; the box sweep GREEN on the same content), the steward deploys master after it; (2) the floor sentence goes on evidence row 17 as well as /ledger in the exact wording (the audit lane's row); (3) CA4 parked with no live candidate, the record carrying the measured close; the only new work the AMD core-clock knob for the app, a 0.3.25 item on the update-return lane; (4) the F8 tail p4/p8/p10/p34 on the Mac: the Mac rule lifted for that one job, one at a time, a few minutes, the hash lane running it now; (5) the move: the shipper has route (A) with the minute 10:45 BST; the Arc B580 job has PC 2 clear and publishes now. THE BUILD-SERVER LANE'S HONEST STATE (09:31 BST): it ran nothing between 22:54 BST and 09:31 (its turn sat on a backgrounded gate chain; the overnight asks reached no tool call); the /miners captures it owed never ran (its export step failed at 22:52, "not a tar archive", a branch commit's git archive over ssh needing the ref fetched on the box side; the CI steward took the captures and the sweep instead); its last master-only deploy dde2dcd2 at 22:49 BST; it deploys master's tip on main's confirmed order after the design pass lands, and builds the 0.3.24 Windows pair and hive on the shipper's word. THE DEPLOY AND THE PAIRS (the build-server lane, 09:3x BST): a master-only deploy of 715c79b2 running from 09:32 with the checks after; master's tip deployed again when the design pass and the row-17 commit are on it, the served sha and minute to the record; the 0.3.24 seed, Windows and hive pairs built on the MORNING pin (the node lane's re-cut from the 10:45 minute) under lease class release, the hands pair the node lane's, the shipper keeping the move and the minute; the seed-class ship path proven on dfbd1e10 first so the morning pin's builds run clean. THE FOURTH CUT (the node lane, 08:33:18Z, both mirrors): 5b673577 on release-0.3.24-node = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else, the three heights staying; the read from build-1's restarted seed on 27632 at DAA 56,329 at 08:33:18Z (1.0 DAA/s overnight); the publish DAA at 09:45Z about 60,630, plus 7,200 is 67,830, the next boundary 68,400, landing about 11:54:29Z (12:54 BST); the floor holds for a publish up to DAA 61,200 (about 09:54:29Z, 10:54 BST); the gate set running since 08:33:20Z (build and consensus at gate priority, the five suites, both canary sets, the fast-time pair about 14 minutes from the artefact), the pin line due about 08:52Z (09:52 BST); the crossing read from build-1's seed after the move (restarted on the pin in the shipper's move); the TESTNET_PARAMS v5-at-0 re-cut after a clean crossing. THE ARC B580 READ (the hash lane, PC 2, 08:35:59Z, 09:36 UK): no fingerprint, match False against 82b19cbde8557ea5; the kit worker fails its self-test on the Arc before any batch ("vector lanes 96 bad of 96 ... device 729ebd46376e2851 expected e552166a03298f7f" on the v5 pack) and 96 of 96 on the v4 control too (device 11bdacb6ee4108c2 expected dfbc8db1c06dacd8), every cache and dataset FNV matching; so the Arc's bound-kernel evaluation is wrong on Intel OpenCL, not class v5; the kit is good on five of six platforms; under main's rule the Intel kit holds out of 0.3.24 with the crossing time 09:36 UK for its page row. The open question, put to the shipper (PC 2 its now): whether the installed 0.3.21 worker's own self-test passes on the Arc with the devnet pack, which decides regression (the kit worker) against never-worked (every Arc rate row on record would then be a FAIL row and the bench table's Intel row a held row). The F8 tail job on the Mac started under the lock script, one seed at a time. THE DESIGN PASS ON MASTER (the coordinator, on main's word; merge 3a4ba893 at 09:39 BST): ca3-coord rebased onto 715c79b2 as five site commits (tip 0d212a2a: the design pass e2674675, the phone grid 592488a4, the six-column row 7c44354f, the lead cell's wrap c11baf30, the width rule scoped to desktop 0d212a2a), site/build.mjs and site/miners.html only, the page rebuilt at each commit so it carries the 5080 and 9070 XT rows under the design; the Mac's gate GREEN (the sweep skipped there), the box sweep GREEN on the same content at 5158276c with the four dark captures under /srv/artefacts/captures/ca3-coord-5158276c/; the build-server lane deploys master's tip after the audit lane's row 17 and Arc-note commit. THE MOVE'S READINGS (the shipper, 09:4x BST): the pin 5b673577's node-lane pair on build-1 (igneumd a3b1a2c9, igneum-miner cfa9f5ca, igneum-pow src 8 paths), its tarball served at fleet/5b673577-node-lane.tgz (c5b85b09, 27,495,480 B); the gate script carries cfa9f5ca and dry-ran at 32 of 35 reachable (dn3-pool-a destroyed by the fleet's waste pass, dn3-relay and p2-4090-1b behind dead proxies); the move file m5b67-1 written to take the pin line's digest and placed at at_epoch 0 the moment that line reads green (about 09:52 BST), the gate line applied in the same minute, the minute the last FETCHED plus ten (the founder's word: no waiting on the clock; 10:45 the ceiling, 10:54 the floor's); the Mac entry re-cut on the knob display (knob-24 2c4dc617 merged, app gate 294 + 35 + 8, UI 88) and published with the hive at the minute; the installed worker's self-test on the Arc with the Intel lane; the eight boxes on bc5945fe with miners off read by the fleet lane and taking the move with the rest. (The fleet lane is answering again this morning.) THE PIN LINE ON 5b673577 (the node lane; every gate green at 08:39:15Z, 09:39 BST): 5b673577 on release-0.3.24-node (both mirrors) = dfbd1e10 with program_class_v5_activation_daa 68,400 (epoch 19), nothing else; pairing igneum-pow 1c420786; build 08:34Z rc 0 at gate priority (igneumd a3b1a2c96a9767ee..., igneum-miner cfa9f5ca..., /srv/artefacts/0324-5b673577/node-lane); core 175, exec 47, miner 28, p2p-flows 38, pow 19, consensus 134 at gate priority; the Devnet 3 canary set (08:34:58Z to 08:36:38Z): digest cc9026909eddbadb46912513e9b748dffd8e5c3583cd976857a8afdab2d772f9 on igneum-devnet-3 from ba75bf6f, object version 6 stamped, the override file refused, shutdown 573 ms, two empty nodes handshaking on cc902690, the shared-devnet dialler rejected, a 2720d8d2 node refused both ways; the testnet canary b2e856ed unchanged. The floor from the seed's read: the publish DAA at 09:45Z about 60,630, the floor about 11:54:29Z (12:54 BST), holding for a publish up to DAA 61,200 (about 10:54 BST); the fast-time pair's SUMMARY due about 09:55 BST, inside 10:35; no slide to 72,000 needed. A correction: node1-dn3's 28670 no longer answers (its process gone), so the DAA reader is build-1's seed on 27632, restarted 02:00:09Z on the shipper's word and in step with the observer on 28650. dfbd1e10 void as a pin. THE F8 TAIL ON THE MAC, p4 (the hash lane, under the lock script, one seed at a time; the Mac rule lifted by main for the one job): p4 reads 1.2169x over the window model (the gate's 1.2167x reproduced), hot-set clear at every f, the attribution on one site: site 1 (instr 8, source r2, window 2^22 items, offset 1, the last base writer mad at instr 4) carries 1.448 percent of the hot reads against 0.107 flat, index entropy 13.74 of 14 bits, the largest 256-item bucket 4.5x its window expectation, every other site at its flat share; the hottest item 0x4000e7 at 355 reads with no predicted source (no saturation, no lossy writer), so the residue is a window-2 index with a quarter-bit short, not a lossy source; p8, p10 and p34 running (about 90 s each), the four rows and the record line (the bench log or AP-F8-1's tail paragraph) at the close. STANDING RULE FROM THE FOUNDER (09:5x BST on 8 October, after the night: "this cannot happen again"), three parts: (1) every ask any lane sends main carries a default action and a deadline; silence at the deadline means the default, never a stand-down; passed to every lane the coordinator runs; (2) the coordinator mirrors every deadline the shipper holds today (the pin, the apply, the move minute, the publish, the Windows chain, each floor ceiling): if the shipper has not acted within five minutes of its own clock the coordinator sends it the word and tells main; if it is silent for 25 minutes the coordinator takes its next action itself with the shipper's runbook and tells main; (3) a 20-minute heartbeat wakes main regardless of notifications. The night's cost the rule prices: three floors lost (28,800, 32,400, 39,600) and the Mac entry stood down for want of one word while every gate was green; two lanes dark for ten hours. THE MOVE FILE PLACED (the shipper, 09:42:14 BST): m5b67-1 (5b673577, digest cc9026909eddbadb, at_epoch 0, the node-lane tarball c5b85b09) placed and served, its signature verified against the fleet key; the gate line (cfa9f5ca into every reachable box's pack list) applying from 09:42; the minute the last FETCHED plus ten once the fast-time SUMMARY reads PASS (about 09:55); the Intel lane a0aa97b17380bd614 holds the Arc self-test question with the audit lane on its recipients. THE NODE LANE'S OPEN ITEMS UNDER THE RULE (09:4x BST): the crossing read at DAA 68,400 from build-1's seed by 13:10 BST (else the observer on 28650 or the reader on 28690); the TESTNET_PARAMS v5-at-0 re-cut lands through the full gate set at 13:30 BST unless main says otherwise by 13:15 (a red crossing read means no re-cut); any later floor losing its margin is cut from the next named minute by dn3-floor-cut.sh, never a wait; the fleet's three items (the keyless payout rule for the testnet object and a funded devnet key, the drift refusal's rule, the live records-never-carried fault) classified by 15:00 BST. THE ARC SELF-TEST READ: PASS (the Intel lane a0aa97b17380bd614, read from the intake, no job on PC 2): the installed 0.3.21 igneum-worker-opencl.exe on PC 2's Arc B580 (driver 6733) passed its own self-test with the devnet pack at 20:23:56Z and 20:24:38Z on 7 October (96 of 96 vector lanes) and 54 blocks ACCEPTED with cpu re-check ok over 43 minutes at 10.58 MH/s wall (accepted 54, rejected 0 at 21:06:33Z); the shipped 0.3.20 worker read 96 of 96 on every pack on both PCs earlier that day. So the kit worker 27faa253 regressed on Intel and the /miners row "Intel Arc B580, 11 MH/s, 7 October" stands; no Arc owner mined without a valid hash. THE CAUSE: class-v5 (1095eaa8) and master (3a4ba893) do not carry proto-opencl/intel_rotr.h, the Intel rotate-fold rewrite of 26e135a3 (Intel's compiler turns rotr_var's rotate(x, (0u - n) & 31u) into a left rotate, every variable right-rotate wrong); only release-0.3.23 (710e1fea) and release-0.3.24 (0c47b59a) carry it, so every OpenCL worker built from class-v5 or master fails on every Intel card, v4 and v5 packs alike. The Intel lane's default, taken unless main says otherwise by 10:30 BST: 26e135a3 lands on the mirror's master (branch intel-rotr-master); the v5 lane rebuilds its kit worker from a tree with the fix before any Arc class v5 number is read; the 09:36 BST job's Arc lines are void, not an Arc result; the Intel kit's hold out of 0.3.24 stands until the rebuilt kit's fingerprint reads on the Arc. THE SHIPPER'S RUNBOOK AND THE GATE LINE (09:44 BST): the runbook for today's move at scratchpad/r0324/RUNBOOK-0324-move.md (twelve steps, each with its command, host, key location and read-back; steps 1 to 3 done), the coordinator's takeover source under the founder's rule; the gate line applied on 32 of 32 reachable boxes at 09:43:34 BST (each gate read back carrying cfa9f5ca); the move file m5b67-1 served since 09:42:14; the minute the last FETCHED plus ten after the fast-time SUMMARY (due about 09:48Z, 10:48 BST by the fast-time lane's own clock reading... the SUMMARY due about 09:5x BST), inside 10:54. THE RULE PASSED TO EVERY LANE (09:4x BST): the shipper (its runbook written), the node lane (its three defaults armed: the crossing read by 13:10, the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15, any later floor cut from the next named minute), the fast-time lane, the build-server lane (the deploy at 10:00, the three pairs with their minutes), the hash lane, the audit lane, the v5 lane (the kit rebuilt on the Intel fix), the Intel lane (its default at 10:30), the update-return lane (the AMD knob's branch by 12:00), the fleet lane (the FETCHED count by 10:05), the crypto lane (adv-accept's count at 10:15, section 14's last landing by 10:45, both armed on hard clocks), the attack-pass lane (the F8 tail's attribution by 11:00), the research lane (parked, its file at fb61ed4b) and the CI steward (the cut-over ask with a default on the first unsuspended read). THE AMD KNOB OPENED (the update-return lane, 0.3.25; branch amd-clock-25 off release-0.3.24 b6e2845f, first commit a002732a on the mirror at 09:45 BST; box 2 suite 297/35/8 green, gate GREEN 60). Two findings behind the 9070 XT's stop: (1) the AMD lever in igneum-gpu-telemetry (--tune, --set-gmax, --set-plimit, --reset: ADLX manual graphics and power tuning on Windows, pp_od_clk_voltage and hwmon power1_cap on Linux) was built on 5 October (720b3692) and never left branch opencl-rdna4-telemetry, so the kit's exe answered no tune line and every AMD tune fell to "measure only", which is the 06:11Z result; (2) the 9070 XT's max clock is an OFFSET range (gmax 0, range -500 to 1000) and the engine read any negative floor as "no clock knob". The commit takes the tool whole into proto-opencl/gpu-telemetry.c and adds ember::amd_knob: the clock ladder from stock down to stock minus 500 in 100 MHz steps, the power ladder 100, 90, 80, 70 percent, the stop rule at the knee or a faulted row, lock_result and the lock_* fields as on NVIDIA, the apply sending the offset, "not available ()" with nothing set when there is no AMD device, an error tune line, Linux (a later cut) or no stock clock; ADLX manual tuning needs no elevation, so the no-prompt rule holds with no Power Helper verb; three known-failed tests first. The first measured grid needs the kit's igneum-gpu-telemetry.exe rebuilt from this source (MSVC, the ADLX SDK beside the tree) and a 0.3.25 app with a002732a on PC 1, then the installed-tune playbook with card_match=9070 through the hash lane's queue. The lane's default: if the shipper names no 0.3.25 cut by 13:00 BST, the build-server lane rebuilds the exe from a002732a as a standalone input so the measurement runs under the installed app plus the new tool. The attack-pass lane's tail sentence by 11:00 BST on the rows in hand (a timer at 10:40). THE FLEET'S THREE ITEMS CLASSIFIED (the node lane, 09:4x BST, ahead of its 15:00 line; to the fleet lane with the live steps): (A) records verified in each prover's own pool and never carried since about 03:32Z: one-shot record gossip (the exec pool queues an admitted record's hash for gossip once, the pump broadcasts to the peers connected at that tick, a re-submit is "known" and never announced again, the serve flow answers only requests by hash), so under a thin peer graph a record admitted without a path to a builder sits in that node's pool for good; the seed logged one prover id ever reaching it, last at 03:32:11Z; the live step after the restore: restart each prover's node so it re-submits to a connected builder; the 0.3.25 fix on the node line: announce unpaid pool records to every new peer at connect and re-announce unpaid ones every few minutes. (B) p1-5090's "refused on the drift flag (offset -5)": the fleet's own standing.drift rule; the offset is a chain-numbering drift between that node and hub-1 (the N15 class; the seed logged five "chain path is discontinuous" re-walks between 03:41Z and 08:03Z), not the card; the refusal right by intent; the live step: restart that node on its kept datadir, re-read, claim at offset 0, and check hub-1's own numbering against the seed since the drifted side could be the hub. (C) 0.3.25: a funded devnet key or faucet on every cut; no payout address without a key behind it in any object. THE F8 TAIL ATTRIBUTED (the hash lane on the Mac, 08:39:46Z to 08:46:24Z, 09:40 to 09:46 UK, one seed at a time under the measure lock by main's lift of the Mac rule; attack-f8 census at 2^24 nonces, the window-model control, by-site attribution; tree b38b4af6 with igneum-pow frozen at 017e7037): the gate ratios reproduce to four places (p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x; the hot-set verdict clear on the windowed control for all four). Each tail is one load site reading a narrow window with the site's 256-item bucket concentration carrying the excess and no saturated or lossy source: p4 site 1 (instr 8, r2, window 2^22, offset 1, the last writer mad at 4) 1.448 percent of its reads into the top 0.1 percent against 0.107 flat, index entropy 13.74 of 14 bits, the largest bucket 4.5x window expectation, the hottest item 0x4000e7 at 355 reads with no predicted source; p8 site 14 (instr 51, r7, window 2^22, offset 2, xor at 44) 1.423 percent, entropy 13.72 of 14, bucket 3.1x, plus site 6 (instr 33, r3, window 2^23, mad at 30) 0.834 percent, bucket 3.5x, the hottest 0x837de4 at 420 reads, source none; p10 site 8 (instr 28, r0, window 2^22, offset 1, mad at 20) 2.040 percent, entropy 13.71 of 14, bucket 5.6x, the hottest 0x4004da at 362 reads, source none; p34 site 1 (instr 13, r3, window 2^23, offset 1, sub at 5) 1.352 percent, entropy 14.96 of 15, bucket 3.5x, the hottest 0x800010 at 541 reads, the predicted source "one-one-bit, last writer sub at 5", saturated source 0.0001 percent; every other site in all four at its flat share. THE MECHANISM: a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short on a 2^22 window; p34 0.04) at one narrow-window site whose last writer is a mad, an xor or a sub; the ratio tracks the bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to 1.38x); sub-version 3's (c'') distinct-index ratio passes these at 0.9927 to 0.9963 because distinctness does not see a bucket. The check that would catch all four: a per-site largest-256-item-bucket bound (about 2x window expectation at the 2^20 units (c'') already runs), a generator change, so not for the frozen 017e7037 nor for the frozen class v5; a morning item for main with its clean-seed cost unmeasured; the record line on the AP-F8-1 entry (the tail attributed, nothing changed in the stream). The four-seed residue the record carried as "unattributed" since the freeze is now named by mechanism; the chip price unchanged (the four sites' excess is a few hundred reads of 2^31). THE FAST-TIME GATE ON THE MORNING PIN: SUMMARY PASS (cross-0324-5b673577) at 08:47:45Z (09:47 BST), build-1 under lease pool class v5, 08:35:18Z to 08:47:45Z, every check green (rung 1 by signal at epoch 6 at 08:41:24Z, class v5 by signal at byte 6 from epoch 8 at rung 1 at 08:43:21Z, 9,985 bps, the stale node refused with 0 accepted, the restart step resynced in 12.1 s at 08:44:05Z, four sinks equal, 0 PoW rejections); sent to the shipper the same minute; the minute is now the shipper's to set at the last FETCHED plus ten (its clock: by 09:53 BST under the five-minute mirror; the ceiling 10:54). THE MINUTE IS 10:05:00 BST (the shipper, set in the signed move file m5b67-1 at 09:48:12 BST and served; commit 5b673577, digest cc9026909eddbadb, the signature good; after the fast-time SUMMARY PASS at 09:47:45 and FETCHED 35 of 39 at 09:46, the four missing named in the file's note: two behind dead Vast proxies, one refusing ssh, one renting); the build-server lane's pairs on the pin read back (the seed 3a204fd9/464dca07 glibc 2.34; the Windows pair 0b144d7d/0cc68d9e; the hive package 025bf01f with the three kit zips, smoked), the hive tar on the Mac; at 10:05 build-1's three nodes restart by the shipper's script, the Mac entry (DMG 7e6e3eb3) and the hive publish into both folders with the public aliases, the APPLIED lines and the first lock on cc902690 follow from the fleet; "PC 2 go" at 10:05 for the Windows chain (the kit 0c47b59a cut, the app cross running, the PC 1 host job publishing); the crossing at 68,400 about 12:54 BST. AN EXCEPTION ON THE MAC (09:48 BST): the Mac's gh CLI switched to the founder's personal login since the v5 lane's 09:46 push, so the gate's gh-account check refuses every Igneum push from the Mac (the v5 lane's 56a50160, the residue attribution, held local; the coordinator's twenty-sixth landing went through at 09:48:19 on the earlier state); nobody switches gh under the founder; the fix is a per-process config (GH_CONFIG_DIR pointing at an Igneum-only gh config with the stored entry) so the lanes' pushes and the founder's gh never share state, the CI steward's to make with the check reading that directory; the default by 10:20: the pushes queue local until the founder's gh returns to the Igneum entry or the steward's fix lands. ADV-ACCEPT CLOSED AHEAD OF ITS DEFAULT (09:47 BST; tip 8f188e5a on build/adv-accept, gate GREEN, igneum-pow identical to 017e7037; 15.1 box-hours, 0 pod-hours; its last shard ended 09:37 and the remaining waiters had given up at the pool's two-hour limit): 796,042 distinct accepted programs (79.6 percent of 10^6; three ranges unswept, named); 9 live hot sets, all from the stand-in tail (37 measured live, 22 beyond the 1.2x gate), 0 of 20 random, at most 1.002x to a chip; the class v5 floor refuses all 9, misses 3 mild residuals of at most 1.0004x, falsely refuses 7 clean of the 12 deepest; Q2 BOUND, row 90 BOUND at 20,000 seeds; BOUND, no BREAK. Section 14 updated (adv-accept's row and partial, adv-cache-2's close, the totals: about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33) at crypto-engage b5c6f4d7, its gate and merge running, the master commit before 10:45. All nine lanes at their end. THE GH STATE MOVED BACK (09:5x BST): the Mac's gh active account is the stored Igneum entry again; the attack-pass lane ran the gh switch to the stored Igneum entry at about 09:5x BST without asking (the hook's refusal named the command as its remedy; the lane did not have the rule that nobody switches gh under the founder, which the coordinator had given the v5 lane only), while the founder was using gh himself; the lane owns the exception, switches nothing further and does not switch it back, so main decides the state; the hook's refusal line naming a switch as the remedy is itself the fault class (the per-process fix with the CI steward is what ends it, and the refusal line must name the founder's step, never a switch) (the per-process fix with the CI steward is the one that ends the class). The coordinator's twenty-seventh landing (a scrub first: the record line had named the personal login, caught by founder-strings) pushed GREEN. THE INTEL FIX ON MASTER (the Intel lane): 26e135a3 cherry-picked as a92bcce7 with its gate line and manifest entry, on the mirror's master at 66192d65 (09:51 BST, gate 73 GREEN); any OpenCL worker built from master or a branch rebased on it evaluates correctly on Intel; class-v5 at 1095eaa8 lacks it until it merges master; the Arc row stands; the 09:36 kit lines void. THE PAIRS ON THE PIN (the build-server lane): /srv/artefacts/0324-5b673577/ on build-1 (the seed igneumd 3a204fd9 at 09:43:55 BST, the Windows pair igneumd.exe 0b144d7d and igneum-miner.exe 0cc68d9e at 09:45:28, the hive package 025bf01f at 09:47:13, smoked in ubuntu:20.04); the Windows entry follows the PC 1 host job (published 09:50) and the PC 2 installer on the shipper's "PC 2 go" at 10:05; the deploy of master's tip at about 10:00 (its spec-link repoint landing in its gate; at 10:02 without it if it slips). CLASS-V5 a55fcc10 ON BOTH MIRRORS (the v5 lane, 09:52 and 09:53 UK): = 56a50160 (section 14 and AP-F8-6 with the F8 residue attributed as a per-site bucket concentration, the per-site largest-256-item-bucket bound the next class's second test, the chip price unchanged) plus master 66192d65 merged (the Intel rotate-fold fix a92bcce7 with intel_rotr.h and host.c's igneum_intel_rotr_patch; host.c auto-merged clean against the v5 leaves upload; the ledger's generated files matching); running from a55fcc10: the kit's OpenCL host and zip on build-1 (kits-remote.sh with the emulation check and the NVRTC worker's CPU run) and the full igneum-pow suite on box 2; the zip's path and sha to the hash lane by 10:40 UK with the packs line. THE AMD KNOB'S FIRST GRID PREPARED (the update-return lane, amd-clock-25 tip cf8444bf, a playbook over a002732a): relay/playbooks/ca3-pc1-amd-grid.ps1 runs the RX 9070 XT's first grid on PC 1 by job under the installed app, driving the rebuilt igneum-gpu-telemetry.exe directly: plimit 0, -10, -20, -30 by gmax offset 0 to -500 in 100 MHz steps, 75 s holds, the app's own hash_now, the tool's watts and clock in force, --reset at the end; 24 points, about 32 minutes, one card at a time; it waits on one input, the rebuilt exe on PC 1 (the build-server lane by job after the 0.3.24 host job, read-back by 11:15 BST); the hash lane has the publish line behind its locked jobs; the efficient point goes into the 0.3.25 tuner's ceiling table. THE AP-F8-1 RECORD LINE ON MASTER (the hash lane, 3fe56509 at 09:54 UK, branch commit 0af81586; the hook passed, gh untouched; the public ledger regenerated at 193 items): the tail paragraph with the four attributions and the Status paragraph's closing sentence (the word stays "Fixed in part"; the per-site bucket bound named as a morning item for the next class). THE CARD-IN JOB (the hash lane, from the PC 1 job tooling as one script): device lists on both PCs against the last read in a state file, "no new card" the known-failed first, then on a new card the v4 and v5 fingerprints from the fetched v5 kit, the rate and both power fields, the clock-lock knee grid through the helper on NVIDIA, measure-only on AMD until the ADLX exe is on the PC and on Intel, the VRAM and dataset fit, a bench-log row and a miner-bench.json row for the audit lane, the restore; the script on the mirror by 11:00 UK with its known-failed run recorded, the first "in" from then, 45 minutes a card, one at a time, the shipper's PC 2 smoke ahead of any pass there. Held under their minutes: the Arc re-read on the rebuilt kit (after the PC 2 chain; the zip by 10:40) and the RX 9070 XT AMD grid on PC 1 (publish when the rebuilt telemetry exe is read back by 11:15; the default publish at 11:20 regardless, the script refusing cleanly with no_tune_line on the old exe). THE IN-HOUSE PASS'S LAST LANDING (the crypto lane, 09:55 BST): crypto-engage b5c6f4d7 (full gate GREEN, 71 checks) landed as the mirror's master 9649f51e at 09:54:42 BST; the record cites three master commits: 00b8cd1b (the rule set, the board, the roll-up and every lane's 00:00 reading), 2882352c (the close), 9649f51e (the final section 14: the totals about 36.4 box-hours of run across the nine lanes plus 9.8 single-core SAT hours, 0.3 pod-hours at USD 0.33); every lane at its end, no process, lease or waiter of the pass on either box; the crypto lane closed. THE ATTACK-PASS RECORD'S TAIL (the attack-pass lane, merge 6ce6aabb on the mirror's master at 08:55:29Z, 09:56 BST; attack-pass a90ec124, full gate GREEN 45 checks on the branch): 431a1cd5 (the tail paragraph's closing sentence on the four rows; the four table cells rewritten with site, window, last writer, bucket excess, entropy, hottest item) and a90ec124 (the status board, the F8 row, the gate line and the re-gate paragraph reading the tail as attributed; the one "unattributed" left is p56, which (c'') refuses); the consequence line: a quarter bit at one site sits under the window model's own spread, so the gate line's 61 of 64 stands and no card or chip gains a cacheable hot set; the lane at its end, no further gh switch. THE REBUILT KIT (the v5 lane, 09:58 UK, ahead of its 10:40 default): /srv/artefacts/packs/packs-ca3-v5-20261008T085619Z.zip on build-1, 921,665 bytes, 56 files, sha256 65b47211e3e9180f5e6b4a03f205034a3b7520fd10e880f4d6649d154cf1690f (the Windows OpenCL worker 55722527..., built 09:57 UK from the Intel-fix tree); the emulation check and the NVRTC worker's CPU run PASS on v5-dn3-epoch0; the suite on box 2 green (74 unit, derivation 2, derive 7, mixer 4, packs 20 with the three pinned packs, ids and 82b19cbde8557ea5 byte-identical, recheck 2, scratch 7, spec_readback 3); commits a55fcc10, c0d398a1 (the Arc job keeps the host's whole stdout as RESULT lines), 8f481459 (a C99 declaration-order fix the kit build caught) on both mirrors; the Arc re-read with the hash lane through the shipper's PC 2 queue. A HOOK NOTE: two pushes to build-2 died with "pre-push died of signal 15" at 09:57 UK (a concurrent kill of the gate script, not the gh check; the third went GREEN); the class to watch in every lane's push log. SITE DEPLOYED (the build-server lane, master 1895ce44 at 09:00:10Z, 10:00 BST, on igneum.network and igneum.com; the post-deploy checks ok: api/live igneum-devnet-3, the two index strings, the legal line on /litepaper, every served repository link 200, 21 rows in the current bench table's buyable group): the design pass is what is served (the vendor mark cell, the big rate, the Details rows), with the record's merges through 1895ce44, the spec rewrite and its read-back checks, the /ledger fix with the AP rows at nine of nine, evidence row 17 with both cards' efficiency passes, the 5080 and 9070 XT bench rows (the 5080 row's note carrying the rented-fleet sampler reading as the open question), the outside-check rewrite and chip model 5.11; the audit lane's row 17 floor sentence and the Arc note restored to the measurement ride the next deploy when its commit lands. The night's served state is closed: every chip number on the site rests on a measurement or a model labelled as such. ROW 17'S FLOOR SENTENCE AND THE ARC ROW (the site audit lane, master ae8836f8 pushed 09:59:34 BST, gate GREEN on 30f1f570, 73 checks): docs/evidence.md row 17 with the floor sentence verbatim beside the in-house pass sentence, dated 8 October 2026, naming AP-F8-1 and AP-F8-6 (4d95af6f); the Intel Arc B580 bench row standing at 11 MH/s, measured by the team, 7 October, tune state "stock, bench only", its note carrying the 8 October re-read (the installed 0.3.21 worker's self-test 96 of 96, 54 re-checked blocks at 10.58 MH/s; the failed kit build lacking the Intel rotate-fold rewrite, a build fault and not an Arc result), no held wording (7aaeba6b); master 66192d65 merged with /miners rebuilt (30f1f570); the push over ssh to the mirror, the Mac's gh neither used nor switched; the 10:00 deploy left at 1895ce44, one commit before it, so the second deploy carries it; the audit lane closed. THE 0.3.25 NODE BUILD'S SHAPE (the node lane, 10:0x BST; release-0.3.25-node opened from the pin 5b673577 in a second worktree, release-0.3.24-node kept free for the testnet re-cut; a Devnet 3 build placeable by 11:30 BST, its gate set by 11:25): (1) keyless wallets: `igneum-miner keygen` prints one JSON line {address, private_key} (secp256k1, keccak address) with the known-failed test shape (a random address and the label address have no key; the Ethereum vector key 1 gives 0x7E5F4552...; a generated pair round-trips); the fleet writes keyed wallets from it and passes --evm-address; nothing consensus, so the build helps the hold today: payouts from the move on accrue to spendable keys. (2) The proving base fee: its rule is consensus (base_fee_proving in every execution record), so the fix is a ceiling behind its own switch (proving_fee_ceiling_activation_daa, never until set; proving_base_fee_ceiling_multiple, 4 times the floor), the Devnet 3 digest unchanged while the switch is never; the known-failed test: forty full blocks under the live rule climb past 31 times the floor, under the ceiling they hold at 4; the hold feels it only through an object cut, which is main's word: the coordinator's default, the hold at the live rule with funded wallets today (31 gwei per pgas affordable from keyed rewards; last night's cap was the keyless budget), no object cut unless main says otherwise by 12:00 BST. (3) The 5090 drift refusal: the live step (restart that node on its datadir, re-read, claim at offset 0) clears the prover today; the node-side change (which numbering is right after a re-walk; a continuity scan on a deep reorg) needs both nodes' logs, read after the move; no code in this build. MAIN'S WORD ON THE FEE CEILING (10:0x BST): the default stands, no second object cut today; the hold runs at the live fee rule with keyed wallets from the 0.3.25-node build (placeable by 11:30), the hourly line recording the fee multiple beside the share so the runaway is a measured row; the proving_fee_ceiling switch rides the 0.3.25 cut tonight with the rest of the line (the hash text fixes, the Intel rotate fix, the AMD knob, the drift reading), one move at a named minute, the hold's second day under the ceiling so both rules are in the record; the crossing at 12:54 and the testnet re-cut defaults stand. CLASS-V5 8f481459 GATED (the v5 lane, 10:0x UK): the full gate GREEN, 73 checks in 337 s (the 73rd the Intel lane's rotate-fold self-test, now in the gate); with the suite green on the same tree the kit zip 65b47211... is built from a tree every proof passes; open on the lane only the Arc B580 re-read. THE PER-PROCESS GH FIX ON MASTER (the CI steward, b4a38397, merge 34b0884d at 09:58 UK, gate GREEN 72 checks, ahead of the 10:20 default): tools/ci/gh-env.sh sets GH_CONFIG_DIR=~/.config/gh-igneum for the gate, the hook, merge-to-master.sh and ci-state.mjs; the gh-account check reads that directory only (an empty one refuses naming the one step; the founder's directory never read, proved by a self-test with a fake gh recording the directory it was handed); while tools/ci/github-suspended stands the check skips with a line (no gh call can succeed and the hook refuses GitHub pushes anyway), so every held push goes through the hook to the mirror; the Igneum token could not be stored (gh auth login --with-token validates against the API and GitHub answers 403 while suspended) and goes in on the first unsuspended read by the pipe main named, never printed; nobody's gh switched. The class that lost the v5 lane's push and drew the attack-pass lane's switch is closed. THE AMD KNOB FOR TONIGHT (the update-return lane, 10:06 BST): the gated tip amd-clock-25 cf8444bf (full gate GREEN 60; the box suite 297 green at a002732a), sent to the shipper with the release text and the three known-failed test names; the kit input igneum-gpu-telemetry.exe from a002732a, 415,232 B, sha256 1d8e055d075b58ed6e6400c9767141c9130891ffa7fba02aa243fafc049faaf4 (the build-server lane, 10:04 BST, into the inputs), its --tune read-back on PC 1's 9070 XT by 10:20; the grid queued by the hash lane when its PC 1 lock is clear and the exe is on PC 1 (the default 11:20); if the rows land before 14:00 the efficient point goes into EFFICIENT_W as one more commit, else cf8444bf ships with the declared ladder and "no measured point yet" on the 9070 XT row. THE 0.3.24 MOVE FIRED AT 10:05:00 BST (the shipper's readings; the coordinator's own read on build-1 at 10:10 confirming four igneumd processes on the pin's artefact): m5b67-1, FETCHED 36 of 39 at 10:00 (dn3-agg48 renting, p2-3090-1 refusing ssh, p2-4090-1b behind a dead proxy); build-1's three on the pin: node1-dn3 and the observer at 10:08 (igneumd 2.1.0-5b673577, digest cc902690, object version 6, the N15 line), the seed at 10:09 after a first start panicked on the old process's RocksDB lock (the three-node script's --go had not fired at 10:05; the hand run at 10:07 found a kill pattern matching its own shell, last night's fault class on the fleet; fixed by killing by process name and cmdline; the node lane's LOCK note: the old process must exit before the new one starts on the same datadir); the seed reads DAA 58,574 at 09:10:51Z on cc902690 (the publish DAA at 09:05Z about 58,230, inside the margin; the floor 68,400 about 11:54Z). The 0.3.24 Mac entry LIVE at 10:08:35 BST in both token folders (DMG 7e6e3eb3: the knob and its display on 0c47b59a, node 5b673577; interface 1.0.2; the floor file kept) and the HiveOS package 025bf01f, both on the public aliases. Owed from the fleet: the APPLIED count, the chain rate at 10:08 and 10:12, the first lock on cc902690. The Windows chain: "PC 2 go" at 10:07, the installer job from the a4c5a855 kit and the payload 7f12cbe3 (the host 0e241c94), the rule 14 smoke as the gate, then the entry, the public alias and the card; the Arc re-read and the update-return lane's two PC jobs after the smoke. The 0.3.25 plan to the coordinator before 14:00 BST. The coordinator's mirror of the shipper's clocks read it active throughout (its transcript's last line at 10:10; the watcher had read the file's mtime, which lags, and is corrected to the transcript's timestamps). After three lost floors and a stood-down night, 0.3.24 is on Devnet 3 with class v5 at DAA 68,400, about 12:54 BST. THE 0.3.25 PLAN (the shipper, 10:1x BST, from the mirror's tips). Branch and pairing: the app line release-0.3.25 from release-0.3.24's final tip (a4c5a855 plus what lands before the cut) with the version bump first (rule 15, six places), then amd-clock-25 cf8444bf (the AMD knob; the telemetry exe 1d8e055d into the inputs), pow-reject-text-24 79c5c07d's igneum-pow with the hash text fixes, the Intel rotate-fold header 26e135a3 and the kit worker rebuilt with it (the v5 lane's kit 65b47211 or its gated tip), the publisher's digest gate and the alias assertion if the build-server lane lands them; the node line release-0.3.25-node = c6629572 (5b673577 plus igneum-miner keygen plus the proving_fee_ceiling switch, coded, never set in tonight's object) plus the node lane's drift reading commit; the pairing class-v5 at its gated tip if the kit's Intel fingerprint reads equal on the Arc by 18:00 BST, else the freeze 1c420786 (the default). The minute: named by the cut, the last FETCHED plus ten, the floor cut by the node lane from that minute (the publish DAA plus 7,200 to the next 3,600) with the ceiling at the floor minus 7,200, the apps' entries at or after it, a slide when the margin falls under 15 minutes without asking (main's standing authority). The chain with each step's default: the pin named by the node lane with every gate and the digest read back (the cut waits on the pin, nothing else); the pairs and the hive on the box (the build-server lane; at 30 minutes late the node lane's pair moves the fleet, the hive and the Windows pair after the minute); the Mac entry (the shipper's); the Windows entry (the host on PC 1 by job, the installer and smoke on PC 2; it follows the move, never gates it); the kits (the v5 kit at the pairing, the Intel kit in only with the Arc fingerprint equal, else out with the crossing time on the row); the card after the Windows entry. The gate set before the file goes: every box suite on the pin, the two canary sets with the mixed-version refusal, the fast-time SUMMARY on the shipped pair, the kaspa-pow pairing read-back, the app crate gate and pre-push on the app tip, the pack-gate line read back on every reachable box, F8 if the pairing moved off 1c420786, the F9/F1 interim at the minute minus five if F8 was rerun. The move's mechanics from today's lessons: the puller takes the pair's miner sha from the move file (the fleet's puller fix), a box with no running box-dn3.sh restarts from a quoted environment (the nine-node fault of 10:05, the fleet's third known-failed shape), build-1's three by process name with the old process's locks released first. Open: the drift reading's commit (not a consensus field by its description); the evening minute from the shipper the moment the pin is green. CARD-IN READY (the hash lane, 10:1x UK; tools/ca3-v4-amend/pc-card-in.ps1 at 3566ecfe): both known-failed shapes recorded on PC 1 (the baseline of 4 cards; "no new card" in 1 s); a relay "in" with the PC publishes one job (55-minute cap) giving the card's key, VRAM and dataset fit, the v5 and v4 fingerprints through the OpenCL kit on every vendor plus the CUDA sub-version 3 row on NVIDIA, the rate with all three power fields, the lock grid through the helper on NVIDIA (300 MHz steps from the maximum, stop at a 3 percent fall) and measure-only rows on AMD and Intel, the app's own row, the bench-log and miner-bench.json rows as RESULT ROW lines, the restore and "next". The Ember tiers' engine half on ember-tiers-25 at 91406944 (local; the push on the box test build's green by 10:45). The Arc re-read's default: 10:50 UK unless the shipper clears PC 2 earlier. MAIN'S WORD ON THE 0.3.25 PLAN (10:1x BST): it runs as written, one addition to the app line: the three-tier Ember Tune, both halves (the hash lane's engine fields and the apply Cmd on ember-tiers-25; the UI lane's tier buttons with rate, watts and the daily saving, sweep on by default at balanced, per-card wired), gated on 0.3.25 before the cut; if either half is not green by 19:00 BST the cut goes without it and the tiers ride 0.3.26, stated in the record; everything else stands, the silence-means-go at 17:00 and the shipper's minute; two readings to main: one when the pin is green, one at the minute. THE FOUNDER'S WORD AT 10:2x BST: push 0.3.25 everywhere as soon as possible; the plan stands in every mechanic, the clock moves: the cut goes the moment its inputs are green, not tonight. The targets: the node line placeable 11:30; the app line assembled by 12:30 (the AMD knob and exe, the hash text fixes, the Intel header and the rebuilt kit worker, the tiers if both halves are green by 12:30, else they ride 0.3.26 and the record says so); the pin green by 13:00; the move at the last FETCHED plus ten but never before the class v5 crossing at 68,400 (about 12:54) has been read clean by the node lane, so the earliest minute about 13:30; Mac and Hive at the minute, Windows behind it within the hour, the card after; the pairing default 1c420786 unless the Arc fingerprint reads equal by 12:30; the defaults and the slide authority stand; main's silence past any of these clocks means go. THE TIERS' UI HALF (the UI lane, 10:52 BST): branch tiers-25 off release-0.3.24 a4c5a855 = the UI commit e6571f60 plus the merge of the hash lane's ember-tiers-25 3408db40 (d3d0704a); the UI tests known-failed first then 73 green on build-2; mock captures of the three states (the measured 5090 and 5080 at Balanced; the install's first minutes with nothing measured and Ember Tune on at Balanced; the M5 Max with no lever as Stock alone with the reason) under ~/Desktop/igneum-previews-2026-10-08/tiers/; the app crate gate and the full pre-push gate running on the merged tip, the gated tip by about 11:15, inside the 12:30 default; tiers-25 fast-forwards onto release-0.3.25 when the shipper opens it from a4c5a855; the live tier numbers come from the engine's own search, not from any table. THE BUILD-SERVER LANE'S CLOCKS (10:1x BST): the 0.3.25 pairs the moment the pin is named (the start script parameterised on the pin); the publisher's digest gate (publish-manifest.sh --node-bin, --network-digest, --move-clock; tools/digest-read.sh) landing on master before 12:30 and riding the app line (the alias assertion not its own); the telemetry exe's --tune read-back on PC 1 DONE at 09:07Z (the 9070 XT tune line: gmax 0 range -500 to +1000, plimit 0 range -30 to +10, factory 1); the second master-only deploy started 10:15 BST on master's tip. A FAULT: PC 2's 0.3.24 Windows installer failed at ISCC because release-0.3.24's .iss still carries the TDateTime line the 0.3.23 fix removed; the one-line fix with the shipper and the update-return lane, the republish on their tip (the Windows entry's default: it follows the move, never gates it). A SPEND TO SURFACE: two new Hetzner boxes provisioning (build-3 HEL1 32 threads, build-4 FSN1 96 threads, in the pool by 10:45), reported by the build-server lane; ordered on the founder's own word in chat ("re order", about 09:5x BST, after he added the credit himself; main clicked the order in his Chrome profile); the standing rule on purchases held; they stay. SITE DEPLOYED AGAIN (the build-server lane, master f98e8e7c at 09:15:33Z, 10:15 BST, on igneum.network and igneum.com; the checks ok): the tip carries ae8836f8 (row 17's floor sentence, the Arc row restored to its measurement) and the record through the twenty-eighth landing; the served state now carries every served change of the night and morning. THE 0.3.25 NODE LINE PLACEABLE (the node lane, 10:1x BST, ahead of 11:30): release-0.3.25-node = c6629572 on both mirrors (the pin 5b673577 plus igneum-miner keygen and the proving-fee ceiling switch coded and never set), pairing igneum-pow 1c420786; every gate green at 09:16:28Z (build 09:13Z rc 0, igneumd 3fadca49..., /srv/artefacts/0325-c6629572/node-lane; consensus 134, pow 19, miner 29 with the keygen test, p2p-flows 38, exec 48, core 177 at gate priority after a first run on a stale file on the box); the Devnet 3 canary (09:13:25Z to 09:15:04Z): digest cc902690 unchanged, byte 6, the override refused, two empty nodes handshaking, the shared-devnet dialler rejected, and the 0.3.24 pin's node handshaking with this build both ways, so the mixed fleet runs through the placement; the testnet canary b2e856ed unchanged. The keygen read-back from the artefact printed an address and a key (the key elided in every transcript and record; a printed private key never enters a message, a log the relay carries, or this file); the fleet writes keyed wallets from it. The defaults: the line's tip at 13:30 BST is c6629572 plus the drift reading's commit only if both nodes' logs reach the node lane by 12:30, else without it; the ceiling-switch field set in the 0.3.25 object from the shipper's minute by the one-go script (the digest moves then; the hold's second day under the ceiling, as main ruled; a re-cut without asking under a 15-minute margin); the crossing line the moment the DAA passes 68,400, a red first; the TESTNET_PARAMS v5-at-0 re-cut at 13:30 unless main says otherwise by 13:15. THE AMD KNOB'S GATED TIP MOVED (the update-return lane, 10:15 BST): amd-clock-25 e2962b89 (full gate GREEN 60, the box suite 298 green) in place of cf8444bf, with the shipper; from the exe's read-back on PC 1: the integrated Radeon's tune line carries every range as a dash and the knob had read it as an offset knob with a one-MHz ladder; it now reads "not available (the driver exposes no tuning interface for this card)", and the 9070 XT's real line (gmax 0, range -500 to 1000; plimit 0, range -30 to 10; stock 3,292 MHz under load) is the test's second half: the ladder 3,192 down to 2,792, the power 70 to 110 percent, offsets on the apply; the grid by 11:20, the efficient point into EFFICIENT_W before 12:30 or the declared ladder ships. THE MOVE'S READ-BACK (the fleet lane, late against its 10:20 minute): APPLIED on the relay at 09:07Z: 24 MATCH by the puller (igneumd 2.1.0-5b673577, digest cc9026909eddbadb, synced; dn3-g1 at peers 24), p1-3080 on cc902690 by 09:10Z; 2 FAILED (dn3-r01, dn3-r02: no saved environment, hand-started yesterday) moved by hand at 09:10:27Z; 9 MISMATCH with no node after the puller's restart (hub-1, dn3-g2, dn3-q04, dn3-q05, dn3-r04, dn3-p02, dn3-p04, dn3-p05, dn3-relay): the saved environment line NET_ARGS=--devnet --devnet-suffix=3 unquoted, so sourcing it ran "--devnet-suffix=3" as a command and the start never reached box-dn3.sh; all nine moved by hand 09:11:58Z to 09:12:24Z with every value quoted, the puller now quoting every value (redeployed 09:16Z on 34 boxes); so 36 of 36 fetched are on 5b673577 and cc902690 by 09:12:24Z (10:12 BST). The first lock on cc902690: checkpoint 1931, block 63510971..., blue score 57,930, at 09:06:32Z on dn3-g1 (4,803 signed, 69.8 percent of active, 66.7 of total); hub-1 logged the same checkpoint at 09:11:43Z after its hand restart and checkpoint 1944 (blue 58,321) at 09:12:39Z. The chain rate: hub-1 read 0 blocks a minute at 09:07Z because hub-1 was one of the nine down; from 09:12Z the tip moves at about 0.4 chain blocks a second as before, and paidShards moves again (11,821, frozen since 03:32Z, to 12,012 at 09:19Z, pool entries 47): carrying resumed with the move, the node lane's one-shot-gossip class confirmed. The proven share at 09:19Z 0.465 cumulative (the hour's own 0.000, the hour being the move); the proving fee 10,000 gwei per pgas last, 50,566 max over 60 blocks (1.0x and 5.1x the floor), the field now on the hourly line. The unfetched: dn3-agg48 (the L40S in its bring-up, applying at its first tick), p2-3090-1 (ssh refused since 21:48Z yesterday, on 2720d8d2 with 4 old-digest peers), p2-4090-1b (its Vast proxy dead, its node down); dn3-relay fetched at 08:48Z and is on cc902690. The eight "bc5945fe" boxes: no such binary (that sha was the reader's own shell); those boxes had no node at all (dn3-g2 dead since 22:49Z, dn3-g1 since 00:46Z, the others overnight, no panic or OOM on any), restarted 08:43Z to 08:53Z, took the move with the rest, and mine where they mine. The keyed-wallet write not started (the 0325 artefact's first mention to the lane at 10:20; box by box after the launch fleet's first boxes are up; the rent running since 09:16Z). p1-5090's drift reads offset -5 again at 09:21Z; hub-1's numbering against build-1's node the next read. Three fault classes for the record from one move: the unquoted environment line (fixed in the puller), the two hand-started boxes with no saved environment, and the eight boxes that had silently lost their nodes overnight with no panic (a watch for a node absent while its box is up is the fleet's next check). THE TIERS GATED FOR THE CUT (the UI lane, 10:20 BST by the Mac's clock): tiers-25 at d3d0704a on the mirror (the UI commit e6571f60 plus the engine half 3408db40 merged, both off release-0.3.24 a4c5a855, a fast-forward onto release-0.3.25): the app crate gate GREEN 299 + 35 + 8 on build-2, the full pre-push GREEN 60 checks with the stamp, the UI tests 73 green known-failed first, the push gate GREEN; the captures under ~/Desktop/igneum-previews-2026-10-08/tiers/; sent to the shipper; two hours inside the 12:30 default; a rebase and re-gate inside the hour if 0.3.25 opens from a later tip. Both halves of the three-tier Ember Tune are in the cut. THE 0.3.25 APP TIP (the shipper, 10:29 BST, two hours ahead of the 12:30 target): e0d4425f on release-0.3.25 (the box gate green): amd-clock-25 e2962b89, tiers-25 d3d0704a (both halves), the Intel header via 9088293a, the node-source pin c6629572; the node pin candidate c6629572 with the digest cc902690 unchanged; the cut list r0325-cut-list.md: the pin named by 13:00, the move no earlier than 13:30 after the 68,400 crossing reads clean; the pairing 1c420786 unless the Arc reads equal by 12:30, the Intel kit on that read. THE 0.3.25 NODE LINE'S TIP MOVED (the node lane, 7bd2940f on both mirrors at 09:24:03Z, every gate green at 09:29:48Z): c6629572 plus the one-shot gossip fix (unpaid proof records re-announced every 120 s; the class confirmed on the live chain after the 09:05Z move); nothing consensus, the Devnet 3 digest cc902690 unchanged on its canary, the 0.3.24 pin's node handshaking both ways, the testnet digest unchanged; build 09:26Z rc 0 (igneumd 16dee9f1..., /srv/artefacts/0325-7bd2940f/node-lane), exec 49, pow 19, core 177, p2p-flows 38, miner 29, consensus 134 at gate priority; it replaces c6629572 as the placeable keygen build and as the tip the ceiling-field cut lands on; the shipper has the line. The drift item is off this line: the fleet's reads were shared-devnet reads (hub-1's node on 26790 at chain block about 190,900; Devnet 3 at 25,900; both answering chain id 4463 below the floor), p1-5090 a shared-devnet prover, and the three numberings at one hash are the snapshot-inherited class (build-1's node1 itself resumed from a snapshot); the fleet rents a fresh-walk node under its standing ceiling to settle which numbering is right, hub-1's restart held until then, the loader change (re-number the resumed range against the DAG) after that read. THE 0.3.25 PAIRS ON 7bd2940f (the build-server lane, from 10:33:11 BST on build-1 under lease class release, /srv/artefacts/0325-7bd2940f/: the seed about 10:36, the Windows pair about 10:38, the hive package with the three kit zips about 10:41, each minute to the shipper and the coordinator); the c6629572 pairs already built (seed f913e3e7, win 42d0dd57, hive 14d86245) stand in their own folder and are not the cut; the publisher's digest gate on master since 10:17, riding the 0.3.25 app line. THE RE-POINTED APP TIP (the shipper): 92f004f1 on release-0.3.25 (e0d4425f plus the node-source pin to 7bd2940f), the push gate GREEN at 10:32 BST, the box gate GREEN at 10:33:25 (303 + 35 + 8); the cut list's pin candidate 7bd2940f; the kit re-cut from 92f004f1 and the pairs on 7bd2940f's artefact with the build-server lane; the Mac node pair and the DMG rebuilding on 7bd2940f under the lock from 10:32:31; the 13:00 pin and the 13:30 earliest minute standing. The 0.3.25 inputs are all green at 10:33 bar the pin's own gate set and the crossing. A SWEEP FINDING FROM MAIN (10:4x BST): on a rented, power-capped RTX A4000 (114 W cap) class v5 reads 26.0 MH/s against v4's 31.4, 17 percent under, the fingerprint equal; the A100 1.3 percent under; every uncapped consumer card level: v5 costs more compute per hash and a compute-limited card pays, which is what a knee lock makes of a card. Two orders with readings by 12:30: (1) the hash lane sends the 5090's v5 pack rows at the 1,300 lock against v4 at the same lock, and the 5080's if they exist; if v5 at the knee loses more than 2 percent, the knee is re-found under v5 and the tiers table says so; (2) the tiers' engine half: a class change (the chain's program class flipping) invalidates the stored tiers and re-runs the search within ten minutes of the crossing, the first-run line saying why; known-failed first (tiers stored under v4 must read "re-measuring for class v5" after the flip, never apply as if current); on 0.3.25 if it fits by the cut, else 0.3.26 with the record saying the v4 tiers may be off by the measured percentage until the re-tune. Per tier: a locked card may lose a few percent of rate at the class v5 crossing until Ember re-tunes; the number is the 5090 row. THE ORDERS PLACED (the coordinator, 10:4x BST): the hash lane's two readings by 12:30 (the 5090's v5 rows at the 1,300 lock against v4 at the same lock, the 5080's if they exist; the knee re-found under v5 if the loss is over 2 percent; the default if the PC 1 queue cannot run it: the A4000's 17 percent stated for a capped card and "unmeasured at the knee on the 5090"; and ember-tiers-25's class key: a class change invalidates the stored tiers and re-runs the search within ten minutes, known-failed first), the UI lane's class-flip state ("re-measuring for class v5", v4 tiers never applied as current after the flip) and knee note by 12:30, the shipper's cut list carrying both on 0.3.25 only if green by the pin at 13:00, else 0.3.26 with the record's sentence that the v4 tiers may be off by the measured percentage until the re-tune. THE 0.3.25 PAIRS ON build-1 (the build-server lane, /srv/artefacts/0325-7bd2940f/): the seed pair at 10:34:44 BST (igneumd c7fc542b, igneum-miner 4494ecc4, glibc 2.34), the Windows pair at 10:36:13 (igneumd.exe 5d1dea23, igneum-miner.exe eee7bdfa), the hive package igneum-hive-0.3.25-7bd2940f.tar.gz at 10:37:49 (sha d977797f..., the three kit zips, smoked in ubuntu:20.04); the kit re-cut from 92f004f1 (sha 676240f6, 424,540 B) staged in both folders, the PC 1 host from it bc8d4f79 (in host.sha256 at the shipper's 24680e1d), the 0.3.25 Windows payload from 24680e1d cutting. Every pair of the cut exists by 10:38; the pin's gate set and the crossing are the only waits. FOUR NEW LANES ON THE FOUNDER'S ORDER (11:00 BST, "build all this today to close this gap"), mirrored by the coordinator as the shipper's clocks are: the explorer (a5ef1d5801084005b; explorer.igneum.network by 16:00), the canonical DEX and the Sepolia certificate verifier (a74a8267813d6ea34; the AMM by 14:00, the swap UI by 17:00, the verifier by 20:00), the builder pages, faucet and grants (adb29da59baf27898; /build and /grants by 15:00, the faucet by 16:00), three reference apps that only work on a proven chain (a2060899d2a27d31c; /light by 16:00, /receipt by 18:00, the Sepolia oracle demo by 21:00); the build-server lane stands up rpc.devnet.igneum.network by 12:00; they do not touch the 0.3.25 cut, the crossing or the fleet, sharing the boxes' lease pools (class measure) and the master-only deploy; a lane silent past 25 minutes gets the word from the coordinator and then main. THE FAST-TIME GATE ON THE 0.3.25 PAIR: SUMMARY PASS (cross-0325-39f127a1) at 09:54:40Z (10:54 BST) on the pair 39f127a1 (the node code and object byte for byte e0644958's; igneum-pow at the freeze 1c420786), build-1 under lease pool class v5, 09:42:25Z to 09:54:40Z, every check green (rung 1 by signal at epoch 6, class v5 by signal at byte 6 from epoch 8 at rung 1 at 9,985 bps, the stale node refused, the restart step resynced in 8 s, four sinks equal, 0 PoW rejections); the ceiling's two new fields absent from the 60x file so the ceiling stayed at never there (the node lane's note); to the shipper the same minute; the pin line names e0644958 and its gates. THE FOUNDER'S WORD AT 11:0x BST ("can we add in any more layers? class rotating? things that would render an ASIC useless as soon as it dropped"): the class v6 design opens today as a rotating family, the research lane and the hash lane under the coordinator, the design doc docs/design/class-v6-rotating-family.md by 18:00 BST with the chip-model rows beside each layer (what it does to k and capex for a fixed-function chip and to the per-joule edge for a GPU-like chip; what it costs every GPU tier, Apple included): (1) per-era draws of the class parameters now fixed by release (the mixer round count within the tested margin, the op-mix weights within the measured safe band, the read width, the program length, the shadow placement), drawn from chain state like the program; (2) the state-derived dataset's size tracking chain-state growth with a floor, so fixed-memory silicon ages out; (3) scheduled family epochs by height (every 180 days by default) with no release; (4) the (c''') acceptance floor and the F8-form uniformity test generalised to each era's parameter draw, redraw on failure, so layers 1 and 3 need no per-era cryptanalysis. Per layer: the gate it needs (the family analysed as a family: the attack board's shape over the testnet period), the known-failed test, an honest line on what a fully general chip still gets. No consensus code this week; the document, the numbers and the gate plan. Per tier for the founder tonight: what each layer does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE ARC RE-READ IN ITS CHAIN (the hash lane, 10:57 UK): no clear came from the shipper, so the default ran at 10:50: the rotate-fold kit's fetch (sha 65b47211) published to PC 2 at 10:51:41, the run (run-ca3-pc2-v5-intel-bench-20261008, the v5 lane's script c0d398a1) in the publish chain behind another lane's publish-jobs.sh sign --deploy from the build-server worktree (the publisher serialises); the fingerprint line by 11:15 if the publisher frees inside ten minutes, else the blocking process named by 11:10. Queued on PC 1 behind the same publisher: run-ca3-pc1-v5lock-5090-20261008 (class v5 against v4 at unlocked, 1,300 and 1,200 MHz, the v5 kit's CUDA packs), its rows by 12:30; the AMD grid after it from about 11:25. The class-key work on ember-tiers-25 started; the v6 cost rows by 16:00 taken. THE TIERS' CLASS-FLIP STATE, THE UI HALF (the UI lane, 10:57 BST): tiers-class-25 at d949e274 on the mirror, off release-0.3.25's tip 24680e1d (the shipper having merged tiers-25 d3d0704a into release-0.3.25 at 111dae69), the crate gate GREEN 303 + 35 + 8 on build-1, the full pre-push GREEN 60, the UI tests 74 green known-failed first (the v4 tiers stayed on the buttons after the flip on d3d0704a); after the flip the table reads "re-measuring for class v5" on every button with the start minute or "queued (within ten minutes of the crossing)", the v4 watts never current, the strip's sentence naming the crossing; the knee note under the table when knee_loss_pct is over 2 percent; the captures tiers-flip-dark.png and -light.png; the fields tiers_class, program_class, tiers_remeasure_at, knee_loss_pct (the shape sent to the hash lane at 10:4x; the engine sha by 12:30); the default: the display rides 0.3.25 inert if the engine half is late and lights up on 0.3.26. THE FOUNDER'S WORD AT 11:1x BST: class v6 is DECLARED with the four layers as its spine (per-era parameter draws, the dataset tracking chain state, scheduled family epochs by height, the acceptance floor generalised to parameters), and deep past-and-future research opens now under the coordinator with serious resources ("see if anything can be optimised, added or invented"; reading public research is in-house, nothing paid or asked of anyone outside): four research lanes today, (A) history (every ASIC-resistant proof-of-work and how it fell or held: Ethash and the E3 and Linzhi chips, ProgPoW's review, RandomX and its chip analyses, Cuckoo, Equihash and the Z9, Argon2 and Scrypt and the Litecoin chips, KawPow, Autolykos, Octopus, kHeavyHash's chips; the exact mechanism each chip used and what the design missed, each mapped to Igneum's layers with "does v6 close it" as a sentence and a number), (B) the hardware future five years out (PIM and processing-near-memory, HBM3e and HBM4, LPDDR6, 3D DRAM, CXL memory pools, wafer-scale, chiplets, FPGA with HBM; for each the chip-model k band against a state-sized dataset and dependent random reads, and the one layer that would blunt it), (C) invention (layers beyond the four, each a paragraph, a known-failed test and a chip-model row: data-dependent program graphs, latency-bound dependent reads tied to the shard proof, randomised memory topology per era, VRAM-size ratchets, proof-carrying hashes sampled by the pool, time-locked parameter commitments, and what the lane invents; rejecting what costs GPUs more than chips), (D) the family gate (how a parameter family is cryptanalysed as a family: sampling bounds, coverage, the F8-form and (c''') tests over the parameter space, the attack board's shape over the testnet period, so layers 1, 3 and 4 can be automatic with a proof of what was tested). Resources: all four boxes under lease class measure, PC 1 by job for card rows, the rented fleet for one-shot measurements inside the ceiling. Deliverables: a first synthesis in docs/design/class-v6-rotating-family.md by 20:00 BST (the four layers priced, every finding from A to D with its number, a ranked list of what v6 adds beyond the four, the honest line on what a fully general chip still gets), the full report by 09:00 tomorrow, one line to main per lane as each lands; per tier at 20:00: what v6 does to a chip on its release day and what it costs a 5090, a 5070 Ti and an M5 Max. THE 0.3.25 APP TIP AND PIN CANDIDATE (the shipper, 10:58 BST): the app tip 9b93e649 (push gate GREEN; the crate unchanged from e0d4425f; the node-source pin to e0644958 and host.sha256 bc8d4f79); the pin candidate the node lane's ceiling cut e0644958 (digest 1b37cb9d, every gate green 10:53, the fast-time SUMMARY PASS 10:54, the floor at DAA 82,800 about 16:53 BST, a publish up to 14:53 without a second cut); the tiers' class-key halves: the UI lane's tiers-class-25 d949e274 green and inert alone, merged with the hash lane's engine sha the moment it lands (12:30), gated as a pair on the release tip, riding only if green by the 13:00 pin; the 0.3.24 Windows take 2 failed at a new place (Inno stopped the app and copied nothing); the update-return lane owns the fix on release-0.3.25 by 12:30, the default the 0.3.25 Windows entry waiting for a clean take 3 while Mac and HiveOS move at the minute. THE FOUR CLASS V6 RESEARCH LANES SPAWNED (the coordinator, 11:0x BST, each with its worktree, its box resources under lease class measure, its clocks and the rules): lane A history (a603a938582c43ab5; the first cut docs/analysis/class-v6/history.md by 15:00), lane B the hardware future (a4f73e2a6f2d1b757; hardware-future.md by 16:00), lane C invention (a5dfe95ee8c47cd0f; invention.md by 17:00), lane D the family gate (a07a99a3788566af2; family-gate.md by 17:00); each feeds the research lane's synthesis docs/design/class-v6-rotating-family.md by 20:00 (its outline by 13:00; the hash lane's per-tier rows by 16:00); the full reports by 09:00 tomorrow; the coordinator's lane mirror carries their clocks. A HELD PUSH AND ITS CAUSE (11:00 BST): the hash lane's push of ca3-v4-amend was refused at 10:58 by the gh-account hook reading the founder's gh (his personal login active again; nothing switched by any lane); the cause is the branch's own hook, which predates the per-process fix (34b0884d): the hook runs the branch's tools/ci, so every branch older than 09:58 must merge the mirror's master before its next push, under which the check reads Igneum's own gh directory and skips under the suspension marker; the rule to every lane. Live: the Arc re-read on PC 2 (published 10:59:47) and the v5lock job on PC 1 (published 10:53, about 12 minutes). THE CLASS V6 OUTLINE ON THE MIRROR (the research lane, docs/design/class-v6-rotating-family.md on counter-asic-4, the commit after fb61ed4b, pushed 10:5x UTC, two hours ahead of 13:00): section 0 the founder's table (per layer, what it does to a fixed-function chip and to a GPU-like chip on its release day, and the 5090, 5070 Ti and M5 Max columns, measured where the night's rows exist, the 5070 Ti scaled until the hash lane's row); the honest frame on top: the four layers render a FIXED-FUNCTION chip useless on the first era its wired value leaves (one tape-out lives one era) and move nothing for the stored-dataset chip with a programmable core except the core's size and the N5 project it forces; that chip keeps 3.6x at zero premium and 2.1x at k = 1 on a 5090 at its knee. The layer table (sections 1 and 2) names the bands each draw takes and the measured rows that set them: the mixer in {4, 8, 16} (x16 open), the op-mix weights within B = 4 with shuffle and mulhi capped (shfl 55.8 pJ per op), the read width in {1, 4} words (w64 excluded by the 5 October rows), the block shape 64 to 256 (never 1,024), N left to the ladder's signal (an unconditional draw retires the Apple tier at 200,000). Open numbers asked of the hash lane with defaults at 16:00: the 5070 Ti row (the rented 5070 scaled), the x16 mixer's verifier and build (the chip model's estimate), two re-weighted shadow packs for the op-mix band (the microbench arithmetic). Layers 2 to 4 and the gate plan are skeletons with their sources named, filling by 18:00 with the four research lanes' cuts, the synthesis by 20:00. THE FOUNDER'S WORD AT 11:2x BST ("all builders are idle, load them up"): build-1 to build-4 filled now and kept above 80 percent all day under the lease pool, class measure behind the release gates, in this order of value: (1) the class v6 family gate's sampling runs for lane D (the F8-form census and the (c''') floor over the parameter bands: the mixer {4, 8, 16}, the op-mix weights within B = 4 with shuffle and mulhi capped, the read width {1, 4}, the block 64 to 256; thousands of drawn eras, the uniformity and bucket tests on each, so the family document carries measured coverage tonight); (2) the attack families at scale on the 0.3.25 pin candidate's igneum-pow (F8 to 256 seeds, F9 and F1 to 10^6 on the frozen 1c420786, the day-key scan to 2^28) as the record's strengthening lines; (3) the invention lane's candidate layers measured as packs as fast as it writes them; (4) the full suite matrix of the 0.3.25 pin on every box as the pre-pin check; (5) the Windows and hive cross builds and the sweep's reruns; the lease tool's pre-emption giving release-class work the cores when the pin's gates need them; one line to main at 12:00 with the load on each box and what runs there, then hourly only if a box drops idle. THE DRIFT CLASS SETTLED (the node lane, from the fleet's fresh-walk node, a shared-devnet node synced from an empty datadir to 191,441 chain blocks at 10:03:45Z): hub-1 and five standing boxes number the fresh chain exactly; seven boxes carry numbering inherited from an exec snapshot taken on a chain that later re-walked (+2: p1-4090, p1-a5000, pool-1, build-1's node1; +3: p2-3090-2; +4: p2-3090-4; +5: p1-5090 and p2-3090-3), and a restart on the kept datadir does not re-walk (p1-5090 at 09:22Z stayed +5); the cost: a prover on drifted numbering signs statements the hub vetoes, so the seven earn nothing from proving until they re-walk, the drift refusal stopping the waste. The node lane's word to the fleet: p1-5090 first, both snapshot files moved aside so the executor re-walks from the DAG, the re-walk timed and read against the fresh node, then the other six in series, hub-1 untouched, build-1's node1 after the 12:40Z move; if the re-walk reads over two hours the six wait for the node-side fix on the next node line (the loader re-numbering a resumed range against the DAG before serving). THE 0.3.25 PIN CANDIDATE CONFIRMED (the node lane): e0644958 on both mirrors (keygen, the re-announce, the ceiling switch at 82,800 in the Devnet 3 object, digest 1b37cb9d, the 0.3.24 pin refused both ways), every gate green at 09:53:01Z, the fast-time SUMMARY PASS at 09:54:40Z on the same object; the publish ceiling DAA 75,600 (14:53 BST); waiting only on the 68,400 crossing reading clean (about 12:54; the node lane's line the moment the DAA passes it); the shipper names the pin at 13:00; the TESTNET_PARAMS re-cut at 13:30 unless main says otherwise by 13:15. LANE C'S FIRST PACK (the invention lane, 11:0x BST by the Mac's clock; its own line read "12:1x", a clock to correct): build-1 takes the igneum-pow build from counter-asic-4 at 5984ffab, then the per-load shadow in its sound form (mx8+shl6912x1: 16 sub-blocks of 432, one pass, the form 20.2a named and never drew) as the first candidate: the acceptance census over 64 seeds and 16 drawn eras against the 16x27 form and the class v4 shape, the pack export, the F8 read at 2^24 and the verifier bench on a leased core, the first read by 13:30; build-2 next for the second candidate (warp-uniform data-dependent block selection); the candidates with no pack form (the VDF commitment, the VRAM ratchet, the pool-sampled witness, the state-tied reads) stay modelled and the 17:00 cut says so; the worktree igneum-wt-v6-invention on class-v6-invention. THE WINDOWS INSTALLER CLASS AND THE 0.3.25 TIP (the shipper, 11:08 BST): the app tip 139c147a (9b93e649 plus install-detach-25 52a34111, packaging/windows and tools/ci only, the crate unchanged; push gate GREEN); the 0.3.24 take 2 class: an installer started under the app's job runner is a child of the engine, and the engine's kill_tree on quit ended it between PrepareToInstall and the copy; the fix re-launches the installer as a one-shot scheduled task outside the job's tree; the 0.3.24 Windows entry skipped; the rule-14 take on PC 2 is the 0.3.25 installer over the running 0.3.21 app, queued ahead of the Arc re-read; the pin candidate e0644958, the DMG 501ba293 staged, the 13:00 pin and the 13:40 provisional minute standing. THE ATTACK FAMILIES AT SCALE (the attack-pass lane, cores held at 11:08 BST, every run under lease pool class measure): box 2 (88 cores): F8 seeds p66 to p257 (192 new, 256 with the gate's p2 to p65) at 2^24 on class v5 at the freeze 1c420786 (the gated binary 0f5c98dc, pairing e5a4ac5978462156; the leaves re-run on 8f481459 if the kit pairing flips), the window-model control, by-site, as three thirds of 64 seeds; box 4 (80 cores held, 16 asked): F9 to 10^6 on 1c420786 (seeds 100,000 to 999,999 in six chunks of 150,000 at 8 threads, four running), F1 to 10^6 class v5 programs on 1c420786 (one census at 40 threads with the progress line and flushed partials, re-drawing the record's first 10^5 on the way as a reproduction check), the F4 day-key scan to 2^28 on 8ca66afa's redraw rule at 8 threads; nothing on build-1 or build-3 (lane D's); the projections: F4 about 2 to 3 hours, F8's 192 seeds about 9 hours, F9's 900,000 and F1's 10^6 about 30 hours each, so the 17:00 default is partials for those two with the lane (d) rows carrying counts so far; any pre-emption by release-class work reported. THE RPC AND THE 0.3.25 PAIRS ON THE PIN CANDIDATE (the build-server lane, 11:0x BST): rpc.devnet.igneum.network up since 11:08 BST (the first of the founder's builder clocks, 52 minutes ahead); the 0.3.25 pairs on e0644958 running on build-1 since 11:06 (the app tip 139c147a), the Windows and hive crosses on build-2, build-3 and build-4 as reproducibility rows at class release by about 12:40; the sweep reruns' list not held by the lane, the default at 12:30: last night's sweep logs on build-1 read for rows that ended without a result line and those rerun at class measure. THE SWEEP RERUNS' LIST (the fleet lane to the build-server lane, 11:1x BST): the fleet ran nothing under the build boxes' lease pool last night (every fleet bench a rented GPU one-shot), so the rows the lease kills cut short are the hash and class lanes' and the build-server lane's default read on build-1 is the right one; the fleet's own rows without a result (A10, A40, A100 40 GB, H100 NVL, H100 PCIe, MI250, RTX 3050, RX 7800 XT, 7900 XT, 7900 XTX, 6900 XT) are provider gaps needing a GPU host, rerun the moment a provider lists one. THE CLASS-FLIP TIERS, BOTH HALVES (the UI lane, 11:13 BST by the Mac's clock, 1 h 47 min inside the 13:00 pin): tiers-class-25 at 081b3ba7 (the display d949e274 plus the hash lane's ember-tiers-25 0a838072, on release-0.3.25's 9b93e649; the field names matched exactly): the crate gate GREEN 305 + 35 + 8 on build-1, the pre-push GREEN 60, the UI tests 78 green known-failed first, the push gate GREEN; with the shipper. A RED ON THE RELEASE TIP, for the shipper and the update-return lane: release-0.3.25's 139c147a is red on one crate test (ota::return_tests::no_relaunch_while_an_installer_runs_and_a_relaunch_when_it_clears, 304 of 305): install-detach's 0c588b09 reshaped the installer's clear step into a multi-line block while the test asserts the one-line literal at app/igneum-app/src/ota.rs:1348; 9b93e649 passes; the fix is the test's literal on the install-detach line; the UI lane built on 9b93e649 so its tip is green alone. A RED FROM THE PIN MATRIX (the CI steward, 11:13 UK): the core suite fails on the pair (the node e0644958 with the app tree 9b93e649): config::params::tests::fast_time_60x_file_is_the_devnet_at_60x panics "override-60x.json lacks the field base_unit_decimals"; the field was added by 0e4ec18a on ca3-v4-node yesterday at 21:45 UK and reached neither master, release-0.3.25 nor the app tip while the node line's test demands it; so every box reads red on core, and the fix is one line on release-0.3.25 (the cherry-pick of 0e4ec18a, or "base_unit_decimals": 8 in infra/fast-time/override-60x.json); sent to the shipper; green so far pow and app on build-1 and build-3; the two new boxes' toolchains read the same as build-1 (Ubuntu 24.04.5, glibc 2.39, the pinned rustc, sccache and lease, no nvcc). The founder's fourth load item paid in its first ten minutes: a red no single-box gate had read. MAIN'S WORD ON THE TWO REDS (11:1x BST): the install-detach fix belongs in 0.3.25 if it can make it, since a Windows install by any path that lets the engine's job runner kill the installer mid-copy is the plug-tune-play fault class (an update a user repairs by hand); the default order: the update-return lane fixes the ota.rs literal by 12:00; if 139c147a plus the fix is green on the crate gate by 12:15 the cut goes from it, else from 9b93e649 with the detach on 0.3.26 and the record saying Windows installs by job stay unreliable until then; the missing 60x field: the CI steward lands the one-line field on master and the release line by 11:45; the pin slides under the shipper's authority inside 14:53. THE DAY-KEY SCAN TO 2^28 (the attack-pass lane; class-v5 8ca66afa's redraw rule, build-4 under lease pool 8 class measure, 379.2 s, census-2p28.md at 10:15Z, 11:15 BST): days with any gain over 1.1x: 0 of 268,435,456 on M1 (median 226), 0 against the mean, 0 on M2, 0 on ROT and RC; the M1 cost mean 225.791, sd 6.073, min 206 (day 27,016 at 1.0971x, the redraw rule's floor: no day under 206 in 2^28), max 258; every weak class on its analytic expectation (ROT any pair summing to 32: 160,354,008 against 161,256,979; RC any zero: 1 against 1.0, at cost 234, no gain; RC with rk = 0: 87 against 72, 1.8 sigma; the two cells under expectation the rule's own refusals). PASS: no chip buys a weak day in the first 735,000 years of days; at most 1.097x on the best day. The row and f4-weakday.md section 10 committed on attack-pass at eabb4b0e, the push held by the branch's old hook (the fix: merge master, under which the check reads Igneum's own gh directory and skips under the suspension marker). THE SWEEP RERUNS' READ (the build-server lane, 11:18 BST): build-1's records hold no hash-lane or v5-lane run the pool cut short (preempt.log: three TERMs all night, every one to an adv-class holder pre-empted by a release gate, not reruns by rule; no reaped.log; builds.jsonl for 18:00Z to 09:00Z 150 rows with no signal end, the non-zero rows the fast-time gate's designed failed cases and build errors; the census and fingerprint suites leaving no builds.jsonl row and no output directory ending without its result); live at 11:17Z the family gate's v5_attempts_census holding 24 cores on build-1 at class measure; the default at 12:30 if neither lane names a run: no reruns, the boxes carrying the e0644958 reproducibility crosses (build-3's Windows pair already read: igneumd.exe 4b0c3aeb, igneum-miner.exe b3da4088) and the gates. The founder's fifth load item is therefore the crosses, not reruns. The attack-pass branch merged master and pushed (460fd9fa, the F4 2^28 row and f4-weakday.md section 10 on the mirror; the hook skipping the gh read with its suspended line; nothing switched). THE FAMILY GATE'S FIRST COVERAGE (lane D, 11:2x BST by the Mac's clock; its own line read "11:3x"): the harness live on build-1 under lease pool class measure (scripts and pinned binaries under /srv/builds/_adv-family-gate/): (1) the base control v5_attempts_census on the shipped class v5 draw over f8-label seeds 1,000 to 11,000, 24 cores since 11:16; (2) the family harness family_gate_era_census (branch family-gate-v5 = class-v5 8f481459 plus the harness, never a chain path; the acceptance keyed on the family's shapes behind IGNEUM_FAMILY_GATE): one drawn era per seed, every layer-1 parameter from the era's own stream (the shadow block {64, 128, 256} x {108, 54, 27}, the mixer {4, 8, 16} recorded, the read width over {1, 4} words, the ten weights within B = 4 with shfl and mulhi never raised), the chain draw through the real rule with every candidate's first failing part, then on the accepted program at the rule's own 2^20 sample the (c'')/(c''') ratio, the largest 256-item bucket per site (ratio and sigma), the index-bit bias per site in sigma; the 16-era smoke run PASSED at 11:21 (about 10 core-seconds per era; 10,000 eras about 28 core-hours). THE WORST READINGS IN THE 16: (a) the index-bit bias read fires HARD on 7 of 16 eras, |z| 130 to 511 at one site, every one at address bit R (the era's stride rotation) or R+1 (era 15 with R = 25 bit 25 z -511 at P(bit) 0.25, a product's bit 0; era 7 R = 17 z -468; era 5 R = 26 z -440; era 13 R = 1 bit 2 z -255, a product's bit 1 at 3/8; era 1 R = 6 z -224; era 12 R = 5 bit 6 z -130; era 6 R = 18 z +256, an or-shaped source at 5/8), the other 9 under |z| 3.8: adv-cache-2's era-stride class measured at the acceptance's own sample on class v5 accepted programs: not diffuse at the bit level, a 25 percent bias on one address bit of one site in about 40 percent of drawn eras, which (c''') does not see (min ratios 0.9954 to 1.0000); a chip holding the favoured half of that site's window serves 75 percent of its reads instead of 50, about 1.6 percent of a hash's reads at f = 1/2 for one site, which does not move the f = 1 verdict but is an auditor's flag on "uniform random reads"; the lever is load_index's form (fold the product's low bits before the rotation), not a floor (a 6-sigma refusal would redraw about 40 percent of epochs): a class v6 design row. (b) The (c'') ratio min 0.9954 (era 7), the rest 0.9965 to 1.0000. (c) Attempts: 14 of 16 accepted at attempt 0 or 1; era 9 (shape 64, width 4, mul 11 and or 8 of 75) took 24 candidates: the lossy corner raises r, the exhaustion number to read per stratum. (d) The largest 256-item bucket: ratios 2.2 to 2.4 at full-window sites are the CLEAN maximum (65,536 Poisson(16) buckets, +4.4 sigma), so the F8-tail bound must be stated in sigma, not ratio (the sigma column in the rebuild). Next: the random stratum (10,000 eras) and the corner strata (the lossy cap, width 4, shape 64, 3,000 each) on build-1's free 64 cores, then build-3 and build-4; the first cut of family-gate.md drafted, the measured coverage table in at 16:xx for the 17:00 cut. THE OTA TEST LITERAL FIXED (the update-return lane, 11:23 BST, ahead of both clocks): ota-test-25 off release-0.3.25 139c147a, tip 53cb2f73 on the mirror, one test-only commit (the test reading the installer's clear step as the begin/end block the detach made it; the detach's behaviour kept), the box 2 crate suite 303 + 35 + 8 passed, 0 failed, the full gate GREEN 60; the shipper's cut tip 139c147a plus this commit, so the install-detach rides 0.3.25 and the PC 2 rule-14 take runs on it. A SHARED-DEVNET FACT FROM THE FLEET (not this lane's, with the shipper and the infra lane): the Hetzner live seed 188.245.5.161:26611 is still on the old override object (digest eada4bda) 1 h 40 min after the 0.3.20 sweep (the fleet never touches Hetzner nodes, so it was outside the sweep); the 0.3.21 wipe canary c22-1 took five digest-mismatch rejects from it; an app with the packaged peers is refused at the seed and syncs through node1 and the hub only, a fresh joiner with only the seed cannot join, the 14 voters and the hub are unaffected; the owner puts the floor file ov16-floor-900000.json (sha 294f1f80) and the c4459193 pin on it. 0.3.21's STAGING (the node lane): the order dry-merges onto 55768f88 with nothing moving to 0.3.22; the late-join fix is 52e96c94 (70e4601e rebased onto 55768f88, exec suite 33 green with both new tests); f067f7c1, b0444f51 and 437f0438 merge clean in order; 2e32d5f6's one conflict (DST_ADDRESS beside pool-finish's DST_BINDING in consensus/core/src/finality.rs) kept both; the live-file digest eada4bda after each (every switch at never); the staging waits on the shipper's sweep-end word; the re-pin held. PC 2 DOWN AGAIN (main, 16:5x UK): the founder takes PC 2 down for cable work (PC 1 back but his desk); both PCs out of the sweep's waves, each updates on its poller on return; no PC job to PC 1; the Windows G1 completed before the outage, nothing reruns. 0.3.21's SECOND GATE LINE on 55768f88 (sha256 279b1b690e854fc9): the ten-minute mixed-version gate beside the 5899f603 pair, 13:37:40Z to 13:47:52Z, SUMMARY PASS (one digest b0afb2ee on five nodes; 223 new and 381 old blocks accepted by the old hub, 0 rejected; counts equal at 319, 486 and 604 through both clean joins and the restart step at 13:45:22Z; no panic); the node lane's two lines on 0.3.21's first candidate complete, in plan 6.9 on ca3-v4-node; the fleet's set on it (the bare-child 12 GB line, the wipe, the kept read, the cases) is the fleet's. 0.3.21's FIRST GATE LINE on 55768f88 (sha256 279b1b690e854fc9, the string read back; pairing igneum-pow 8c728ca3 at byte 5): the digest gate 13:35:41Z to 13:37:19Z SUMMARY PASS (a89be8a7 on both binaries with the peers; db9a85f9 refused, no peer; the live file's eada4bda unmoved); the ten-minute mixed-version gate from 13:37:40Z, line about 13:50Z. The 0.3.21 order as the shipper sent it: 55768f88; f067f7c1 and 70e4601e; b0444f51; 6eb21fc9; db28d331; then the re-pin from 8bdcbdd8 on the coordinator's word; suites between, the digest read after every one; the mirror's release-0.3.20-node back at the pin c4459193, release-0.3.21-node open at 55768f88. THE LATE-JOIN COMMIT (N9's second half, the node lane): 70e4601e on the box mirror as branch proof-hold-fix, from c4459193, two files (igneum/exec/src/proving.rs, protocol/flows/src/v10/proving.rs); the gap was the fetch side on the joiner (the served record ran the native check against the joiner's trailing exec state before anything was stored, the check refused it, the proof was never held, the body rule read "not held" for 20 s and failed the IBD); the fix holds the proof by hash before the checks (the pool entry still needs them) and the serve side says when it holds fewer than asked; the exec suite 32 passed at 13:26Z with the known-failed shape first, the flows check green 13:28Z, igneumd on build-1 at the 0321 worktree path built 13:32Z, sha256 17649eeb2f7d1290, string read back; with the testnet lane (the resume form, B alone); it joins the 0.3.21 staging as its own commit. THE WIPE CANARY ON c19-1, c4459193 (sha 45be9b02d1b002f5, string read back): FORM END rc 0 at 13:50:53Z. Wipe synced 13:35:50Z (57 minutes, inside the 98-minute class); mining 13:36:00Z to 13:47:07Z, 66 mined, 66 accepted, 0 rejected, isSynced true at the tip throughout; the hub holds 41 of its blocks in its last 700 with 0 rejects (13:47:09Z); the restart on its kept datadir at 13:47:15Z: the old process stopped at once (the new process's first lock line seven seconds after the marker; the watchdog held nothing, the b7cc37e7 fault closed), synced again at 13:48:39Z after 84 s, 109 templates read with max 3,432 ms and 0 timeouts; the kept read on pool-1's 0.3.17 copy on the same pod passed at 13:38Z (the rewrite line once, a clean second start). The pin's set on c4459193: the digest gate PASS, the mixed-version gate PASS, the wipe canary PASS, the kept read PASS, the restart PASS, the 12 GB line proves and verifies (paid is a race, not a gate); CASES END from c20-1 (about 14:50Z) is the last pin line. THE INTEROP FACT stands from the void run: the 5899f603 hub accepted 235 object-byte-5 blocks from the 8097d600 node with 0 rejected, one digest on all five nodes on the live sixteen-field file. The gates: the digest test and the kaspa-pow vector test (the amended devnet epoch-0 id 1a4230699a6b9c60 must equal, c120d7963abdcd96 must differ, the v3 control unchanged) on the box; the mixed-version Devnet 2 gate (the amended 0.3.20 node beside a 5899f603 node for ten minutes on the live file without the v4 fields) after the Mac build; the fresh-join canary the 0.3.20 cut's | | Main's rulings (7 October, morning) | no generator change to v4 on the live devnet; the record's null is the window model with numbers, sent by the hash lane to the attack-pass lane so AP-F8-1 re-gates against it; a fault beyond the model (a low-entropy source at site 15) stops at the coordinator with the two options priced (a 0.3.19 class amendment before the flip, or the flip held at the floor), nothing shipping without the founder's word; the tighter tail, an acceptance bound on the hot-set share, is a CLASS V5 item (sent to the v5 lane a6410f3b8abefb762 with the 64-seed census as its gate; the bound's number follows from the model) | ### AP-F4-1, the weak-day MUL draw (the attack-pass lane, 7 October, morning): PASS against v4, a class v5 rule From e6535e0f851eb89543d7d4085815787e24012a77 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:24:47 +0000 Subject: [PATCH 17/19] fast-time: override-60x.json taken whole from ca3-v4-node (81 keys: master's 75 plus base_unit_decimals, pool_split_activation_daa, program_class_v5_activation_daa, proving_base_fee_ceiling_multiple, proving_fee_ceiling_activation_daa, subsidy_per_block_activation_daa; no shared value differs), the file the 0.3.25 node line's test fast_time_60x_file_is_the_devnet_at_60x reads key by key; a single added field moved the red to the next key (8 October 2026, 11:2x UK) Co-Authored-By: Claude Fable 5.1 --- infra/fast-time/override-60x.json | 77 +++++-------------------------- 1 file changed, 11 insertions(+), 66 deletions(-) diff --git a/infra/fast-time/override-60x.json b/infra/fast-time/override-60x.json index db30b7f86..2a36de873 100644 --- a/infra/fast-time/override-60x.json +++ b/infra/fast-time/override-60x.json @@ -13,15 +13,8 @@ "mass_per_tx_byte": 1, "mass_per_script_pub_key_byte": 10, "mass_per_sig_op": 1000, - "block_mass_limits": { - "compute": 500000, - "storage": 500000, - "transient": 1000000 - }, - "block_lane_limits": { - "lanes_per_block": 50, - "gas_per_lane": 1000000000 - }, + "block_mass_limits": { "compute": 500000, "storage": 500000, "transient": 1000000 }, + "block_lane_limits": { "lanes_per_block": 50, "gas_per_lane": 1000000000 }, "storage_mass_parameter": 1000000000000, "deflationary_phase_daa_score": 0, "pre_deflationary_phase_base_subsidy": 50000000000, @@ -79,6 +72,8 @@ "vdf_scheme": 0, "era_vdf_t": 108000000, "fees_v1_activation_daa": 0, + "proving_fee_ceiling_activation_daa": null, + "proving_base_fee_ceiling_multiple": null, "difficulty_v3_activation_daa": 18446744073709551615, "finality_daa_rule_activation_daa": 18446744073709551615, "fork_gate_activation_daa": 18446744073709551615, @@ -88,73 +83,23 @@ "signing_bonus_activation_daa": 18446744073709551615, "signing_bonus_bps": 1000, "finality_leave_activation_daa": 18446744073709551615, - "latency_ladder": [ - { - "reps": 27, - "admissible": true - }, - { - "reps": 35, - "admissible": true - }, - { - "reps": 53, - "admissible": true - }, - { - "reps": 88, - "admissible": false - }, - { - "reps": 173, - "admissible": false - }, - { - "reps": 267, - "admissible": false - } - ], + "latency_ladder": [{"reps": 27, "admissible": true}, {"reps": 35, "admissible": true}, {"reps": 53, "admissible": true}, {"reps": 88, "admissible": false}, {"reps": 173, "admissible": false}, {"reps": 267, "admissible": false}], "latency_ladder_activation_daa": 18446744073709551615, "latency_ladder_window_daa": 120, "exec_restart_state_root": "", - "emission": { - "launch_rate": "3168808781", - "ramp_seconds": 2592000, - "ramp_start_percent": 10, - "step_seconds": 63115200, - "step_decay_q32": 2147483648, - "tail": { - "kind": "cap" - } - }, + "emission": {"launch_rate": "3168808781", "ramp_seconds": 2592000, "ramp_start_percent": 10, "step_seconds": 63115200, "step_decay_q32": 2147483648, "tail": {"kind": "cap"}}, "proving_consensus_verify_daa": 18446744073709551615, "proving_shard_program_id": "", "proving_aggregator_id": "", "sig_scheme": 0, "sig_scheme_activation_daa": 18446744073709551615, "finality_succession_activation_daa": 18446744073709551615, - "latency_ladder_cache_rung": { - "mib": 512, - "admissible": false - }, + "latency_ladder_cache_rung": {"mib": 512, "admissible": false}, "latency_ladder_cache_rung_activation_daa": 18446744073709551615, - "fees": { - "pgas": { - "version": 1, - "cycles_per_pgas": 1000, - "intrinsic_pgas_per_tx": 300, - "modexp_base": 10, - "modexp_per_byte_numer": 1, - "modexp_per_byte_denom": 10 - }, - "block_proving_gas_limit": 120000, - "shard_proving_gas_budget": 30000, - "min_execution_base_fee_wei": 100000000000, - "min_proving_base_fee_wei": 10000000000000, - "initial_execution_base_fee_wei": 100000000000, - "initial_proving_base_fee_wei": 10000000000000, - "base_fee_change_denominator": 8 - }, + "fees": {"pgas": {"version": 1, "cycles_per_pgas": 1000, "intrinsic_pgas_per_tx": 300, "modexp_base": 10, "modexp_per_byte_numer": 1, "modexp_per_byte_denom": 10}, "block_proving_gas_limit": 120000, "shard_proving_gas_budget": 30000, "min_execution_base_fee_wei": 100000000000, "min_proving_base_fee_wei": 10000000000000, "initial_execution_base_fee_wei": 100000000000, "initial_proving_base_fee_wei": 10000000000000, "base_fee_change_denominator": 8}, "peer_directory_activation_daa": null, + "subsidy_per_block_activation_daa": null, + "pool_split_activation_daa": null, + "program_class_v5_activation_daa": null, "base_unit_decimals": 8 } From c64e809fdd387be5cede912c83116068ecd13940 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:26:01 +0000 Subject: [PATCH 18/19] Swap page: the RPC-syncing line (block N of at least 26,972) instead of unreadable reserves while the public Devnet 3 node is behind the pools Co-Authored-By: Claude Fable 5.1 --- site/swap.html | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/site/swap.html b/site/swap.html index 196dc10fb..c6775f3d8 100644 --- a/site/swap.html +++ b/site/swap.html @@ -430,6 +430,7 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:13px;overflo lastQuote=null;$('amount-out').value=''; if(!tin||!tout||tin.key===tout.key){say($('quote-line'),'Pick two different tokens.','');return;} if(amt===null||amt===0n){say($('quote-line'),'Enter an amount for a quote. Slippage limit 1 percent, deadline 10 minutes.','');return;} + if(rpcBehind){say($('quote-line'),'The public RPC is still syncing; quotes come when it reaches the pools.','bad');return;} call(A.router,enc(S.getAmountsOut,['uint256','address[]'],[amt,path(tin,tout)])).then(function(r){ var w=words(r);var out=w[w.length-1];lastQuote={tin:tin,tout:tout,amt:amt,out:out,min:out*99n/100n}; $('amount-out').value=fmt(out,6); @@ -479,7 +480,15 @@ dt{color:var(--ash)}dd{margin:0;font-family:var(--f-mono);font-size:13px;overflo setInterval(refreshProofs,10000); // ---- pools and the faucets ---- + var DEPLOY_BLOCK=26972, rpcBehind=false; function refreshPools(){ + rpc('eth_blockNumber',[]).then(function(n){ + var h=parseInt(n,16);rpcBehind=h';return;} + return readPools(); + }).catch(function(){$('pools-body').innerHTML='The public RPC did not answer. Try again in a minute.';}); + } + function readPools(){ Promise.all(POOLS.map(function(p){return call(A.router,enc(S.getReserves,['address','address'],[p.a,p.b])).then(function(r){var w=words(r);return fmt(w[0],2)+' '+p.sa+' / '+fmt(w[1],2)+' '+p.sb;}).catch(function(){return 'unreadable';});})).then(function(rows){ $('pools-body').innerHTML=POOLS.map(function(p,i){return ''+p.name+''+rows[i]+'';}).join(''); }); From 697fdd23a1f5fe460e9b9973f8d8cfbd42fc6700 Mon Sep 17 00:00:00 2001 From: igneum-labs <337424239+igneum-labs@users.noreply.github.com> Date: Thu, 8 Oct 2026 10:29:17 +0000 Subject: [PATCH 19/19] Class v6 history lane, revision 2: the synthesis lane's reading of the six items folded in, lane B's hardware file cross-cited docs/analysis/class-v6/history.md: layer 2's open ceiling question replaced by the synthesis lane's resolution (the per-year ceiling as layer 2's first constant, the schedule's power-of-two step for the year; the state brings a step forward and never adds one), with the Ethereum-state figure kept as the reason the ceiling exists; the read-width rank reworded on the measured warrant (w16 latency-bound within 2.7 percent on the 5090 and the 9070 XT, within 1 percent on the M5 Max; the band is {4 B, 16 B}, nothing wider); the Jasminer X4 finding cross-cited to lane B's hybrid-bonded DRAM row (hardware-future.md 4.4, 13x to 17x modelled) as that row's shipped precedent at 5x on a planar node; the header records the 11:26 UK landing (eb71eaf3). No number changes. Co-Authored-By: Claude Fable 5.1 --- docs/analysis/class-v6/history.md | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/docs/analysis/class-v6/history.md b/docs/analysis/class-v6/history.md index 1d7b524f4..0a2b802f1 100644 --- a/docs/analysis/class-v6/history.md +++ b/docs/analysis/class-v6/history.md @@ -2,7 +2,7 @@ 8 October 2026, branch `class-v6-history`, the Counter ASIC coordinator's history lane. The founder's word at 11:1x UK: class v6 is declared with four layers as its spine, and the research opens "see if anything can be optimised, added or invented". This file extends and corrects `docs/analysis/asic-resistance-history.md` (5 October 2026, the deep dive: 31 rows, 24 papers, ten lessons, seven ranked additions); it does not repeat that file's rows. What is new here: the exact MECHANISM of every chip (what it specialised: the memory, the hash core, the instruction mix, the parameter fixity), what each design missed, the timeline from announcement to chip to response, and for each one the mapping to class v6's four layers with "does v6 close it" in one sentence and one number. Every figure about another chain cites a URL with the date it was read, or is labelled approximate. Every Igneum figure names the repo file. Reading public research is in-house; nothing is paid or asked of anyone outside. -First cut landed before 15:00 UK on 8 October (the table, the mechanisms, the three lessons); the full report by 09:00 UK on 9 October. A line goes to the coordinator and to the synthesis lane at each landing. +First cut landed on the mirror's master at 11:26 UK on 8 October (merge 4c58ad65: the table, the mechanisms, the three lessons), ahead of the 15:00 clock; this revision folds in the synthesis lane's reading of the six items (`docs/design/class-v6-rotating-family.md` section 7b) and cross-cites lane B's hardware file; the full report by 09:00 UK on 9 October carries any row the other lanes move. A line goes to the coordinator and to the synthesis lane at each landing. ## 0. One page @@ -22,7 +22,7 @@ First cut landed before 15:00 UK on 8 October (the table, the mechanisms, the th 2. **Automatic change beats the human fork only where it costs the chip a redesign, and the one parameter that does is the memory.** Grin's six-monthly tweaks held because each was a new algorithm the lane was scheduled to retire; Monero's forks lost on the second lap; Ethash's DAG growth is the one scheduled change in the record that killed a shipped chip (the E3, when the DAG passed its 4 GB of DDR3). Layer 2 is that lesson made a rule, and its rate decides everything: at 2 GiB plus 0.5 GiB a year a 32 GB chip board outlives the chain, so layer 2 as declared ages out the honest 8 GB card before any chip unless the floor is set against DRAM cost per gigabyte, not against chain state. 3. **A steered or biased address pattern is always found after launch unless the test lives in the acceptance rule, and every new draw needs its own null.** ProgPoW's seed, MTP's blocks and class v4's lossy sources were the same attack three times; layer 4 puts the test where it must be, and its cost is one census per era draw on the node (2.2 s per candidate at 2^20 today) with the null re-derived for every drawn parameter, because the window model that defines "uniform" changes with the read width and the program length. -**What the history says to add** (section 5): the issuance clock and the share-pattern detector (unchanged from the 5 October ranking, still unbuilt); a layer 2 floor stated in DRAM dollars; the read width kept out of the era draw unless a width other than 4 bytes is measured latency-bound on every vendor; the reserve ordered by what a sequencer chip cannot fold into firmware. +**What the history says to add** (section 5): a layer 2 floor and a per-tier ceiling stated before its rate (the synthesis lane has since made the ceiling layer 2's first constant); the issuance clock and the share-pattern detector (unchanged from the 5 October ranking, still unbuilt); the read width's draw band stopped where the honest card stops being latency-bound ({4 B, 16 B} on the measured warrant, nothing wider); the op-mix band bounded by the per-vendor energy table; the reserve ordered by what a sequencer chip cannot fold into firmware. ## 1. The chips, one row per mechanism @@ -77,7 +77,7 @@ One section per proof of work. Each carries the mechanism, the miss, the timelin **The timeline.** Hash live July 2015; the E3 at 32 months (announced) and 36 (shipped); the first chip over 2x at 65 months (Linzhi, December 2020); 5x at 75 months (Jasminer, October 2021); the Merge at 86 (15 September 2022). The E3's death by DAG growth: Classic first, at epoch 328 (DAG about 3.56 GB, March 2020), then Ethereum, with a 30 March 2020 firmware stretching the DDR to about block 11.4 million (about October 2020): 20 to 27 months after shipping. Ethereum's responses: Zamfir's April 2018 poll (57 percent for an anti-chip fork); EIP-1057 created 2 May 2018, a 93 percent community vote in April 2019, audits delivered September 2019, "accepted" on 21 February 2020, EIP-2538's opposition on 25 February, then stagnant; the share claim in the EIP's own text: "as much as 40 percent of the Ethereum network may now be secured by ASICs" (undated inside a 2018 to 2020 document; no year-by-year series exists). After the Merge: Classic's hashrate went 64 to 183 TH/s in one day; today Classic reads 129.9 TH/s and ETHW 2.15; at USD 0.10 per kWh every Ethash chip in the table loses money (E9 Pro minus USD 5.28 a day), and a later wave (iPollo V1 3.6 GH/s at 3,100 W, June 2022; Jasminer X16-P 5.8 GH/s at 1,900 W, August 2023) holds Classic (https://hashrateindex.com/blog/how-much-ethereum-mining-hashrate-can-other-blockchains-absorb/ ; https://2miners.com/etc-network-hashrate ; read 8 October 2026). -**The mapping to v6.** Ethash is class C in full, and its chips are the f = 1 chip of `chip-model-v3.md` section 5 at three points on the packaging ladder: commodity DRAM on a board (E3, E9: 1x to 4x), memory sized and placed per chip (Linzhi: 2x), DRAM bonded to the logic (Jasminer: 5x, the model's HBM3 row). None of v6's four layers touches a chip of this class: the program, the mixer, the op mix, the family schedule and the acceptance floor are all firmware or configuration to a controller that stores the dataset; the only layer that reaches it is layer 2, and only when the dataset passes the chip's board, which at 2 GiB plus 0.5 GiB a year is year 60 for a 32 GB board (section 4.2). Does v6 close it: **no**. The number: 5.1x per joule on GDDR7 and 7.5x on one HBM3 stack at zero premium against the 5090 unlocked, 3.6x at the 5090's 1,300 MHz knee, 2.1x at the knee with the class v4 shadow at k = 1 (`counter-asic-4-research.md` section 0); the history's measured band for exactly this chip class is 1.0x (E3) to 5.1x (Jasminer), and Jasminer's number is the model's HBM-class row reached in 2021 on a 40 nm logic die. The one thing Igneum has that Ethash did not: the honest card is latency-bound at 4-byte reads, not bandwidth-bound at 128, so the chip's energy per read is the activate's 909 pJ plus a 32-byte atom (2.0 nJ on GDDR7 against the card's measured 8.7 to 10.9 nJ marginal), which is where the 5.1x comes from, and the shadow is the only term on the card's side of that ratio. +**The mapping to v6.** Ethash is class C in full, and its chips are the f = 1 chip of `chip-model-v3.md` section 5 at three points on the packaging ladder: commodity DRAM on a board (E3, E9: 1x to 4x), memory sized and placed per chip (Linzhi: 2x), DRAM bonded to the logic (Jasminer: 5x, the model's HBM3 row). None of v6's four layers touches a chip of this class: the program, the mixer, the op mix, the family schedule and the acceptance floor are all firmware or configuration to a controller that stores the dataset; the only layer that reaches it is layer 2, and only when the dataset passes the chip's board, which at 2 GiB plus 0.5 GiB a year is year 60 for a 32 GB board (section 4.2). Does v6 close it: **no**. The number: 5.1x per joule on GDDR7 and 7.5x on one HBM3 stack at zero premium against the 5090 unlocked, 3.6x at the 5090's 1,300 MHz knee, 2.1x at the knee with the class v4 shadow at k = 1 (`counter-asic-4-research.md` section 0); the history's measured band for exactly this chip class is 1.0x (E3) to 5.1x (Jasminer), and Jasminer's number is the model's HBM-class row reached in 2021 on a 40 nm logic die. The forward line of the same mechanism is lane B's (`docs/analysis/class-v6/hardware-future.md`, master 34f63b3c, section 4.4 and its table row for fine-grained hybrid-bonded DRAM on logic: 13x to 17x per joule at zero shadow, modelled, on the 2028 to 2030 roadmaps); the Jasminer X4 is that row's shipped precedent, five years early and at 5x on a planar node, which is the reason to read lane B's 13x to 17x as a ceiling a first product will not reach and a second one will approach. The one thing Igneum has that Ethash did not: the honest card is latency-bound at 4-byte reads, not bandwidth-bound at 128, so the chip's energy per read is the activate's 909 pJ plus a 32-byte atom (2.0 nJ on GDDR7 against the card's measured 8.7 to 10.9 nJ marginal), which is where the 5.1x comes from, and the shadow is the only term on the card's side of that ratio. ### 2.3 RandomX (Monero, 30 November 2019): the chips, RandomX v2, and the X9's withdrawal @@ -232,7 +232,7 @@ The arithmetic that binds. A chip's memory is bought by the board, and today's p | 24 GB | 44 | one HBM3 stack; the 4090 and the M5 Max class | | 32 GB | 60 | the f = 1 GDDR7 chip's board; the 5090 | -So layer 2 as a rate "tracking chain state" ages out fixed-memory silicon only if the dataset grows faster than a chip generation's memory headroom, and every rate that does that retires the honest small cards first by the same table. The E3 is the only case in the record where a growth rule beat a chip, and it beat a chip that had under-provisioned memory by a factor the card fleet also hit (4 GB). The rule that would hurt the f = 1 chip is one that keeps the dataset above what one board of commodity DRAM holds at the chip's price point, and that rule is unaffordable for the honest fleet. The honest reading: layer 2 is the right lever class (the memory is the one parameter a stored-dataset chip cannot read as firmware), and its value is set by the floor and the ceiling, not by the tracking: a floor keeps the dataset above every SRAM die (class B stays closed: 2 GiB is 1,000 mm^2 of SRAM even at N5), and a ceiling keeps it under the honest tiers' memory. Between those two lines the chip's board holds whatever the card holds, and the growth rate changes nothing for it. What "tracking chain state" adds over the fixed schedule is governance (no release decides the size) and the class v5 link (the dataset is built from the state, so the size follows the state's record count naturally); it is not an anti-chip rate. Open for the synthesis lane: if the chain's state grows the way Ethereum's did (approximate, from memory: Ethereum's account and storage state passed 100 GB in its eighth year), a dataset that tracks it literally outgrows every consumer card inside the chain's first decade, so the ceiling is the number to fix, in gigabytes per tier per year, before the rate. +So layer 2 as a rate "tracking chain state" ages out fixed-memory silicon only if the dataset grows faster than a chip generation's memory headroom, and every rate that does that retires the honest small cards first by the same table. The E3 is the only case in the record where a growth rule beat a chip, and it beat a chip that had under-provisioned memory by a factor the card fleet also hit (4 GB). The rule that would hurt the f = 1 chip is one that keeps the dataset above what one board of commodity DRAM holds at the chip's price point, and that rule is unaffordable for the honest fleet. The honest reading: layer 2 is the right lever class (the memory is the one parameter a stored-dataset chip cannot read as firmware), and its value is set by the floor and the ceiling, not by the tracking: a floor keeps the dataset above every SRAM die (class B stays closed: 2 GiB is 1,000 mm^2 of SRAM even at N5), and a ceiling keeps it under the honest tiers' memory. Between those two lines the chip's board holds whatever the card holds, and the growth rate changes nothing for it. What "tracking chain state" adds over the fixed schedule is governance (no release decides the size) and the class v5 link (the dataset is built from the state, so the size follows the state's record count naturally); it is not an anti-chip rate. Resolved by the synthesis lane (11:2x UK, `docs/design/class-v6-rotating-family.md` section 7b on `counter-asic-4`): layer 2's rule carries a per-year ceiling as its first constant, the fixed schedule's power-of-two step for that year, so the chain's state can bring a step forward and never add one; "retires cards before chips" is recorded as the reason the ceiling exists. Under that rule the table above is the ceiling's own schedule, and the honest 8 GB tier's year-12 line stands whatever the state does. What this file adds for the record: if the chain's state grew the way Ethereum's did (approximate, from memory: the account and storage state passed 100 GB in its eighth year), a dataset tracking it with no ceiling would have outgrown every consumer card inside the first decade; the ceiling is what makes layer 2 a governance rule and not a fleet-retirement rule. ### 4.3 Layer 3: scheduled family epochs by height, every 180 days, no release @@ -268,7 +268,7 @@ Reading. Layer 4 is the right answer to this class and the only one of the four |---|---|---|---|---| | 1 | **State layer 2's floor and ceiling in gigabytes per tier, before its rate.** The floor above every SRAM die (today's 2 GiB holds); the ceiling under the honest tiers' memory on a stated glide (the 8 GB tier's life is the first number) | the E3 is the only chip a growth rule ever killed and it was the chip with the fleet's own memory limit; Scrypt-N was abandoned because its growth was public and slow; Autolykos's growth is untested; a dataset that tracks chain state literally outgrows every card if the state grows the way Ethereum's did (approximate) | none at the floor; everything at the ceiling | genesis rule, with the card-lifetime table | | 2 | **The clock and the detector**, unchanged from the 5 October ranking and still unbuilt: the per-program rate spread and nonce pattern on the observer (the method that found Monero's chips at 85 percent), plus a share-by-key-and-template instrument (what found Qubic until it randomised), plus the issuance trigger at about USD 50 K a day | every chip in the record was on its chain before it was announced (Monero 2017, Zcash's three groups, SChernykh's 2021 reading of the X5) | none | before the public testnet | -| 3 | **Keep the read width out of the era draw** unless a width other than 4 bytes is measured latency-bound on all three vendors; a draw over {4 B, 16 B} is harmless and worthless (w16 moved the chip's cost not at all) | the Ethash chips' whole edge was the bandwidth lever Ren and Devadas name; w64 made the 5090 bandwidth-bound | a 47 percent loss on the 5090 at w64 | spec 1.13.1's allowed set stays {1} | +| 3 | **The read width's draw band stops where the honest card stops being latency-bound**: the synthesis lane keeps {4 B, 16 B} on the measured warrant (w16 latency-bound within 2.7 percent on the 5090 and the 9070 XT, within 1 percent on the M5 Max); w64 is out (the 5090 bandwidth-bound at 71.9 MH/s); the draw costs every chip nothing (the same 32-byte atom at either width) and is a governance value only | the Ethash chips' whole edge was the bandwidth lever Ren and Devadas name; a width that makes the card bandwidth-bound hands the chip that lever | within 2.7 percent at w16; 47 percent at w64 | the band {4 B, 16 B} in the class v6 spec; nothing wider without a per-vendor latency-bound measurement | | 4 | **Bound the op-mix draw by the per-vendor energy table, not only by the rate spread**: the 5090 pays 55.8 pJ per shuffle against 11.3 per add, so a shuffle-heavy era taxes the honest card up to 5x per instruction for no better k | X16R's drawn order cost the chip nothing and the fleet nothing; Igneum's draw can cost the fleet watts | up to 2x the premium per instruction at the band's edge | the band's definition in the class v6 spec | | 5 | **Order the reserve by what a sequencer cannot fold into firmware**, mm8 last; and name the random item-derivation program (the 5 October addition 2) as the one reserve item whose semantics are not knowable at genesis | the kHeavyHash chips and the 5090's own 1.4 to 4.1 pJ per int8 MAC; RandomX's SuperscalarHash is the one idea Igneum has not taken, and it acts on the f = 0 chip only | none at launch | reserve ordering, genesis | | 6 | **Generalise layer 4 with a null per drawn parameter**: the window model re-derived per era, the census per draw, and a stated ceiling for the shadow-written residue per shadow placement | lesson 3 | 2.2 s per candidate once an epoch on a node | the class v6 acceptance rule |