diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md
index df2fbe4a..16303253 100644
--- a/docs/plans/counter-asic-3-status.md
+++ b/docs/plans/counter-asic-3-status.md
@@ -423,7 +423,17 @@ Every failing seed is ONE low-entropy load site; the mechanism is lineage-blind
 
 Known-failed test class_v4_distinct_ratio_rejects_the_low_entropy_band green on box 2 (12.95 s): p15 attempt 3 (52638ea2e8b0fd68) site 2 at 0.943; p18 attempt 2 (9a37e9489d8ba698) site 6 at 0.927; p19 attempt 0 (79d7441de0689223) site 15 at 0.933; p56 attempt 2 (486a8ad2701ec3b5) site 2 at 0.965; p23 attempt 1 (d65122675f16a1c7) draws site 7 from r5 and passes, and with r6 put back is refused by (a') UnfreshLoadSource and, run anyway, by the ratio at 0.836. Stream unchanged from ddacfbd3 (re-export diff 0 on all eight packs): id a785001687d8688a, the seven fingerprints and the packs zip sha256 4f2445c5... as recorded. THE SUITE AT 017e7037 on box 2: 103 of 103 (64 lib + 7 derive + 4 mixer + 19 packs + 2 recheck + 7 scratch, 3 diag ignored), rc 0; the lib tests 140.8 s against the 41 s whole-suite line at ddacfbd3, because every class v4 draw in the tests pays the 2^20 ratio pass on its chosen candidate (2.2 s each): a CI-time cost, not a node cost (one pass per epoch draw). CI run 37639406567 on 017e7037 success. THE CENSUS AT 017e7037 (box 2, 4,096 chain-shaped seeds plus F8's p1 to p3, draws in parallel across the cores; tools/ca3-v4-uniform now draws across available_parallelism since the serial run became a four-hour job): 4,099 programs, 0 lossy-sourced load sites of 65,584 (57,322 injecting, 8,262 bijective), 0 exhaustions; attempt histogram 0:1297 1:899 2:609 3:416 4:298 5:197 6:125 7:92 8:54 9:46 10:21 11:14 12:13 13:9 14:6 16:2 17:1, mean 2.086, max 17 (ddacfbd3 read 1328/917/622/409/284..., mean 1.998, max 17): the ratio refuses about 4 percent of candidates that pass every other test, one extra attempt on about one seed in twelve, the worst case unchanged at 17; devnet epoch-0 at attempt 1, id a785001687d8688a; F8's p2 attempt 5, p3 attempt 1. The attack-pass lane's class check: ddacfbd3's shadow-executed verdicts against 8bdcbdd8 on 598,678 chain-shaped seeds move 11,990 (2.0 percent) to a different attempt, 0 exhausted, max 32; on 017e7037 the pairing holds and the 64-seed gate at 2^24 plus the 10^6 exhaustion count run to about 16:05Z. The hash lane's ledger lines 2a111fb1 on origin and build (the GitHub 500s cleared on their own), CI run 37642582140 success at 15:25Z; the hash lane has nothing in flight. Nothing on the hash side stops a cut of sub-version 3 at 017e7037 for 0.3.22; then the attack-pass lane's 64-seed gate at 2^24 and 10^6 exhaustion count. Owed as before: G2, G3, the ladder re-measure, AMD (PC 1), the 2019-class core, the fleet and PC 2 G1 on the sub-version 3 packs (a re-run of a known result). release-0.3.21-node STAGED (the node lane, 16:0xZ): 96161037 on the mirror, at the shipper's sweep-end word, in the final order on 55768f88 (c631c64b, 52e96c94, f067f7c1, b0444f51, 437f0438, 2e32d5f6, f95178a1 cherry-picked, a6864e36 and d8bceca5), pairing igneum-pow 8c728ca3 at byte 5, no re-pin. The whole set on the tip green: consensus-core 126, the consensus crate whole (lib 120, both integration targets), igneum-miner 25, kaspa-pow 17, igneum-exec 36, the three checks, 15:49 to 15:57Z on build-2. Binary built on build-1 at 15:58Z, igneumd sha256 f99340b3cf4ce32e, the string read back; the digest eada4bda on the previous live object and 4bbbe816 on the published floor file (so the 0.3.20 floor file is published), as the pin reads them. The node lane's digest and mixed-version gates on it from 16:00Z; the fleet's set from the same minute; plan 6.9 carries the steps and the two box-input rules the staging added. THE 0.3.20 RECORD (the shipper, UTC). Cases: the dc141409 run on c18-1 CASES END rc 0 at 12:37:18Z (both halves PASS); the c4459193 run on c20-1 CASES END rc 0 at 14:54:02Z with its relay half void (one RunPod host, no hairpin) and the poison half PASS (13 rejected, 0 accepted); main's (b): the publish on the dc141409 cases plus the diff argument (the class byte the only touch in the cases' territory) and c19-1's tip-following, with the separate-host relay re-run on c19-1 on the live digest as the halt condition and the Discord card's gate (CASES END about 16:25Z). PUBLISH 14:54:17Z (15:54 BST): manifest 0.3.20, mac DMG sha256 73796c5f..., Windows installer 45b2f3fb..., the hive tar d9dd12df... on the public alias. THE FLOOR FILE: program_class_v4_activation_daa 900,000 (the live DAA 294,073 at 14:51Z; 294,073 + 604,800 = 898,873 rounded up to the 3,600 boundary), window 86,400 unchanged, file sha256 294f1f80...; the digest 4bbbe8162ea9fff277aa5b16b4ffad9e2262ba5e6a5acac7b697ff77211e7328 read on c4459193's binary before the cut and on the hub's handshake line after. The pin c4459193, igneum-pow 8c728ca3, byte 5; the sweep's node pair a80ed39c / 70a5180f (the build-server lane's native build of the same tree). THE SWEEP complete 15:39Z (16:39 BST): wave 1 the hands (observer-node, node1 at 14:57Z), the Mac (15:23Z on its poller, interface 1.0.1), the hub, pool-1 and the eight heaviest voters (15:04 to 15:21Z); wave 2 the four lighter voters (15:23 to 15:26Z); wave 3 Devnet 2's four pods and bps-seed (digest 4a0b8726 unmoved); the first new-side lock 9313 at 15:28:08Z after a 23-minute split; every lock line "sweep complete before 13 October 09:00 UK (the margin; the floor is now DAA 900,000)". 0.3.17 nodes remaining: PC 1 and PC 2 (offline for the project lead's cable work; their apps update on their pollers on return) and the three testnet seeds (on the testnet, not on the devnet's floor); no devnet node of the fleet, the hands or the Mac on 0.3.17. EARLIEST FLIP: the floor at DAA 900,000 is about 7.0 days of DAA from the publish (605,927 DAA at about one a second), so about 14 October 23:00Z (15 October 00:00 BST) at the earliest, and only once the seven 86,400-DAA windows read 95 percent of blue weight signalling byte 5; the live floor of 831,600 (13 Oct 08:00Z) is replaced by the published file on every swept node, so the chain never flips to the 6 October stream. Per tier: a solo miner on the Mac or a swept box mines on; a PC miner on 0.3.17 is refused by digest when it next connects until its app updates (its poller does this on return; nothing by hand); the pool and the hub are on the pin; the auditor reads one object, one digest, one floor.
 
-SUB-VERSION 3 PASSES BOTH GATES (the attack-pass lane, the close line dated 7 October 2026): class v4 sub-version 3, commit 017e70376489251e18564c0abce7e466e606c8b3 on ca3-v4-amend, pairing id a785001687d8688a (verified by the harness). F8's 64-seed census (p2 to p65) at 2^24 nonces each, chain path, window-model control, box 2, 14:51Z to 16:00:20Z: PASS, 60 of 64 under 1.2x (0.9915x to 1.144x); the only four over are the named tail, unattributed and chased: p10 1.5036x (identical to sub-versions 1 and 2, hottest item 0x4004da at 362 reads, no predicted source), p8 1.3776x (0x837de4, 420 reads), p34 1.2505x (0x800010, 541 reads, the one-bit value through sub at 5), p4 1.2167x (0x4000e7, 355 reads); every strong seed gone (p23 4.82x, p19 3.32x, p15 2.57x, p18 2.50x, p56 2.01x all under 1.2x); the worst ratio on the stream 1.50x; the tail's hottest items carry 355 to 541 reads of 2^31 (one to two per 2^22 items above the mean), no chip consequence. The exhaustion gate as it is: the chain-path count on 017e7037 runs slowly (the draw evaluates (c'') at 2^20 per candidate: 20,532 seeds in 59 minutes, 0 exhausted, 0 panics, max attempt 29), so a 10^6 count is two days away and is not waited on; the substance is met by construction (the 256 cap and the last resort unchanged from 8bdcbdd8, 0 of 10^6 there) and by measurement at 017e7037 on 24,631 seeds (0 exhausted, max 29; r about 0.68); the count runs on as a strengthening line. The node's epoch draw now costs about 2 attempts at 2.2 s each, 4 to 5 s per epoch on slower cores, once an hour. The hash lane's ledger at 6941da1d. SUB-VERSION 3 AT 017e7037 IS THE FROZEN GENERATOR FOR 0.3.22 (byte 7), THE TAIL RULED (main, 18:1x UK): the four seeds (p10 1.5036x with its hottest item at 362 reads, p8 1.3776x at 420, p34 1.2505x at 541, p4 1.2167x at 355, of 2^31 reads) are accepted as the window model's unattributed residue at 1.22x to 1.50x with nil chip consequence; the AP-F8-1 row CLOSES with that wording and the hottest-item counts beside it; the 10^6 count on 017e7037 runs on as the strengthening line. The attack-pass plan's start 15 October or the morning after. 0.3.21's FIRST CANDIDATE 96161037 (sha256 f99340b3cf4ce32e, string read back) passed the node lane's two gates on its own binary: the digest gate 15:59:54Z to 16:01:32Z PASS (a89be8a7 with the peers right; db9a85f9 refused, no peer); the mixed-version gate 16:01:53Z to 16:12:05Z PASS (one digest b0afb2ee on five; 252 new and 352 old accepted, 0 rejected; counts equal at 316, 493 and 604 through both clean joins and the restart step; no panic); the node side complete; the fleet's set on the same binary (the kept start with the ids gate, the wipe canary, the cases rerun, the 12 GB line, N15's restarts of p1-5090 and p2-3090-3) is what the shipper's pin word waits on; byte 5 throughout, nothing on the class v4 seam moved. DEVNET 3 NOW (the project lead through main, 17:3x BST): 0.3.22 is the Devnet 3 release (a fresh network object igneum-devnet-3 with every activation at 0 and no override file, the era VDF fork, class v4 sub-version 3 at byte 7 from genesis), the node building on build-1, genesis by 17:30Z (18:30 BST). THE HANDOFF (17:3x BST, inside the 17:40 BST line, to the shipper and the node lane): igneum-pow 017e70376489251e18564c0abce7e466e606c8b3 (the audit-freeze-2026-10-07 tag; 2a111fb1 and 6941da1d above it docs only), object byte 7, PROGRAM_SUBVERSION_V4 = 3, devnet epoch-0 program id a785001687d8688a (must-differ c120d7963abdcd96, 1a4230699a6b9c60, a788661687db4bb3), kit packs-ca3-v4-sub3 zip sha256 4f2445c50c58d76a5544023492d8b858d0b07c5e372d31f9c90c4ce51f829154, the eight fingerprints as recorded (equal on Metal, Apple OpenCL, the fleet 5090 and PC 2's 5090), both attack-pass gates GREEN, suite 103 of 103, CI success; the open items stated as open: the four-seed tail under main's ruling (the attribution for 0.3.23), the 10^6 count as a strengthening line, the 4 to 5 s epoch draw, the owed measurements that do not gate Devnet 3. NO FURTHER HASH CHANGE RIDES ON 0.3.22; anything from the tail goes to 0.3.23. The node lane stages the re-pin on the 0.3.21 tip as its own commit (the fork commit and the kaspa-pow suite line owed to this record); the hash lane's sub-version 3 pairing follows it. THE PUBLIC CHIP TEXT REWRITTEN LAUNCH-FIRST (the project lead: "I thought we were making it 2.1 from launch?"; master 9b996d06, docs/plans/counter-asic-3-public-text-2026-10-07.md): the testnet and mainnet objects set program_class_v4_activation_daa 0, so class v4 is live from genesis and the launch number is 2.1x to 3.9x on day one; the three texts and evidence row 17 lead with that (labels kept), then class v5, then the 5x to 9x only as the class v3 baseline the work started from, the devnet's activation height a devnet fact only; no em dashes, no prize mention, eight columns; with the site lane to apply in the same deploy as the padding sweep (its commit and deploy time owed to this record). STOP-THE-LINE ON MASTER's POW SUITE (main through the CI steward, 18:4x BST): red since 16:31 BST (runs a4eaf76 and 1210158d; program_ids_differ_between_class_v3_and_class_v4_of_one_seed and cpu_recheck_equals_the_worker_reference_for_class_v3_and_class_v4 fail on packs-ca3-v4/v4-devnet-epoch0) because era-vdf's merge 0e2d6b1c put the sub-version 1 line's generator on master against tests/recheck.rs's pins, never CI-run. Ruling: the hash lane merges ca3-v4-amend's tip 6941da1d to master through merge-to-master.sh within 15 minutes, provided its epoch-0 id equals the frozen object's; the coordinator verified the condition (6941da1d's igneum-pow byte-identical to 017e7037; recheck pins 0xa785_0016_87d8_688a with the three must-differ ids; PROGRAM_SUBVERSION_V4 = 3), so no revert; master is the 0.3.23 line after this, 0.3.22 keeps its pin at 017e7037. EXECUTED: ca3-v4-amend on master as merge cf7d6ccb (pushed 16:45:48Z through merge-to-master.sh, try 1; the branch gate GREEN, 55 checks in 317 s on 874e945d; master ac8ed2f1 merged into the branch first with docs/fud-ledger.md keeping both sides and igneum-pow taken wholesale from 6941da1d, byte-identical to 017e7037); master's CI run 37654633279 on cf7d6ccb in progress from 16:46:10Z, the pow job's conclusion owed. Master's igneum-pow is now the frozen sub-version 3 generator. THE 0.3.22 RE-PIN STAGED (the node lane, 16:38Z): fork commit bd710a36 on release-0.3.22-node (igneum-pow 017e7037 archived beside the fork as igneum-pow-amend, byte 7, epoch-0 id a785001687d8688a must-equal, the three must-differ ids, PROGRAM_SUBVERSION_V4 read as 3); the line's candidate fa7f854f (the re-pin, the era VDF merge f2ecf452, the igneum-devnet-3 object with program_class_v4_activation_daa 0 and a one-day signal window, so byte 7 is stamped and hashed from genesis); the kaspa-pow suite on build-2 from 16:38Z, its line owed; the hash lane has the commit for its pairing run. THE kaspa-pow SUITE ON THE 0.3.22 CANDIDATE 21d8f454 (build-2, 16:47:10Z): GREEN 17 of 17; the pairing test reads PROGRAM_SUBVERSION_V4 = 3, epoch-0 id a785001687d8688a, the three must-differ ids hold, the chain draw at attempt 1 against class v3's 0 on the test header; the candidate's digest for igneum-devnet-3 ab9af79f...ed23; the node stamps object 7 (block version 1794) from genesis. MASTER GREEN ON THE POW FIX: CI run 37654633279 on cf7d6ccb success at 16:54:51Z, the igneum-pow job success; master's igneum-pow the frozen 017e7037 byte for byte; build-1's amend checkout synced to 874e945d (its packs-ca3-v4 carries the sub-version 3 kit). THE CHIP TEXTS LIVE (the site lane, site-ui-5): sections 1 to 4 applied verbatim in b34d1932, master cc593483, live on igneum.network at 16:47Z (home row 03 "under class v4 from the first block"; /litepaper#chip-model the new paragraph and the reordered table with the class v3 row last; /miner the new line; /evidence row 17 "2.1x (k = 1) to 3.9x" in eight columns); the litepaper's abstract rewritten launch-first too; tools/ci/ledger-text-check.mjs X35 asserts the launch-first sentence on the home page and "so the launch number is the class v4 row" on the litepaper; no prize mention. /claims (the litepaper's limits section) rewritten launch-first by the coordinator in the litepaper's chip bullet (2.1x to 3.9x under class v4 from the first block with its labels, class v5, then the class v3 baseline "never the launch state", the recompute chip under 1x, the X9 history), /claims rebuilt from it, the ledger text check 61 of 61, the padding and overlap sweeps green in the 56-check gate, master 0a999646 at 17:01:47Z. DEVNET 3's FIRST GO (the fleet lane): dn3-g1 on 21d8f454 up 16:50:21Z from an empty datadir, digest ab9af79f matching the build-1 read, genesis a6fa348e executed (chain id 4463), object 7 / block version 1794, era VDF from DAA 0, no override file, mining from 16:50:42Z; the dn3_ observer running since 16:38:29Z; DN3_GO_DATE=2026-10-07 in /srv/hands/dn3/dn3.env, the hash-origin timer's first dn3_ report 8 October 08:30Z. BUT NO BLOCK ACCEPTED: dn3-g1 refuses every block its miner finds with "the block timestamp is too far into the future: block timestamp is 1791417600001 but maximum timestamp allowed is 1791392281213" (16:57:51Z): 1791417600001 ms = 2026-10-08T00:00:00.001Z, the template stamping genesis + 1 ms, so the igneum-devnet-3 genesis object carries 8 October 00:00Z, one day after the intended 7 October 00:00Z; every block is seven hours in the future and refused before PoW on any node with any miner or pack; the pack is not the suspect (the node draws the epoch's program under its own igneum-pow and hands the pack to the worker; the exported kit is for G1 and recheck only). The fix is the node lane's (a past genesis timestamp, moving the genesis hash and the digest) and a rebuild; the fleet reruns the go within the minute of the new pair; if the corrected build lands by about 17:20Z the pair lock is about 17:35Z (18:35 BST). Spend at 17:00Z about USD 410 of 1,000. GITHUB SUSPENDED (17:0xZ): every push to origin refused with "Your account is suspended" (403); the API reads 404 for the user igneum-labs and the repo (the shape of a suspended account, not a revoked token); with the project lead (the ticket route given by main); until GitHub answers, every lane pushes to the build-1 mirror (/srv/igneum.git over ssh) and the mirror is the record, merges landing through the box gate stamp under the shipper's exception window. This branch is on the mirror at a33a7859 (17:09Z); master's last origin landing 0a999646 (17:01:47Z). DEVNET 3 MINES (the fleet lane, 17:08Z): dn3-g1 on the corrected candidate 69d1b56e (genesis 7 October 00:00Z, hash 4020cb43..., digest 83eb50cdf2eda4cb...) accepted its first block at 17:06:19.920Z, 3 blocks by 17:06:44Z, 0 rejected; dn3-g2 joining, the pair's first common lock about 17:15Z (18:15 BST), then the late joiner and the canary's ten minutes; the hands pair for 69d1b56e landed 17:05:54Z, the shipper ruling whether the genesis bytes are the node lane's or the hands'. Per tier: the first chain that hashes class v4 sub-version 3 (byte 7) from genesis is live on the fleet; the launch-first chip texts describe it exactly. DEVNET 3's EPOCH-0 PROGRAM ID (the node lane): fce15bf61030be57 on igneum-devnet-3 (genesis 4020cb43...b925), read in the 0.3.22 miner's "cache ready" line on build-1 at 17:10:39Z and on dn3-g1 at 17:12Z, the node accepting 85 of 85 GPU blocks; a785001687d8688a stays the shared devnet's epoch-0 id (genesis edc4fa84) and the kaspa-pow pairing pin; the pairing is unchanged, the id follows the seed; the fleet's pack-id gate reads fce15bf61030be57 at epoch 0 and the node's per-epoch value after; the exported kit's eight packs are the shared devnet's seeds, so a Devnet 3 kit, if the hive wants one, is a re-export over 4020cb43's seeds (the hash lane's, on request). DEVNET 3's PAIR AGREES (the fleet lane, 17:19Z): dn3-g2 up 17:17:58Z on the same bytes, digest and genesis, synced from dn3-g1 and mining from 17:18:15Z; the pair agrees on every determined finality checkpoint (20, 21, 22 on identical blocks in both logs); 702 blocks, 0 rejected on either, by 17:19:08Z. The first LOCK lands about 19:10Z (20:10 BST): the object's finality window is 7,200 DAA ("window filling, 681 of 7200") at about 1 block/s, the object as cut, nothing wrong. Two independent nodes agree on the chain since 17:18Z; finality signatures start two hours in; the hash-origin daily (dn3_ tables, DN3_GO_DATE 2026-10-07) counts from today once the observer units are enabled on the shipper's go. The shipper rules whether the genesis is declared on the pair's agreement or on the first lock; the late joiner, the empty-datadir canary (ten minutes) and the spare placing on the hands pair; the standing boxes' second nodes after the joiners on the shipper's word. Spend 17:20Z about USD 412 of 1,000. GITHUB RULE (main through the shipper, 18:02 BST): no push, fetch or poll of GitHub from this lane or its sub-lanes, not even a retry; the box mirror on build-1 and build-2 is origin and the box gate stamp the verdict until it lifts. DEVNET 3's EPOCH-0 PACK EXPORTED (the hash lane): program.json 0xfce15bf61030be57, attempt 0, sub_version 3 (dn3-g1 drew the export path's seed); igneum-pow at 874e945d (byte-identical to 017e7037 and master's), --epoch-hex and --era-hex the genesis 4020cb43...b925, class mx8-eraaf3a9139+sh256x27, day bytes "igneum-day/" || le64(20733) (the chain's rule day = timestamp_ms / 86,400,000; the program and id day-independent, the dataset and fingerprint rolling with the UTC day); Metal fingerprint e510ad92b4d24846 at 2^24 from base 0, Apple OpenCL equal, vectors 3 of 3; on build-1 /srv/artefacts/packs/v4-devnet3-epoch0/ and .zip sha256 e025750f71175ed14d6e2a24e387ebbf1979b1cd0faee9139c41a7671165b334, the sub3 kit zip beside it (4f2445c5...); in the 0.3.22 hive package order with the kit. [user]'S THREE ORDERS (through main, 18:3x BST), with his correction (18:4x BST: everything in-house, nothing external): (1) BUILD CLASS V5 NOW on the 0.3.23 line (program class v5, its own activation height, object byte 6): the generator and verifier in igneum-pow (+0.2 ms per warp against the 10 ms gate), the node side (the state commitment into the dataset derivation, the stateless/stale-chip rule known-failed first on a stale dataset), the hot-set, weak-day and shadow-redundancy rules routed to this class, the attack-pass families re-run on v5, the kit for Metal, CUDA, OpenCL/AMD and Intel with fingerprints equal, the crossing on Devnet 3 by height after its gate, miner cost rows per card; the v5 lane resumed with the order, owing tonight the design-to-code gap in one list and a UTC clock for the first v5 pack. (2) CRYPTANALYSIS IN-HOUSE: no outside firms, no paid lots, no briefs to anyone; the three targets (the mixer M_r, the chained cache, the acceptance rule) attacked by three adversarial lanes that have never worked on the hash code, each given only an outsider's inputs (the public kit, the frozen object, the spec, the attack-pass harnesses), a written attack plan first, a budget of box hours, a report in the attack-pass shape (the claimed break or the bound reached, reproducible); the Counter ASIC lane the defender, main ruling disputes; plans within two hours, first results by tomorrow evening; a coordinator lane spawned for it (docs/plans/cryptanalysis/in-house-pass.md, funding.md's lots rewritten to the internal pass). The served sentence "the cryptanalysis plan buys three external lots" rewritten on the litepaper, /claims and evidence row 17 to the internal adversarial pass, labelled internal, the one outside check staged on its escrow and the publish word (the served text never names the prize or the project lead, by the forbidden-strings gate). (3) OWED MEASUREMENTS on PC 1: released tonight after the shipper's 0.3.21 host build (the line "PC 1 released" about 18:05Z, the window to 17:30Z on 8 October, an elevated job allowed under the standing rules); the hash lane prepares the three job files on the sub-version 3 kit (the 9070 XT G1 and ladder rows, item 6's step costs on AMD, the 5090 clock rows elevated, dr736 if it fits), published one at a time on the coordinator's relay of the line, --cards-off on the measured card alone, the desk left usable; the rows land in the public table with their labels. Nothing in (1) or (3) reaches outside the fleet, the boxes and the PCs. DEVNET 3's PROOF WINDOW OPENED 17:57:05Z (the fleet lane): coverage from 17:44:23Z (dn3-x1, an RTX 3060 12 GB, the sm_86 SP1 floor, from DAA 0; the first segment 1024..1031 submitted 17:45:49Z, 86.1 s end to end); the first records PAID by dn3-x1's key, segments 1256..1263 (0.7036 IGN) and 1272..1279 (1.5993 IGN) in block 1403, paidShards 10 at 17:57Z; the 24 hours run to 17:57Z on 8 October; sizing: a 3060 proves a segment in 84 to 86 s (about 42 an hour) against 450 segments an hour at 1 block/s, 11 cards for a share of one, 15 placed (14 more 3060 12 GB pods on distinct Vast machines, USD 1.07/h together); a runbook rule found: a Devnet 3 block builder carries a proof record only after its own node verified it, so every Devnet 3 miner node needs IGNEUM_PROOF_VERIFIER on the host verifier (the four miners ran bare for 11 minutes, 72 records pending and 0 carried; dn3-j1 restarted with it at 17:54:37Z and carried the first). THE UTILISATION TABLE (the fleet lane, 18:00Z, for the project lead): 36 rented boxes plus the Hetzner seed, 37 GPUs, USD 197 a day at this size, today's total about USD 432 of 1,000; idle in the hour 12 by the letter, 9 of them Devnet 3 boxes rented in the last 15 minutes and syncing, 3 the 0.3.21 warm set standing between windows; the fleet's recommendation: stop Devnet 2's four boxes (USD 59.76 a day; Devnet 3 is now the staging chain) on the shipper's word, fold the 0.3.21 warm set (USD 12.72 a day) into Devnet 3's relay trio when 0.3.21's last line closes. THE PUBLIC BENCH TABLE (main's order): /miners now carries every measured card, 37 rows, with watts, MH per watt, the class v4 cost per card and the tuned state (the 5090 stock 136.1 MH/s at 350 W and tuned 127.71 at 226.8 W; the 4070 at its tune point; the 9070 XT, the M5 Max (no lever), the 5060 Ti, the Arc B580, and the fleet's rented-card sweep of 7 October: the 5080 71.16 at 143.4 W, the H100 248.70 at 385.6 W, the B200 416.35 at 855.6 W, the 4090 hands row 52.25 at 183.1 W, down to the 4060 Ti 20.10 at 77.5 W; every fleet row stock, bench only, labelled measured by the fleet or reported by the fleet). CLASS V5's DESIGN-TO-CODE GAP (the v5 lane, 19:00 UK; section 13 of docs/design/class-v5-stored-state.md at c17bc04b): igneum-pow's class v5 rebased onto the frozen sub-version 3 (v5 and its rungs draw under (a'), (c''), the shared-operand rule, the 256 cap and the last resort by merge); the fork rebased onto release-0.3.23-node with class v5 pinned to object byte 6 counted EXACTLY (byte 7 is v4 sub-version 3; the bytes are no longer ordered by class); the gap about 40 agent hours (17 the v5 lane's): the +0.2 ms per warp on the Mac (1 h), the AP-F4-1 and AP-F1-1 rules with their known-failed cases (3 h each), the attack-pass families on the v5 stream (4 h), the kits (Metal and CUDA hosts uploading leaves.bin, 2 h each; OpenCL/AMD and Intel 3 h), G1 on the 5090 and the Mac (2 h), the Devnet 3 crossing by height (3 h), the miner cost rows (3 h), the snapshot wire's day streams (3 h), the pool's per-epoch state fetch (2 h), the proof witness (4 h), the spec text (2 h); the first v5 pack's clock (Devnet 3's genesis as epoch and era seed, day 20,733) follows the suite and the day-stream bin. THE IN-HOUSE ADVERSARIAL PASS (the crypto lane, 19:05 UK): crypto-engage deb0011e on the mirror; docs/plans/cryptanalysis/in-house-pass.md (outsider inputs and withheld files, the three lanes, the budget 8 box-hours each and 24 total with a ceiling of 48 on the coordinator's word, the review roles, the clock, the label rule); funding.md's lots rewritten to the internal pass with no cash row and every firm name removed; the three lanes adv-mixer, adv-cache and adv-accept spawned 19:1x UK with outsider-only inputs, plans due 21:10 UK, first results by 18:00 UK on 8 October; the Devnet 3 epoch-0 pack added to their inputs. THE PC 1 QUEUE PREPARED (the hash lane, c1a1f1d7, nothing published): the kit fetch (zip sha256 1d441f5a..., the 8 sub-version 3 packs plus the ladder packs sh256x13/27/53/88 and sh64x52); the class v4 efficiency pass on the 5090 (elevated, --cards-off, a 17-step clock-lock grid from unlocked to 1,400 MHz with the four Ember caps on it, v4 then v3 at each step for about 60 s, the fingerprint on every step, nvidia-smi at 1 Hz, -rgc in finally; nvidia-smi has no voltage-offset lever, the lock walks the driver's V/F curve; about 40 minutes; the owed 5090 clock rows subsumed); the same on the 5080 (its unlocked row the stock point against the fleet's 71.16 MH/s); G1 on the 13 packs and the AMD ladder rows on the 9070 XT (the ladder packs' Mac fingerprints sh256x13 ff8f707210659eb1, sh256x27 892b6d55a7ddcfcb, sh256x53 663c73c61f62cc54, sh256x88 ec7ca430a061fa59, sh64x52 9dd010f79d8ca9f4); item 6's step costs on AMD (run e); the 5080 full Ember Tune (not elevated, the Power Helper carries the rights); the 9070 XT tune pass (the tune_line before and after on RESULT lines); dr736 not in the queue (its packs are not in this tree). The protocol: each exit line to the shipper, the next published on its ack, the shipper's "PC 1 released" line (after the 0.3.21 window host build) opens the queue. [user]'S STANDING RULE (through main, 19:1x UK): no gaps between tasks, the builders run continuously, no lane idle while another has queued work; the Counter lane reports exceptions and clock readings only. [user]'S WORD ON CLASS V5 AND THE CRYPTANALYSIS (19:2x UK): push both tonight. Class v5 split across six parallel lanes: (a) the v5 lane, the generator and verifier with the verifier cost against the 10 ms gate (the first v5 pack cut 18:06:39Z on build-1 under igneum-pow b4059975: Devnet 3's genesis as epoch and era seed, day 20,733, the genesis state's 11 leaves under state root 7e37a9fb19b154d32daf5bf30a50d339a75029fbc9eec9ea20e95439dba5a311, generator 5 attempt 0, program id e5a4ac5978462156, Metal fingerprint 82b19cbde8557ea5 at 2^24 from base 0 on the M5 Max at 18:07Z, the three vector warps bit-exact, the cache FNV 7334fa46e5d972eb, 70 unit tests green, the v5 chain draw equal to the amended v4's instruction for instruction; the pack proto-cuda/packs-ca3-v5/v5-dn3-epoch0 with state.igsd1 and leaves.bin); (b) the node lane, the state commitment into the dataset derivation and the stale-dataset test known-failed first, the digest-compat fields, the day streams, the proof witness; (c) a new kits lane, Metal, CUDA, OpenCL/AMD and Intel with fingerprints equal and a kit zip on build-1; (d) the attack-pass lane, F1, F4, F8 at 2^24 and F9 on the v5 stream on both boxes (held for main's own line to it, since it takes tasking from main only); (e) a new fast-time lane, the ladder climb plus the v5 crossing with a stale node, the digest-compat and the restart cases; (f) the shipper and the node lane, the Devnet 3 crossing as class v5 by activation height on the 0.3.23 node with the apps' kits in the same release, every gate kept; the target the v5 crossing on Devnet 3 tonight, the clock a reading when (b)'s fork draws e5a4ac5978462156 for Devnet 3's epoch 0 and (e)'s cases pass. The cryptanalysis: the lanes tripled (three per target on different question classes), CPU and GPU pods rented within the fleet's ceiling through the fleet's warm-pod shape, the sweeps back to back, first results by 23:00Z (00:00 BST) with the box-hours and pod-hours spent and the bound reached stated honestly (the crypto lane's rule set at crypto-engage 921ea607 carries the no-gaps rule). THE EVENING's EXCEPTIONS AND PLACEMENT (19:2x to 19:5x UK): build-2's mirror master was stale (a4bca198, without the pow fix) until the hash lane fast-forwarded it to build-1's 001e32ec at 18:25:42Z; the three adversarial branches cut from the stale tip are ordered to merge build-1's master before any run or report (master's igneum-pow 017e7037 byte for byte on both mirrors). the project lead's read: build-1 at 11 percent CPU, build-2 at 56; the word both to near max, the lease pool to about 88 cores a box; the cryptanalysis sweeps, the attack-pass family runs on v5 and the v5 kit builds placed on build-1 explicitly, back to back, both boxes above 80 percent until the queues empty. The cryptanalysis tripled: seven of nine lanes running (the three originals re-scoped to one question class each, plus mixer day-key, mixer statistical distinguisher, cache hot-set, acceptance header-grinding); adv-cache-3 (chain-break) and adv-accept-3 (draw exhaustion) blocked by the session's 20-concurrent-subagent cap (only the project lead raises CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS), re-spawned as slots free; plans by 20:25 and 21:10 BST, first results by 00:00 BST; the sweep queue on build-2 live. Pods: no CPU-only pod type exists, and the fleet lane rents none for the pass until the project lead's word reaches it on its own channel; the boxes' idle cores at nice 10 carry the sweeps; adv-accept-2's GPU locality row stays an analytic bound tonight. MAIN's RULINGS (20:0x UK): pods cleared, the fleet lane holding the project lead's word with a USD 200 cap for the pass tonight, each request as purpose, lane, hours and pod type, adv-accept-2's GPU locality row first; the 20-concurrent cap stays (the Mac crashed under agent load this morning; the standing rule keeps running agents near ten), the two unspawned lanes take slots as they free and their questions queue as sweeps on the boxes meanwhile. PC 1: released 18:04:37Z, the kit fetch landed 18:16:36Z, the MSVC host built and off; main's word 19:5x UK: publish the class v4 efficiency pass this minute, then the queue in order. THE EVENING's READINGS (19:3x to 19:5x UK). THE SITE: deployed from the box mirror (GitHub dark) by the build-server lane's tools/site-deploy-from-mirror.sh (a git-less export of the site tree at the mirror commit, node site/build.mjs, vercel deploy from the export; reads nothing from GitHub): master 04c4d9bc live at 18:45:40Z with the 37-row bench table; the sortable table (every header sorts, default MH per watt; the current class as the table, 31 rows; the six earlier-class rows collapsed; the note on why the rate fell from the genesis program to class v3 and v4) landed on the mirror's master at ee5214dc at 18:46Z, its deploy ordered. THE PC 1 EFFICIENCY PASS: the first publish (18:27:57Z, elevated) failed exit 251 at 18:31:14Z, the UAC prompt unanswered in two minutes (nothing measured, the card restored); republished as -b at 18:39:23Z through the installed app's Power Helper task (the project lead's one approval of 6 October, the same channel as the app's Ember tune; the verbs dev, lgc and rgc only; rgc in the finally; the fallback measured-only rows if the helper is absent). [user]'S STANDING RULE (through main, 19:4x UK, recorded): no PC job may raise a UAC prompt or need a click, ever; elevation only through the installed app's Power Helper task; without it the job reports "no elevation path: Power control off on <machine>" with the measured-only rows and never prompts; the "no prompt" assertion goes into the job runner's publish check (a job with --elevated and no helper path refused at publish; the hash lane commits it); the Arc driver retry cancelled tonight under the same rule; the project lead away until about 20:40Z (21:40 BST), one table to him then through main. CLASS V5 KITS (lane (c), 18:34Z to 18:41Z): 82b19cbde8557ea5 read on Metal (M5 Max, 14.163 MH/s GPU time, the build 30.6 ms GPU with the 11 leaves), Apple OpenCL (15.265 MH/s wall), the Metal one-click serve path (a v5 job mined at 18:37:32Z) and the CUDA worker's CPU emulation (--check PASS, 96 of 96 lanes); the fleet 5090 or 4090 card reading asked of the fleet lane; the kit zip packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e (919,129 bytes, 56 files: three packs with leaves.bin, Linux and Windows NVRTC and OpenCL workers, both bench scripts, SHA256SUMS) on build-1; branch v5-kits 3c9ee884; the hosts upload the leaves and free them after the build, no igneum-pow change; the 9070 XT bench placed on PC 1 right after the 9070 XT G1; the Intel row deferred tonight (no B580 on a PC, the Arc path needs a click). THE IN-HOUSE PASS: all nine lanes started (adv-cache-3 at 19:41 BST, adv-accept-3 as a slot freed); plans on the mirror from adv-mixer 1d720654, adv-cache 476e4516, adv-accept d2bc4dc8, adv-cache-2 3d9bcece, adv-accept-2 9b86d4e2; first results from adv-cache (2c7bb6b4: the storage-against-recompute curve monotone toward the full store, f = 1/2 at 0.875x of the full-store chip's rate under equal silicon and 1.26x the ops; batching loses to the stride store from 8 MiB; Q1a, Q2 and Q3 bounds clear; 0.2 box-hours) and adv-mixer (e35556af: Q1 algebraic structure BOUND, no composition cheaper than 8x, algebraic degree at least 16 after one application and saturated after two, 9,360 ops per item stand; the 2^24 day census' best day a 1.17x FPGA multiply-datapath day, 1 in 2^24; 0.4 box-hours); adv-accept-2's reading from the code: the header reaches the hash only through the init words and never the program or the dataset, so a grind cannot amortise across a stream, its bound analytic, the GPU confirmation on the fleet's A5000 pod (rented on the fleet lane's own authority under the word it holds). The crypto lane's rulings: the SIGSTOP yield retired (it kept every sweep stopped while suites held the slots), nice 10 on cores 8 to 95; logs and pid files outside the worktree mirror after build-remote.sh's sync deleted adv-cache's first logs; the lanes' doc paths added to the export exclude list. The crypto lane originates no pod rent on an agent relay (a purchase needs the project lead's own turn); nothing waits on it. Every lane's numbers are the lane's until the defender checks the logs. Box reading: build-1 at load about 161 on 96 cores at 18:45Z, the near-max the project lead asked for; the 8 reserved cores protect the release builds. 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 project lead 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 |
+SUB-VERSION 3 PASSES BOTH GATES (the attack-pass lane, the close line dated 7 October 2026): class v4 sub-version 3, commit 017e70376489251e18564c0abce7e466e606c8b3 on ca3-v4-amend, pairing id a785001687d8688a (verified by the harness). F8's 64-seed census (p2 to p65) at 2^24 nonces each, chain path, window-model control, box 2, 14:51Z to 16:00:20Z: PASS, 60 of 64 under 1.2x (0.9915x to 1.144x); the only four over are the named tail, unattributed and chased: p10 1.5036x (identical to sub-versions 1 and 2, hottest item 0x4004da at 362 reads, no predicted source), p8 1.3776x (0x837de4, 420 reads), p34 1.2505x (0x800010, 541 reads, the one-bit value through sub at 5), p4 1.2167x (0x4000e7, 355 reads); every strong seed gone (p23 4.82x, p19 3.32x, p15 2.57x, p18 2.50x, p56 2.01x all under 1.2x); the worst ratio on the stream 1.50x; the tail's hottest items carry 355 to 541 reads of 2^31 (one to two per 2^22 items above the mean), no chip consequence. The exhaustion gate as it is: the chain-path count on 017e7037 runs slowly (the draw evaluates (c'') at 2^20 per candidate: 20,532 seeds in 59 minutes, 0 exhausted, 0 panics, max attempt 29), so a 10^6 count is two days away and is not waited on; the substance is met by construction (the 256 cap and the last resort unchanged from 8bdcbdd8, 0 of 10^6 there) and by measurement at 017e7037 on 24,631 seeds (0 exhausted, max 29; r about 0.68); the count runs on as a strengthening line. The node's epoch draw now costs about 2 attempts at 2.2 s each, 4 to 5 s per epoch on slower cores, once an hour. The hash lane's ledger at 6941da1d. SUB-VERSION 3 AT 017e7037 IS THE FROZEN GENERATOR FOR 0.3.22 (byte 7), THE TAIL RULED (main, 18:1x UK): the four seeds (p10 1.5036x with its hottest item at 362 reads, p8 1.3776x at 420, p34 1.2505x at 541, p4 1.2167x at 355, of 2^31 reads) are accepted as the window model's unattributed residue at 1.22x to 1.50x with nil chip consequence; the AP-F8-1 row CLOSES with that wording and the hottest-item counts beside it; the 10^6 count on 017e7037 runs on as the strengthening line. The attack-pass plan's start 15 October or the morning after. 0.3.21's FIRST CANDIDATE 96161037 (sha256 f99340b3cf4ce32e, string read back) passed the node lane's two gates on its own binary: the digest gate 15:59:54Z to 16:01:32Z PASS (a89be8a7 with the peers right; db9a85f9 refused, no peer); the mixed-version gate 16:01:53Z to 16:12:05Z PASS (one digest b0afb2ee on five; 252 new and 352 old accepted, 0 rejected; counts equal at 316, 493 and 604 through both clean joins and the restart step; no panic); the node side complete; the fleet's set on the same binary (the kept start with the ids gate, the wipe canary, the cases rerun, the 12 GB line, N15's restarts of p1-5090 and p2-3090-3) is what the shipper's pin word waits on; byte 5 throughout, nothing on the class v4 seam moved. DEVNET 3 NOW (the project lead through main, 17:3x BST): 0.3.22 is the Devnet 3 release (a fresh network object igneum-devnet-3 with every activation at 0 and no override file, the era VDF fork, class v4 sub-version 3 at byte 7 from genesis), the node building on build-1, genesis by 17:30Z (18:30 BST). THE HANDOFF (17:3x BST, inside the 17:40 BST line, to the shipper and the node lane): igneum-pow 017e70376489251e18564c0abce7e466e606c8b3 (the audit-freeze-2026-10-07 tag; 2a111fb1 and 6941da1d above it docs only), object byte 7, PROGRAM_SUBVERSION_V4 = 3, devnet epoch-0 program id a785001687d8688a (must-differ c120d7963abdcd96, 1a4230699a6b9c60, a788661687db4bb3), kit packs-ca3-v4-sub3 zip sha256 4f2445c50c58d76a5544023492d8b858d0b07c5e372d31f9c90c4ce51f829154, the eight fingerprints as recorded (equal on Metal, Apple OpenCL, the fleet 5090 and PC 2's 5090), both attack-pass gates GREEN, suite 103 of 103, CI success; the open items stated as open: the four-seed tail under main's ruling (the attribution for 0.3.23), the 10^6 count as a strengthening line, the 4 to 5 s epoch draw, the owed measurements that do not gate Devnet 3. NO FURTHER HASH CHANGE RIDES ON 0.3.22; anything from the tail goes to 0.3.23. The node lane stages the re-pin on the 0.3.21 tip as its own commit (the fork commit and the kaspa-pow suite line owed to this record); the hash lane's sub-version 3 pairing follows it. THE PUBLIC CHIP TEXT REWRITTEN LAUNCH-FIRST (the project lead: "I thought we were making it 2.1 from launch?"; master 9b996d06, docs/plans/counter-asic-3-public-text-2026-10-07.md): the testnet and mainnet objects set program_class_v4_activation_daa 0, so class v4 is live from genesis and the launch number is 2.1x to 3.9x on day one; the three texts and evidence row 17 lead with that (labels kept), then class v5, then the 5x to 9x only as the class v3 baseline the work started from, the devnet's activation height a devnet fact only; no em dashes, no prize mention, eight columns; with the site lane to apply in the same deploy as the padding sweep (its commit and deploy time owed to this record). STOP-THE-LINE ON MASTER's POW SUITE (main through the CI steward, 18:4x BST): red since 16:31 BST (runs a4eaf76 and 1210158d; program_ids_differ_between_class_v3_and_class_v4_of_one_seed and cpu_recheck_equals_the_worker_reference_for_class_v3_and_class_v4 fail on packs-ca3-v4/v4-devnet-epoch0) because era-vdf's merge 0e2d6b1c put the sub-version 1 line's generator on master against tests/recheck.rs's pins, never CI-run. Ruling: the hash lane merges ca3-v4-amend's tip 6941da1d to master through merge-to-master.sh within 15 minutes, provided its epoch-0 id equals the frozen object's; the coordinator verified the condition (6941da1d's igneum-pow byte-identical to 017e7037; recheck pins 0xa785_0016_87d8_688a with the three must-differ ids; PROGRAM_SUBVERSION_V4 = 3), so no revert; master is the 0.3.23 line after this, 0.3.22 keeps its pin at 017e7037. EXECUTED: ca3-v4-amend on master as merge cf7d6ccb (pushed 16:45:48Z through merge-to-master.sh, try 1; the branch gate GREEN, 55 checks in 317 s on 874e945d; master ac8ed2f1 merged into the branch first with docs/fud-ledger.md keeping both sides and igneum-pow taken wholesale from 6941da1d, byte-identical to 017e7037); master's CI run 37654633279 on cf7d6ccb in progress from 16:46:10Z, the pow job's conclusion owed. Master's igneum-pow is now the frozen sub-version 3 generator. THE 0.3.22 RE-PIN STAGED (the node lane, 16:38Z): fork commit bd710a36 on release-0.3.22-node (igneum-pow 017e7037 archived beside the fork as igneum-pow-amend, byte 7, epoch-0 id a785001687d8688a must-equal, the three must-differ ids, PROGRAM_SUBVERSION_V4 read as 3); the line's candidate fa7f854f (the re-pin, the era VDF merge f2ecf452, the igneum-devnet-3 object with program_class_v4_activation_daa 0 and a one-day signal window, so byte 7 is stamped and hashed from genesis); the kaspa-pow suite on build-2 from 16:38Z, its line owed; the hash lane has the commit for its pairing run. THE kaspa-pow SUITE ON THE 0.3.22 CANDIDATE 21d8f454 (build-2, 16:47:10Z): GREEN 17 of 17; the pairing test reads PROGRAM_SUBVERSION_V4 = 3, epoch-0 id a785001687d8688a, the three must-differ ids hold, the chain draw at attempt 1 against class v3's 0 on the test header; the candidate's digest for igneum-devnet-3 ab9af79f...ed23; the node stamps object 7 (block version 1794) from genesis. MASTER GREEN ON THE POW FIX: CI run 37654633279 on cf7d6ccb success at 16:54:51Z, the igneum-pow job success; master's igneum-pow the frozen 017e7037 byte for byte; build-1's amend checkout synced to 874e945d (its packs-ca3-v4 carries the sub-version 3 kit). THE CHIP TEXTS LIVE (the site lane, site-ui-5): sections 1 to 4 applied verbatim in b34d1932, master cc593483, live on igneum.network at 16:47Z (home row 03 "under class v4 from the first block"; /litepaper#chip-model the new paragraph and the reordered table with the class v3 row last; /miner the new line; /evidence row 17 "2.1x (k = 1) to 3.9x" in eight columns); the litepaper's abstract rewritten launch-first too; tools/ci/ledger-text-check.mjs X35 asserts the launch-first sentence on the home page and "so the launch number is the class v4 row" on the litepaper; no prize mention. /claims (the litepaper's limits section) rewritten launch-first by the coordinator in the litepaper's chip bullet (2.1x to 3.9x under class v4 from the first block with its labels, class v5, then the class v3 baseline "never the launch state", the recompute chip under 1x, the X9 history), /claims rebuilt from it, the ledger text check 61 of 61, the padding and overlap sweeps green in the 56-check gate, master 0a999646 at 17:01:47Z. DEVNET 3's FIRST GO (the fleet lane): dn3-g1 on 21d8f454 up 16:50:21Z from an empty datadir, digest ab9af79f matching the build-1 read, genesis a6fa348e executed (chain id 4463), object 7 / block version 1794, era VDF from DAA 0, no override file, mining from 16:50:42Z; the dn3_ observer running since 16:38:29Z; DN3_GO_DATE=2026-10-07 in /srv/hands/dn3/dn3.env, the hash-origin timer's first dn3_ report 8 October 08:30Z. BUT NO BLOCK ACCEPTED: dn3-g1 refuses every block its miner finds with "the block timestamp is too far into the future: block timestamp is 1791417600001 but maximum timestamp allowed is 1791392281213" (16:57:51Z): 1791417600001 ms = 2026-10-08T00:00:00.001Z, the template stamping genesis + 1 ms, so the igneum-devnet-3 genesis object carries 8 October 00:00Z, one day after the intended 7 October 00:00Z; every block is seven hours in the future and refused before PoW on any node with any miner or pack; the pack is not the suspect (the node draws the epoch's program under its own igneum-pow and hands the pack to the worker; the exported kit is for G1 and recheck only). The fix is the node lane's (a past genesis timestamp, moving the genesis hash and the digest) and a rebuild; the fleet reruns the go within the minute of the new pair; if the corrected build lands by about 17:20Z the pair lock is about 17:35Z (18:35 BST). Spend at 17:00Z about USD 410 of 1,000. GITHUB SUSPENDED (17:0xZ): every push to origin refused with "Your account is suspended" (403); the API reads 404 for the user igneum-labs and the repo (the shape of a suspended account, not a revoked token); with the project lead (the ticket route given by main); until GitHub answers, every lane pushes to the build-1 mirror (/srv/igneum.git over ssh) and the mirror is the record, merges landing through the box gate stamp under the shipper's exception window. This branch is on the mirror at a33a7859 (17:09Z); master's last origin landing 0a999646 (17:01:47Z). DEVNET 3 MINES (the fleet lane, 17:08Z): dn3-g1 on the corrected candidate 69d1b56e (genesis 7 October 00:00Z, hash 4020cb43..., digest 83eb50cdf2eda4cb...) accepted its first block at 17:06:19.920Z, 3 blocks by 17:06:44Z, 0 rejected; dn3-g2 joining, the pair's first common lock about 17:15Z (18:15 BST), then the late joiner and the canary's ten minutes; the hands pair for 69d1b56e landed 17:05:54Z, the shipper ruling whether the genesis bytes are the node lane's or the hands'. Per tier: the first chain that hashes class v4 sub-version 3 (byte 7) from genesis is live on the fleet; the launch-first chip texts describe it exactly. DEVNET 3's EPOCH-0 PROGRAM ID (the node lane): fce15bf61030be57 on igneum-devnet-3 (genesis 4020cb43...b925), read in the 0.3.22 miner's "cache ready" line on build-1 at 17:10:39Z and on dn3-g1 at 17:12Z, the node accepting 85 of 85 GPU blocks; a785001687d8688a stays the shared devnet's epoch-0 id (genesis edc4fa84) and the kaspa-pow pairing pin; the pairing is unchanged, the id follows the seed; the fleet's pack-id gate reads fce15bf61030be57 at epoch 0 and the node's per-epoch value after; the exported kit's eight packs are the shared devnet's seeds, so a Devnet 3 kit, if the hive wants one, is a re-export over 4020cb43's seeds (the hash lane's, on request). DEVNET 3's PAIR AGREES (the fleet lane, 17:19Z): dn3-g2 up 17:17:58Z on the same bytes, digest and genesis, synced from dn3-g1 and mining from 17:18:15Z; the pair agrees on every determined finality checkpoint (20, 21, 22 on identical blocks in both logs); 702 blocks, 0 rejected on either, by 17:19:08Z. The first LOCK lands about 19:10Z (20:10 BST): the object's finality window is 7,200 DAA ("window filling, 681 of 7200") at about 1 block/s, the object as cut, nothing wrong. Two independent nodes agree on the chain since 17:18Z; finality signatures start two hours in; the hash-origin daily (dn3_ tables, DN3_GO_DATE 2026-10-07) counts from today once the observer units are enabled on the shipper's go. The shipper rules whether the genesis is declared on the pair's agreement or on the first lock; the late joiner, the empty-datadir canary (ten minutes) and the spare placing on the hands pair; the standing boxes' second nodes after the joiners on the shipper's word. Spend 17:20Z about USD 412 of 1,000. GITHUB RULE (main through the shipper, 18:02 BST): no push, fetch or poll of GitHub from this lane or its sub-lanes, not even a retry; the box mirror on build-1 and build-2 is origin and the box gate stamp the verdict until it lifts. DEVNET 3's EPOCH-0 PACK EXPORTED (the hash lane): program.json 0xfce15bf61030be57, attempt 0, sub_version 3 (dn3-g1 drew the export path's seed); igneum-pow at 874e945d (byte-identical to 017e7037 and master's), --epoch-hex and --era-hex the genesis 4020cb43...b925, class mx8-eraaf3a9139+sh256x27, day bytes "igneum-day/" || le64(20733) (the chain's rule day = timestamp_ms / 86,400,000; the program and id day-independent, the dataset and fingerprint rolling with the UTC day); Metal fingerprint e510ad92b4d24846 at 2^24 from base 0, Apple OpenCL equal, vectors 3 of 3; on build-1 /srv/artefacts/packs/v4-devnet3-epoch0/ and .zip sha256 e025750f71175ed14d6e2a24e387ebbf1979b1cd0faee9139c41a7671165b334, the sub3 kit zip beside it (4f2445c5...); in the 0.3.22 hive package order with the kit. [user]'S THREE ORDERS (through main, 18:3x BST), with his correction (18:4x BST: everything in-house, nothing external): (1) BUILD CLASS V5 NOW on the 0.3.23 line (program class v5, its own activation height, object byte 6): the generator and verifier in igneum-pow (+0.2 ms per warp against the 10 ms gate), the node side (the state commitment into the dataset derivation, the stateless/stale-chip rule known-failed first on a stale dataset), the hot-set, weak-day and shadow-redundancy rules routed to this class, the attack-pass families re-run on v5, the kit for Metal, CUDA, OpenCL/AMD and Intel with fingerprints equal, the crossing on Devnet 3 by height after its gate, miner cost rows per card; the v5 lane resumed with the order, owing tonight the design-to-code gap in one list and a UTC clock for the first v5 pack. (2) CRYPTANALYSIS IN-HOUSE: no outside firms, no paid lots, no briefs to anyone; the three targets (the mixer M_r, the chained cache, the acceptance rule) attacked by three adversarial lanes that have never worked on the hash code, each given only an outsider's inputs (the public kit, the frozen object, the spec, the attack-pass harnesses), a written attack plan first, a budget of box hours, a report in the attack-pass shape (the claimed break or the bound reached, reproducible); the Counter ASIC lane the defender, main ruling disputes; plans within two hours, first results by tomorrow evening; a coordinator lane spawned for it (docs/plans/cryptanalysis/in-house-pass.md, funding.md's lots rewritten to the internal pass). The served sentence "the cryptanalysis plan buys three external lots" rewritten on the litepaper, /claims and evidence row 17 to the internal adversarial pass, labelled internal, the one outside check staged on its escrow and the publish word (the served text never names the prize or the project lead, by the forbidden-strings gate). (3) OWED MEASUREMENTS on PC 1: released tonight after the shipper's 0.3.21 host build (the line "PC 1 released" about 18:05Z, the window to 17:30Z on 8 October, an elevated job allowed under the standing rules); the hash lane prepares the three job files on the sub-version 3 kit (the 9070 XT G1 and ladder rows, item 6's step costs on AMD, the 5090 clock rows elevated, dr736 if it fits), published one at a time on the coordinator's relay of the line, --cards-off on the measured card alone, the desk left usable; the rows land in the public table with their labels. Nothing in (1) or (3) reaches outside the fleet, the boxes and the PCs. DEVNET 3's PROOF WINDOW OPENED 17:57:05Z (the fleet lane): coverage from 17:44:23Z (dn3-x1, an RTX 3060 12 GB, the sm_86 SP1 floor, from DAA 0; the first segment 1024..1031 submitted 17:45:49Z, 86.1 s end to end); the first records PAID by dn3-x1's key, segments 1256..1263 (0.7036 IGN) and 1272..1279 (1.5993 IGN) in block 1403, paidShards 10 at 17:57Z; the 24 hours run to 17:57Z on 8 October; sizing: a 3060 proves a segment in 84 to 86 s (about 42 an hour) against 450 segments an hour at 1 block/s, 11 cards for a share of one, 15 placed (14 more 3060 12 GB pods on distinct Vast machines, USD 1.07/h together); a runbook rule found: a Devnet 3 block builder carries a proof record only after its own node verified it, so every Devnet 3 miner node needs IGNEUM_PROOF_VERIFIER on the host verifier (the four miners ran bare for 11 minutes, 72 records pending and 0 carried; dn3-j1 restarted with it at 17:54:37Z and carried the first). THE UTILISATION TABLE (the fleet lane, 18:00Z, for the project lead): 36 rented boxes plus the Hetzner seed, 37 GPUs, USD 197 a day at this size, today's total about USD 432 of 1,000; idle in the hour 12 by the letter, 9 of them Devnet 3 boxes rented in the last 15 minutes and syncing, 3 the 0.3.21 warm set standing between windows; the fleet's recommendation: stop Devnet 2's four boxes (USD 59.76 a day; Devnet 3 is now the staging chain) on the shipper's word, fold the 0.3.21 warm set (USD 12.72 a day) into Devnet 3's relay trio when 0.3.21's last line closes. THE PUBLIC BENCH TABLE (main's order): /miners now carries every measured card, 37 rows, with watts, MH per watt, the class v4 cost per card and the tuned state (the 5090 stock 136.1 MH/s at 350 W and tuned 127.71 at 226.8 W; the 4070 at its tune point; the 9070 XT, the M5 Max (no lever), the 5060 Ti, the Arc B580, and the fleet's rented-card sweep of 7 October: the 5080 71.16 at 143.4 W, the H100 248.70 at 385.6 W, the B200 416.35 at 855.6 W, the 4090 hands row 52.25 at 183.1 W, down to the 4060 Ti 20.10 at 77.5 W; every fleet row stock, bench only, labelled measured by the fleet or reported by the fleet). CLASS V5's DESIGN-TO-CODE GAP (the v5 lane, 19:00 UK; section 13 of docs/design/class-v5-stored-state.md at c17bc04b): igneum-pow's class v5 rebased onto the frozen sub-version 3 (v5 and its rungs draw under (a'), (c''), the shared-operand rule, the 256 cap and the last resort by merge); the fork rebased onto release-0.3.23-node with class v5 pinned to object byte 6 counted EXACTLY (byte 7 is v4 sub-version 3; the bytes are no longer ordered by class); the gap about 40 agent hours (17 the v5 lane's): the +0.2 ms per warp on the Mac (1 h), the AP-F4-1 and AP-F1-1 rules with their known-failed cases (3 h each), the attack-pass families on the v5 stream (4 h), the kits (Metal and CUDA hosts uploading leaves.bin, 2 h each; OpenCL/AMD and Intel 3 h), G1 on the 5090 and the Mac (2 h), the Devnet 3 crossing by height (3 h), the miner cost rows (3 h), the snapshot wire's day streams (3 h), the pool's per-epoch state fetch (2 h), the proof witness (4 h), the spec text (2 h); the first v5 pack's clock (Devnet 3's genesis as epoch and era seed, day 20,733) follows the suite and the day-stream bin. THE IN-HOUSE ADVERSARIAL PASS (the crypto lane, 19:05 UK): crypto-engage deb0011e on the mirror; docs/plans/cryptanalysis/in-house-pass.md (outsider inputs and withheld files, the three lanes, the budget 8 box-hours each and 24 total with a ceiling of 48 on the coordinator's word, the review roles, the clock, the label rule); funding.md's lots rewritten to the internal pass with no cash row and every firm name removed; the three lanes adv-mixer, adv-cache and adv-accept spawned 19:1x UK with outsider-only inputs, plans due 21:10 UK, first results by 18:00 UK on 8 October; the Devnet 3 epoch-0 pack added to their inputs. THE PC 1 QUEUE PREPARED (the hash lane, c1a1f1d7, nothing published): the kit fetch (zip sha256 1d441f5a..., the 8 sub-version 3 packs plus the ladder packs sh256x13/27/53/88 and sh64x52); the class v4 efficiency pass on the 5090 (elevated, --cards-off, a 17-step clock-lock grid from unlocked to 1,400 MHz with the four Ember caps on it, v4 then v3 at each step for about 60 s, the fingerprint on every step, nvidia-smi at 1 Hz, -rgc in finally; nvidia-smi has no voltage-offset lever, the lock walks the driver's V/F curve; about 40 minutes; the owed 5090 clock rows subsumed); the same on the 5080 (its unlocked row the stock point against the fleet's 71.16 MH/s); G1 on the 13 packs and the AMD ladder rows on the 9070 XT (the ladder packs' Mac fingerprints sh256x13 ff8f707210659eb1, sh256x27 892b6d55a7ddcfcb, sh256x53 663c73c61f62cc54, sh256x88 ec7ca430a061fa59, sh64x52 9dd010f79d8ca9f4); item 6's step costs on AMD (run e); the 5080 full Ember Tune (not elevated, the Power Helper carries the rights); the 9070 XT tune pass (the tune_line before and after on RESULT lines); dr736 not in the queue (its packs are not in this tree). The protocol: each exit line to the shipper, the next published on its ack, the shipper's "PC 1 released" line (after the 0.3.21 window host build) opens the queue. [user]'S STANDING RULE (through main, 19:1x UK): no gaps between tasks, the builders run continuously, no lane idle while another has queued work; the Counter lane reports exceptions and clock readings only. [user]'S WORD ON CLASS V5 AND THE CRYPTANALYSIS (19:2x UK): push both tonight. Class v5 split across six parallel lanes: (a) the v5 lane, the generator and verifier with the verifier cost against the 10 ms gate (the first v5 pack cut 18:06:39Z on build-1 under igneum-pow b4059975: Devnet 3's genesis as epoch and era seed, day 20,733, the genesis state's 11 leaves under state root 7e37a9fb19b154d32daf5bf30a50d339a75029fbc9eec9ea20e95439dba5a311, generator 5 attempt 0, program id e5a4ac5978462156, Metal fingerprint 82b19cbde8557ea5 at 2^24 from base 0 on the M5 Max at 18:07Z, the three vector warps bit-exact, the cache FNV 7334fa46e5d972eb, 70 unit tests green, the v5 chain draw equal to the amended v4's instruction for instruction; the pack proto-cuda/packs-ca3-v5/v5-dn3-epoch0 with state.igsd1 and leaves.bin); (b) the node lane, the state commitment into the dataset derivation and the stale-dataset test known-failed first, the digest-compat fields, the day streams, the proof witness; (c) a new kits lane, Metal, CUDA, OpenCL/AMD and Intel with fingerprints equal and a kit zip on build-1; (d) the attack-pass lane, F1, F4, F8 at 2^24 and F9 on the v5 stream on both boxes (held for main's own line to it, since it takes tasking from main only); (e) a new fast-time lane, the ladder climb plus the v5 crossing with a stale node, the digest-compat and the restart cases; (f) the shipper and the node lane, the Devnet 3 crossing as class v5 by activation height on the 0.3.23 node with the apps' kits in the same release, every gate kept; the target the v5 crossing on Devnet 3 tonight, the clock a reading when (b)'s fork draws e5a4ac5978462156 for Devnet 3's epoch 0 and (e)'s cases pass. The cryptanalysis: the lanes tripled (three per target on different question classes), CPU and GPU pods rented within the fleet's ceiling through the fleet's warm-pod shape, the sweeps back to back, first results by 23:00Z (00:00 BST) with the box-hours and pod-hours spent and the bound reached stated honestly (the crypto lane's rule set at crypto-engage 921ea607 carries the no-gaps rule). THE EVENING's EXCEPTIONS AND PLACEMENT (19:2x to 19:5x UK): build-2's mirror master was stale (a4bca198, without the pow fix) until the hash lane fast-forwarded it to build-1's 001e32ec at 18:25:42Z; the three adversarial branches cut from the stale tip are ordered to merge build-1's master before any run or report (master's igneum-pow 017e7037 byte for byte on both mirrors). the project lead's read: build-1 at 11 percent CPU, build-2 at 56; the word both to near max, the lease pool to about 88 cores a box; the cryptanalysis sweeps, the attack-pass family runs on v5 and the v5 kit builds placed on build-1 explicitly, back to back, both boxes above 80 percent until the queues empty. The cryptanalysis tripled: seven of nine lanes running (the three originals re-scoped to one question class each, plus mixer day-key, mixer statistical distinguisher, cache hot-set, acceptance header-grinding); adv-cache-3 (chain-break) and adv-accept-3 (draw exhaustion) blocked by the session's 20-concurrent-subagent cap (only the project lead raises CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS), re-spawned as slots free; plans by 20:25 and 21:10 BST, first results by 00:00 BST; the sweep queue on build-2 live. Pods: no CPU-only pod type exists, and the fleet lane rents none for the pass until the project lead's word reaches it on its own channel; the boxes' idle cores at nice 10 carry the sweeps; adv-accept-2's GPU locality row stays an analytic bound tonight. MAIN's RULINGS (20:0x UK): pods cleared, the fleet lane holding the project lead's word with a USD 200 cap for the pass tonight, each request as purpose, lane, hours and pod type, adv-accept-2's GPU locality row first; the 20-concurrent cap stays (the Mac crashed under agent load this morning; the standing rule keeps running agents near ten), the two unspawned lanes take slots as they free and their questions queue as sweeps on the boxes meanwhile. PC 1: released 18:04:37Z, the kit fetch landed 18:16:36Z, the MSVC host built and off; main's word 19:5x UK: publish the class v4 efficiency pass this minute, then the queue in order. THE EVENING's READINGS (19:3x to 19:5x UK). THE SITE: deployed from the box mirror (GitHub dark) by the build-server lane's tools/site-deploy-from-mirror.sh (a git-less export of the site tree at the mirror commit, node site/build.mjs, vercel deploy from the export; reads nothing from GitHub): master 04c4d9bc live at 18:45:40Z with the 37-row bench table; the sortable table (every header sorts, default MH per watt; the current class as the table, 31 rows; the six earlier-class rows collapsed; the note on why the rate fell from the genesis program to class v3 and v4) landed on the mirror's master at ee5214dc at 18:46Z, its deploy ordered. THE PC 1 EFFICIENCY PASS: the first publish (18:27:57Z, elevated) failed exit 251 at 18:31:14Z, the UAC prompt unanswered in two minutes (nothing measured, the card restored); republished as -b at 18:39:23Z through the installed app's Power Helper task (the project lead's one approval of 6 October, the same channel as the app's Ember tune; the verbs dev, lgc and rgc only; rgc in the finally; the fallback measured-only rows if the helper is absent). [user]'S STANDING RULE (through main, 19:4x UK, recorded): no PC job may raise a UAC prompt or need a click, ever; elevation only through the installed app's Power Helper task; without it the job reports "no elevation path: Power control off on <machine>" with the measured-only rows and never prompts; the "no prompt" assertion goes into the job runner's publish check (a job with --elevated and no helper path refused at publish; the hash lane commits it); the Arc driver retry cancelled tonight under the same rule; the project lead away until about 20:40Z (21:40 BST), one table to him then through main. CLASS V5 KITS (lane (c), 18:34Z to 18:41Z): 82b19cbde8557ea5 read on Metal (M5 Max, 14.163 MH/s GPU time, the build 30.6 ms GPU with the 11 leaves), Apple OpenCL (15.265 MH/s wall), the Metal one-click serve path (a v5 job mined at 18:37:32Z) and the CUDA worker's CPU emulation (--check PASS, 96 of 96 lanes); the fleet 5090 or 4090 card reading asked of the fleet lane; the kit zip packs-ca3-v5-20261007T183921Z.zip sha256 e6c088bb34fecdc3ff297dbb06438a14ade7d8c55273357726d28f7a1334a25e (919,129 bytes, 56 files: three packs with leaves.bin, Linux and Windows NVRTC and OpenCL workers, both bench scripts, SHA256SUMS) on build-1; branch v5-kits 3c9ee884; the hosts upload the leaves and free them after the build, no igneum-pow change; the 9070 XT bench placed on PC 1 right after the 9070 XT G1; the Intel row deferred tonight (no B580 on a PC, the Arc path needs a click). THE IN-HOUSE PASS: all nine lanes started (adv-cache-3 at 19:41 BST, adv-accept-3 as a slot freed); plans on the mirror from adv-mixer 1d720654, adv-cache 476e4516, adv-accept d2bc4dc8, adv-cache-2 3d9bcece, adv-accept-2 9b86d4e2; first results from adv-cache (2c7bb6b4: the storage-against-recompute curve monotone toward the full store, f = 1/2 at 0.875x of the full-store chip's rate under equal silicon and 1.26x the ops; batching loses to the stride store from 8 MiB; Q1a, Q2 and Q3 bounds clear; 0.2 box-hours) and adv-mixer (e35556af: Q1 algebraic structure BOUND, no composition cheaper than 8x, algebraic degree at least 16 after one application and saturated after two, 9,360 ops per item stand; the 2^24 day census' best day a 1.17x FPGA multiply-datapath day, 1 in 2^24; 0.4 box-hours); adv-accept-2's reading from the code: the header reaches the hash only through the init words and never the program or the dataset, so a grind cannot amortise across a stream, its bound analytic, the GPU confirmation on the fleet's A5000 pod (rented on the fleet lane's own authority under the word it holds). The crypto lane's rulings: the SIGSTOP yield retired (it kept every sweep stopped while suites held the slots), nice 10 on cores 8 to 95; logs and pid files outside the worktree mirror after build-remote.sh's sync deleted adv-cache's first logs; the lanes' doc paths added to the export exclude list. The crypto lane originates no pod rent on an agent relay (a purchase needs the project lead's own turn); nothing waits on it. Every lane's numbers are the lane's until the defender checks the logs. Box reading: build-1 at load about 161 on 96 cores at 18:45Z, the near-max the project lead asked for; the 8 reserved cores protect the release builds. THE NO-PROMPT RULE IN CODE (the hash lane, ca3-v4-amend 6d84e99f on the mirror): packaging/ota/publish-jobs.sh refuses --elevated at publish (exit 3, the rule and the helper path named); tools/ci/playbook-quit-check.sh rule 3 fails any job script launching with -Verb RunAs or runas /user outside a comment; self-tests (a RunAs launch fails, a Power Helper task start passes, the publisher refuses the 18:27Z line as published, the known-failed case); the pre-push gate carries it. THE CLASS V4 EFFICIENCY PASS, MID-RUN (run-ca3-pc1-v4-eff-5090-20261007-b, PC 1, the 5090 alone, the Power Helper task running with no prompt, every fingerprint matching, 60 s steps, memory 13,801 MHz throughout; rows at 18:51Z):
+
+| Core clock | v4 MH/s | v4 W | v4 MH/W | v3 MH/s | v3 W | v3 MH/W |
+|---|---|---|---|---|---|---|
+| unlocked (sm 2,838 to 2,850) | 136.84 | 475.5 | 0.288 | 136.59 | 330.2 | 0.414 |
+| 2,850 lock | 136.89 | 478.9 | 0.286 | 136.61 | 330.3 | 0.414 |
+| 2,781 | 136.87 | 457.6 | 0.299 | 136.62 | 317.9 | 0.430 |
+| 2,700 | 136.85 | 443.2 | 0.309 | 136.54 | 312.4 | 0.437 |
+| 2,550 | 136.77 | 402.9 | 0.340 | | | |
+
+Reading so far: the class v4 premium at the unlocked clock is +145 W on this card tonight (against the +80 W of the 6 October measurement at a different cap state), and 73 W of it is recovered at the 2,550 MHz lock for 0.05 percent of rate; the rate is memory-bound and flat as the clock falls, so the project lead's thesis holds so far; the steps 2,472 down to 1,400 run to about 19:20Z, the TABLE at the close. THE IN-HOUSE PASS's FIRST FINDING (adv-accept a22d5ba0, 19:51 BST, 1.6 box-hours; the defender's review: confirmed and bounded, no dispute on the numbers): one accepted class v4 sub-version 3 program (seed 100767 in the F8 label space, program id 9d68e6286fc817d4, attempt 2) passes every part of the frozen rule including (c'') (its distinct-index ratio 0.9988 at 2^20, inside the clean spread) and on the live dataset at 2^24 nonces reads a hot set: the top 0.1 percent of items take 2.05x the window model's share (X_f +0.155 percent, X_f/f 1.55; the largest 64-item bucket +294.8 sigma), attributed to one site (instruction 23, source r6, a quarter window; r6 last written by a mad at instruction 4, zero in 0.0126 percent of evaluations, under (c')'s 1 percent) whose hot items are the images of small source values under the era stride at multiples of 2^19; the other sites 0.00 to 0.21 percent against 0.10 expected. The chip price, the lane's and the defender's alike: a 1 MB on-die copy of the hot items serves 0.31 percent of loads instead of 0.15, a 1.002x gain, no chip-model row moves. What is new: the stand-in ratio is a cheap seed selector (the four lowest-ratio seeds of 4,600 accepted programs all beyond 1.2x live, 1.29x to 2.05x; 23 random accepted programs at most 1.043x), and the finding attributes one of AP-F8-1's unattributed tail by value. Routing (main): FINDING (bounded) on sub-version 3, published whole, no change to the frozen object or Devnet 3; MAIN'S ORDER: class v5's acceptance rule carries the fix before tonight's v5 object freezes (a per-site hot-item test at live scale, or a value-level source test, whichever reaches the lineage-fresh constant; known-failed first on seed 100767's shape under the v5 state-derived dataset; the selector's false-positive rate read); if it cannot be in tonight's object with the suite green, the v5 crossing clock moves with the reason. adv-cache FINAL (49ef7747, 0.55 box-hours, 0 pod-hours): every row BOUND (the defender's review: Q1b, Q1c and Q4 BOUND, the Q1b curve equal to its log row for row; Q2 and Q3 readings for adv-cache-2 to confirm); its three sibling definitions implemented as a harness for adv-cache-3, not run. Plans on the mirror from all eight started lanes; adv-accept-3 spawned as the cap gave way. 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 project lead 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 project lead'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