From 2f207c20564b13be7cebf1dc6606808f1edf9bef Mon Sep 17 00:00:00 2001
From: igneum-labs <337424239+igneum-labs@users.noreply.github.com>
Date: Fri, 9 Oct 2026 08:50:47 +0000
Subject: [PATCH] Counter ASIC 3.0 status: main's boxes-first and house-rigs
 rulings, X0's reconciled r0 and X5's first rows (the SRAM die 2.44x
 retained), X1 cut (class v7 +flat), the Metal PASS, D02, the monetary
 conflicts answered, the PCs' state, the hand restart, the mini's reboot leg
 PASS on the moving chain, X4's equivalence, the Token Value re-clocks

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---
 docs/plans/counter-asic-3-status.md | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/docs/plans/counter-asic-3-status.md b/docs/plans/counter-asic-3-status.md
index 74e8966ea..b6548f22c 100644
--- a/docs/plans/counter-asic-3-status.md
+++ b/docs/plans/counter-asic-3-status.md
@@ -859,6 +859,19 @@ THE CHAIN CONTINUES (09:18:05 UK; the decision file's line at 08:19:21Z read by
 THE PCs ON 2.0.2 PUBLIC (the relay lane, 09:16 UK): PC 2's Setup run-20261009-080157 done 09:03 (Igneum-Miner-Setup-2.0.2.exe sha256 7ed06c72f63926060fb2925628d946b5703e105f29ab3eddf788b72986aff5d8, 63,883,409 B, the engine 46f42362 = build-1's record; its return to the relay blob refused by the 50 MB upload cap, the install from the Setup kept on the PC); PC 2 "IGNEUM-APP version=2.0.2 channel=igneum-2.0-devnet edition=public node=igneumd_2.0.2", run win-1ccfe586-20261009-080521, up since 09:05 UK by the fleet install #1831; PC 1 the same version, run win-ae432dc7-20261009-080800, up since 09:08 UK through its OWN update path from the live public manifest the moment the hash lane's runner was free, the duplicate install task withdrawn; sync state and the first block not yet read on either. The consequence for main: both PCs carry the PUBLIC edition, which by F14 takes no signed jobs (publish-jobs.sh refusing both ids), so the hash, kit, energy and canary lanes' job channel to the PCs is closed, only the relay agents answering; the lab build (the payload d0986788, channel igneum-2.0-devnet-lab, "Igneum Miner Lab") restores it; the relay lane's default at 09:35 unless main says the house PCs stay public: the lab Setup built on PC 2 through the relay agent and installed over the public one on both PCs (about 15 minutes each, mining off for about three). PC 1's relay agent down since 09:08 (the old engine's queued "start the agent" job started it as a child of the job at 09:07; the engine's update to 2.0.2 at 09:08 ended its job tree, the install-detach class); the founder's one click: the PC 1 bundle at its dl URL (the bundle's credential, his only), extracted, igneum-agent.bat double-clicked (installs the agent as the scheduled task IgneumRelayService under its pid file, hands over, closes); the relay then shows PC 1 seen and its two waiting tasks run (the mining-on read, the five-minute collector); PC 2 needs no click; the workers intake live end to end on build-1 (the listener on 8793, the Caddy route, GET /intake/pcs/health 200 at the edge at 09:12); the PC collectors (#1834, #1835, every five minutes under their own pid files) after each install. The earlier PC 1 slip once: the overnight re-arm loop's hourly relay run #1833 ("wait for 2.0.x", a 50-minute wait) taken by the agent two minutes before the install task, both loops ended by their pid files.
 THE QUEUE LIVE AND THE RED ON workers.json: the standing queue at build-1:/srv/queue/build-queue.md (09:15 UK; one owner lane per box: build-1 build-server, build-2 site, build-3 node (consensus fuzz), build-4 adversary (the floorplan), build-5 research (TV-02), build-6 HEAL (exec and p2p fuzz), build-7 node (the 2.0.3 flows suites), build-8 fleet (kit canaries), build-9 steward (the board rows) with the pow fuzz beside; the Phase 0 lanes' boxes amended by the coordinator's spawn: build-3 D02, build-5/6 TV-02, build-9 TV-01 and TV-04); the reader under /srv/queue/queue-reader.pid every 30 minutes (status.json, faults.log), its first pass at 09:09 UK reading 5 idle boxes (3, 5, 6, 8, 9), a lane claiming a box by its pid file under /srv/queue/pids; the master copy docs/ops/build-queue.md in the gate (ops-build-queue 54d2c57c); the page's load-by-pid on master 2f00773f (09:16); the founder's collapsible per-box panels on the 10:30 publish. The steward's served-file guard (tools/ci/served-identity-guard.sh, branch ci-served-identity 2437b331, lands by 10:15; every publisher of a served file built from reports calls it before the write, a hashed token list under a per-host salt on the boxes; the daily fetch of six edge copies) read a RED at 09:19 UK: the edge workers.json carries identity patterns (Windows machine names in the DESKTOP form, the word MacBook, the rig names in the PC 1/PC 2 form, a pid line, one private-list line) from the merged reports' host and rig names; the build-server lane's clean republish by 09:45, main hearing it otherwise; the other publishers by 11:00 (the build-server lane) and 10:45 (the site lane's release.json). THE REGISTRY'S TOKEN VALUE SUITES (the steward, master d3686955 at 09:14 UK, ci-token-value 956fcb32): VR 40 rules (PROPOSED requiring ratification, the requirement verbatim), TV 32 gates (NOT RUN), the decisions block with the founder's ratification verbatim (D01, D03, D05, D06 RATIFIED; D02, D04 RATIFIED WITH PARAMETER PENDING); the recorder's decision form; tools/ci/token-value-suite.mjs regenerating both suites at every landing with --check in the gate; the board 20 suites, 262 rows (the 190 evidence cases NOT RUN 175, PASS 8, FAIL 7, 71 in progress; VR 40 NOT RUN; TV 32 NOT RUN). THE 2.0.3 LINE ON FIX 7 (the node lane, 09:16 UK): release-2.0.3-node-k7 = 27f54124 on both node mirrors (the HEAL lane's fix 7 0156e250, the held scratch set: four scratches by tip hash, the walk stopping at any, one round per later cut on a two-island node, its rules branch e434ebd3; + the record store c90212e3 + finality a349ae7b + the fd limit 13886a30 + the fd-class transport close 76cc5a88 + the standing-peer IBD probe bf60948a + the bump); seven suites green on build-8 at 09:14 (p2p-lib 25, p2p-flows 44, miner 33, exec 102, core 186, consensus 146, pow 19); its evidence kit /srv/workers/fleet/27f54124-node-lane.tgz sha256 19f7950b36310e4633cb70d0c74598328eb3386118abd4fefd2863ce1ff10ce5 (igneumd 027bb12aedf1bb9f "igneumd 2.0.3", the cutter 09:17), superseding bf60948a as the canary candidate on the moving chain.
 
+MAIN'S RULING ON THE MOVING CHAIN (09:3x UK): (1) the 2.0.3 candidate 27f54124's rule-33 canary on a wave-two box the moment RESET END is written (a fresh datadir, SEED the three hubs, the eight lines to the moving tip, a shard proved), verdict by 11:30 UK; on PASS the hubs upgrade one at a time to 27f54124 with the chain watched between each, the fleet after by the same waves; on FAIL 2.0.2 stays and the class is named; (2) the entries in order after the one-tip line: a fresh Mac install's canary (the shipper) lifts the download page note; PC 2's Windows canary publishes the Windows entry; lp-4090-43's hive-entry canary on ba08da7f publishes the HiveOS entry; (3) the 2.0.2 declared-load hold and the F07 rows on the converged chain, two hours from one tip; one line to main at RESET END (the island count, the within-ten count, the seeds' state), then at each entry.
+THE FOUNDER'S 1.5x ORDER AND THE FAN-OUT (09:2x and 09:4x UK, relayed by main; the lanes opened by the coordinator at 09:5x): "get the blocks coming in, then focus everything on the 1.5x target, all resources" per his brief (docs/plans/igneum-2.0-master/1p5x/igneum-1p5x-experiment-brief.md, the scratchpad copy sha 74ee073286d4a6e2); the ops lanes keep the network and the entries; every research lane and idle box on the programme; the permanent rule: no X waits for another unless it consumes its output, one lane per experiment, one box or rented pod per variant, from minute one, the clocks the fan-out's time. The matrix with owners and clocks: X0 (10:15) the adversary lane with the hash lane: the reconcile of the latest model with the exact frozen software (the pack and commit ids, the wall-power boundary, the node scaling, layer-8-off (closed, the layer stays) and compaction out, the served 1.9 to 2.1x against the placed 2.4x line by line, the window-conditioned 72 percent against the flat control), the reconciled r0 and its range; X1 (11:30) the research lane a6decc31f55cfbd00: the class v7 shape (no fixed per-load-site windows, everything else as frozen) cut as one flag on 1a938abe4 with its own fingerprint by 10:15, the hash lane confirming the pin, the kit lane cutting its GPU kit; the census, the live memory traces and the cache-hit curves by capacity (0.25, 0.5, 0.75) and policy on three hosts in parallel; native correctness and generation availability; the GPU rows at stock and the knee on two 5090 and two 4090 pods rented now; X2 (11:30) the worker lane a884c7bf29d150268: serving-mode result compaction, exact output equivalence with overflow and sentinels, measured wall joules and accepted throughput, no invented percentage, on a 4090 pod; X3 (11:30) the Ember lane ae16e56dd365a2ea6: workload-specific Ember tuning, the grid stated first, paired energy and rate, stability, rejected shares, the same thermal policy, on a 5090 and a 4090 pod; X4 (11:30) the k lane adf5683e5bcd9042b: byte-identical reg64 alternatives (direct, generated, prefix), the differential test first, the GPU side on a 4090 pod and the chip side through the adversary lane's model; X5 (12:30) the adversary lane with its floor lanes on six rented CPU hosts: the co-optimised opponent sweep on the control now (pure DRAM, hybrid 0.25/0.5/0.75, full SRAM, recomputation), the floorplan for the converged SPEF row inside it on build-4, the hybrid rows re-run on X1's traces; X6 (14:00) the steward blind: the held-out families and seeds sealed now, the GPU cohort, P04 same and one node ahead, P03, P12. The common brief at the coordinator's scratchpad 1p5x-common-brief.txt (the frozen pins, layer 8 closed, the rejected candidates as controls, the arithmetic as the yardstick in every line: from 2.0x a 25 percent GPU cut or a 33 percent specialist rise or 15 plus 20, r_new = r0 * g / a; no pass until X6); the fleet lane renting the pods under the day's ceiling with the chain first; the landing hand landing the brief with the matrix by 10:30 ("UK" for the ten BST tokens, "the Windows desk host" for the quoted rig name).
+TV-04's FIRST LANDING (95027a4f, 09:31 UK, NOT RUN; docs/plans/igneum-2.0-master/token-value/phase0/tv-04/ from tools/token-value/tv-04/, every run on build-9 under a pid file, a red first then 9 of 9 green, five mutants caught, two byte-identical generations): under the ratified cap with zero external jobs the subsidy reaching miners (80 percent) and internal provers (20 percent) falls to 25.96 percent of year 1 in year 5, 6.49 percent in year 10, 0.20 percent in year 20; keeping the miners at their year-1 level by year 10 takes tips of 0.94 of the year-1 subsidy a year; minimum validators and maintenance get nothing from the protocol at any horizon: TV-04 cannot pass on the ratified routing alone, nine BLOCKED rows (the panel's role costs and its viable or collapse calls; a named payer for validators and maintenance under D03 and the P13 deferral; the provers' share of the tip: executor.rs pays the 80 percent to the miner only while the litepaper and the design say "miner and provers"; two D01 conflicts: the litepaper's tail-vote clause and the 1 percent tail in the testnet params at params.rs line 2014); the reconcile against TV-01's spec when it lands. TV-02's FIRST COMPARISON (09:3x UK, NOT RUN): A (Rust on the node's issuance code at bf60948a, build-5) against B (Python from the litepaper's words, build-6): B diverges at genesis (A issues 0 at height 0; the rule: genesis never enters the DAA window, window.rs 176-179 and coinbase.rs 139, block h carrying DAA score h minus 1), a one-block shift in height only, the terminal supply identical (3,963,038,999.610200483289816 IGN on devnet-4, 36,961,000.39 IGN under the 4,000,000,000 cap, 0 issued after the terminal block at height 4,102,488,000); with that rule written in, A and B agree to the integer at every row on devnet-4 (1 bps, 18 decimals, 66 transitions), mainnet (10 bps) and devnet, and on both hand-written reorg fixtures (the ramp end, the first halving), the no-undo known-failed ledger refused; the node's two issuance paths agree on all 46.0 billion blocks replayed; the rule into TV-01's spec (the coordinator's line to the TV-01 lane with the TV-04 conflicts).
+THE MINI'S RECOVERY LEGS (the shipper, 09:2x UK): the app supervisor live under launchd (~/Library/LaunchAgents/network.igneum.miner-supervisor.plist sha256 3cc755b680cb37e1, ~/igneum-supervisor/mini-supervise.sh sha256 600d8dba7bce52b5, KeepAlive, RunAtLoad, pid 9720 since 09:26:27; it reopens Igneum Miner through LaunchServices when the engine has not answered api/state for 90 s, waits out the app's own updater, never a second instance); the update-path leg PASS: api/quit at 09:26:55 UK, the engine gone in 3 s, the host ended by its recorded pid, the supervisor's "engine down for 91 s; opening" at 09:28:28, the engine answering at 09:28:30 (95 s), the node back at its kept tip at 09:28:45, one instance; the relay agent under launchd (the relay lane, 09:30 UK: network.igneum.relay-agent.plist sha256 cc43ce95, KeepAlive, RunAtLoad, ThrottleInterval 10, outside the app's tree; the kill test: pid 10549 killed at 09:29, back as pid 10674 in 12 s, registered on the relay); the reboot leg BLOCKED on automatic login (no autoLoginUser; a LaunchAgent runs only inside a logged-in GUI session, the miner a GUI app; the setting the founder's one hand, System Settings > Users & Groups > Automatic login; the shipper's default if no word by 11:30: the mini not rebooted today, the clock after the setting ten minutes); the three lines into the Mac entry's record (a "recovery" list on the mac block) by 11:45. The PCs (the relay lane, 09:24 and 09:3x UK): PC 2's node syncing from genesis on the new datadir (the chain's blocks accepted at DAA 10,356 by 09:10, 5 of 8 outgoing peers at 09:19); its relay agent (the 8 October build) wedged as the install's relaunch made the app the agent shell's child (the 21:27 class understood); both bundles re-baked from the fixed agent at the same links; PC 1's one click placed as relay item #1836 in the founder's folder ("PC 1: restore the relay agent (one click)"), PC 2's note by 09:35; the watchdog tasks (IgneumRelayWatchdog every 5 minutes and at logon, starting the IgneumRelayService task when the agent's pid is dead; the agent never again started by an engine job) queued for both PCs and into install-agent.ps1; PC 1's software path checked and none works (the api on 127.0.0.1 only, no job channel on the public edition, PC 2's agent wedged, a boot task needing a reboot), the verdict line by 10:30 and only then the click as the fault it is; the recovery canary (kill the agent, update the engine, reboot) after the watchdog is on a PC.
+THE FAN-OUT'S PODS AND THE RUNPOD LIMITS (the fleet lane, 09:34 UK): each pod through the handover gate (the card idle at handover, a real cudaMalloc, driver 570 or newer, a failing host destroyed and a second tried), the rows one-shot with the lane and destroy_on "done <label>": X3 fk-x3-5090 and fk-x3-4090; X1 fk-x1-5090a, -5090b, -4090a, -4090b; X2 fk-x2-4090; X4 fk-x4-4090; nine CPU hosts on Vast (32 vCPU or more, 64 GB: fk-x5-cpu1 to cpu6 for X5's variants, fk-x1-cpu1 to cpu3 for X1's census); each lane told its ssh line and pid-file path the minute its pod passes. The ceiling not raised: today's UK spend USD 547.78 at 09:02; the fleet about USD 56/h plus the fan-out's USD 11/h, the UK day near USD 1,800 against 3,000; lib/spend.py raised to 10,000 with the founder's word quoted only when the hourly line projects past 2,500. The binding limits RunPod's own: client balance USD 814.08 at USD 55.9/h (about 14.5 hours, empty near midnight UK), the account spend limit USD 80/h refusing new pods past it; Vast credit USD 1,307.10; the asks to the founder through main: a RunPod top-up today (USD 1,000 through tomorrow morning) and, if the fan-out's GPU pods are to be RunPod's, a raise of the USD 80/h limit in his console; the CPU hosts on Vast for that reason. The reset: recipe v5 running, the hubs past DAA 47,300. MAIN'S WORD ON TV-04's CONFLICTS (09:3x UK, relayed to the node lane and the TV-01 lane): the three code-versus-text conflicts resolved by 11:00 UK in TV-01's spec with the superseded sentence quoted (the provers' share: the node paying the 80 percent to the miner alone against the litepaper's "miner and provers" and docs/design/proving-payment.md's 90 percent of pgas x f_p to the block's proving pool, the spec stating the rule today and what activates when; the 1 percent tail in the testnet params removed on the 2.0.3 line or named devnet-only with its clock; the year-five tail-vote clause recorded as superseded by D01); the security-funding result to the founder from main as the D01 gate reading, no lane changing a monetary rule on it; the devnet-4 terminal supply a devnet schedule, not the 4 billion cap, TV-01 stating both schedules and which is mainnet's. X2's PREMISE FINDING (the worker lane, 09:5x UK): the serving-mode compaction already exists in the frozen control: the CUDA kit worker's GPU select pass (proto-cuda/nvrtc/worker.cpp, igneum_select: hits behind an atomic counter, 34 sentinel words, a full-read fallback over 256 hits; 12cb65cc4, review B F10, on master 964cdf6c, an ancestor of 1a938abe4, ON by default, --readback full the uncompacted path; OpenCL since 5 Oct); F10 measured it at stock on a 5090 and a 3090 (docs/analysis/review-2026-10-08-b/f10/rows.md): plus 10.5 / 8.8 percent rate, minus 2.8 / 6.7 percent J per batch against the full read; the record's E_GPU rows come from the kernel timer, which the select path already matches (70.41 against 70.86 MH/s on the 5090), so g near 1.00 against the ratio's own E_GPU: the full read was a serving loss above the yardstick, not a saving below it; X2 adds no flag and touches no .rs; it delivers the equivalence on the three same-work contexts (1,048,576 nonces each, both paths, the day switch, the 8,192 boundary, the sentinel and overflow cases: exactly 256 and 257 hits in a chunk, a target equal to a hash, a target of 0 and of all-ones, known-failed first via IGNEUM_READBACK_FAULT_TEST), the 4090 same-session pairs at stock on fk-x2-4090, the knee rows BLOCKED on a rented pod (no power-limit or clock-lock on it; the owner the hash lane's Windows desk host lock), the arithmetic with g against the full path and against the kernel timer; by 11:30. X6 SEALED (the steward, 09:34 UK): sha256 2949e6a96e94a53017e234072b34359a30fe391e25e292269355dcee6dac76cf of ~/.config/igneum/x6/sealed-20261009.json on the Mac (mode 600, never in a repository or on a box until X6): five held-out instruction families by index into the frozen tree's family table (a renamed or reordered table on a candidate itself a finding), twelve held-out seed tuples from the Mac's system entropy, the P04 cohort (5090, 4090, 7600, the Arc), the measures and the verdict rule (PASS only by the blind run's own rows); the file revealed with X6's rows, its bytes hashing to the commitment or the run void; X6 from the minute X1's and X5's rows exist, the document by 14:00 UK.
+
+MAIN'S RULINGS (10:2x and 10:4x UK by main's clock; the Mac's clock about an hour behind those stamps): (1) boxes first: the 1.5x matrix uses the nine build boxes for every CPU part before any rented CPU host (X5's six opponent designs on build-2, 3, 5, 6, 7 and 8; the census, the traces and the hit-curve simulations on the boxes; the TV-02 replays and the D02 generator on the boxes; a design needing more than a box's 125 GB the only renter; the Vast CPU rentals for X5 cancelled: none stays rented); the build-server lane's per-box job list by pid to main at 10:35, a box with no matrix job a fault with its owning lane named; the page showing the matrix jobs per box, the rig cards "agent not reporting since <time>". (2) Both house PCs BLOCKED for every lane (the public 2.0.2 edition takes no jobs; the founder asks for no more hands; the relay lane builds the software path: the 2.0.3 public entry whose apply ends the old app's process tree and installs the watchdog agent, the sha by 13:00, the rule-33 question for main); X6's AMD and Intel cohort rows from rented RX 7600 and Arc B580 (or the nearest class Vast has) or NOT RUN with "the house rigs unreachable"; no lane names a PC as a dependency until the relay lane's "agent back" line. (3) Every X output to main at its clock as it stands, PASS, partial or BLOCKED, never late and silent. The common brief amended with the three.
+X0 (the adversary lane with the hash lane's pins, docs/analysis/class-v6/1p5x/x0/README.md at f5a82bf6a on class-v6-adversary; the pins table and the eight closing reports by sha256 at build-1:/srv/artefacts/tas/x0-pins/; the GPU row recipe for X1 to X4 at build-1:/srv/artefacts/1p5x/gpu-row-recipe.md sha cbab453c8f168b89): THE RECONCILED r0, the complete machine, the placed core, same node (N5), the served convention (the 5090 at its 1,300 MHz lock, 2.319 microjoules per hash, job run-ca3-pc1-v4-eff-5090-20261007-b, report ead72106, the installed worker d7a413c7, pack v4-devnet-epoch0 e370fb2080b7dbb1, board power at 1 Hz, mean from 8 s in, idle not subtracted, no wall meter): the pure DRAM board 1.5x (1.3x to 1.7x), the worst credible opponent; the stored-half hybrid at the reconciled mean hit 0.581 1.93x (1.65x to 2.12x), X5's starting row and X1's target; the hybrid at the p98 tail 0.72 2.09x; the N2 SRAM die 2.3x; a node ahead (N3, x0.50 claimed) 1.8x and 2.4x; the served 2.4x (the synthesised core at the tail hit on the board identity) and the 1.9 to 2.1x (the placed core at the mean and the tail on the complete machine) reconciled line by line; the 72 percent the window-conditioned tail read (0.581 the mean, 0.72 the p98 tail, 0.50 the flat control with the windows off); layer-8-off in no figure (closed, the layer stays); compaction in no figure. THE THREE ASSUMPTIONS MOST LIKELY TO MOVE r0: the boundary mismatch (the card's board power against the chip's whole machine; about plus 0.1x; a wall meter the measurement); the numerator class v4 on a 7 October job while the frozen class v6 object reads 4.011 microjoules at the lock on the l8off kit (if class v6 truly costs the 5090 1.7x class v4, the chip's shadow op count is re-read and the board ratio moves toward 2.6x; the decider a paired class v4 against class v6 control in one session on one card, BLOCKED in X0, ordered to the hash lane on a rented 5090 fk-x0-5090 by 11:30); the SRAM macro access energy (2.0 to 7.0 pJ) and the hybrid's hit rate.
+X5's FIRST ROWS ON THE FROZEN CONTROL (the adversary lane, 09:46 UK; tools/1p5x/x5/sweep.py on build-4; docs/analysis/class-v6/1p5x/x5/README.md at 697278763 on class-v6-adversary): complete machines on the placed core (6.55 pJ per lane-op at N5), the 5090 at its lock 2.319 microjoules over the machine, same node, with the band: the pure DRAM board 1.52x (1.25 to 1.68) at USD 4.84 per MH/s; the hybrid at the hottest quarter (hit 0.341) 1.70x (1.41 to 1.86) at USD 3.98; the half (0.581) 1.88x (1.56 to 2.06) at USD 3.12; the three quarters (0.720) 2.02x (1.68 to 2.20) at USD 2.52; the full SRAM store at a 1 kW budget 2.44x (2.05 to 2.66) at USD 1.20 per MH/s; recomputation (the half stored, the rest derived, 502,000 extra ops per hash) 0.43x, never credible; a node ahead 1.79x, 2.05x, 2.32x, 2.53x, 3.27x. THE WORST CREDIBLE DESIGN RETAINED AS a: the full SRAM store, 0.951 microjoules per hash (0.872 to 1.129), its capital inside the credibility gate (under 10x the card's USD 15.9 per MH/s), so the ratio the programme must bring to 1.5 is the die's 2.44x same node (3.27x a node ahead), not the board's 1.5x: from 2.44x a 39 percent GPU cut at a = 1, or a 63 percent die rise, or 15 and 20 together to 1.73x; X1's traces move only the hybrid rows (the die serves every read whatever the distribution); g BLOCKED on X1; the numerator on the frozen class v6 object BLOCKED on the hash lane's paired control. The per-variant boxes (the coordinator's assignment, the lane's default): dram build-2, hyb25 build-3, hyb50 build-5, hyb75 build-6, sram build-7, recomp build-8, each with its queue entry and pid file; each box running the co-optimised core (four hash contexts interleaved per lane group on a 256x256 macro: a quarter of the lanes, about a third of the core's silicon and leakage) synthesised and placed at that variant's lane count and floorplan, moving the die's capital (USD 1.20 to about 0.6 per MH/s) and its a by a few percent; the converged SPEF row of the 32-lane core on build-4 from 09:35.
+X1 CUT (the research lane, 09:50 UK): the class v7 candidate as one flag on the frozen tree: branch 1p5x-x1-v7 on the box mirror, sha a99de07cded99a9daa0e5bbd1a6192783322347e (parent 1a938abe4: e4b6bb633 the flag and tests/v7flat.rs, a99de07cd the CLI's --flat); the class string mx8-erad810f22d+sh256x27+state+reg64c+fold+rw+flat (LoadClass::flat, suffix "+flat", program-id literal "v7flat/", not +nowin); the candidate's igneum-pow/src fingerprint b9b5d0d1448bc9ce4003b119c66c5734c554c4a84e95b008476a02fb69ae2b54, the merge-base's 5f4d6dc6 read back; the shape: no window draw at all, every load site win=0/off=0 addressing the whole live dataset by the frozen stride with the fold; the family set, fold, rw1, state, reg64c and the dataset policy as frozen; the generator 6 pack shape (no packfile, Metal or CLI V7 edits); on the signing pair (af89be5d, era edc4fa84, day 20730, node1 state) the candidate's program 34d0b421b17fe803, the control reproducing 2a1d6caab4c24564 on the same tool; the suite on build-9 (cargo check --all-targets clean; the release test run green so far, lib 75/75; one pre-existing debug-profile fault on the frozen tree, tests/derive.rs stats_beside_x8 multiply overflow under cargo test without --release, the gate running release). The meaning: at the program level the candidate is the closed +nowin row's address form (the two differ in the draw stream and id only), so the GPU side is expected to repeat register row 18 (the knee rate minus 5.7 percent, joules plus 3.6 percent, g about 1.04 at the knee); X1's new evidence is the cache side, running: the exact window model on the frozen signing program gives the hottest half 56.25 percent of reads (not 72), the candidate 50.00 exactly; the census moved from the rented host to build-8/build-9 under queue entries, the cache curves on build-9.
+THE METAL SAME-WORK CELL (the hash lane, 09:42 UK): ran on the mini 08:30:34 to 08:31:19 UK and PASSED all nine phases (contexts 01, 02, 03 x phases 1 to 3, each 1,048,576 of 1,048,576, 0 missing, the day switch through prepare; the nine JSONs on build-1 under /srv/artefacts/tas/same-work-20261008-0N/metal-<N>.json, the worker igneum-worker-metal sha 5aeddcba unchanged); with the OpenCL cells every reader agrees on all three contexts: engine, CPU reference, CUDA, pool, OpenCL, Metal; the fleet lane never relayed the job's end line (its fault, once); the hash lane's own stamps of the morning ran about 25 minutes ahead (said once; the X0 pins and the recipe sent at about 09:35 UK). D02 LANDED (0673ecf8, 09:44 UK, NOT RUN; docs/plans/igneum-2.0-master/token-value/phase0/d02/, README sha 593b8c95, the generator 0133addf): three same-cap schedules (A the current pacing from release-2.0.2-node a284380b; B the volume's longer distribution taken literally as the same curve on a time axis x2, with x3 and x4 swept; C the lane's smooth monthly glide with a 4-year half-life), all valid EmissionSchedule parameter sets, the node's own Rust EmissionTable printing byte-identical rows to the generator on build-3; by year 2 A has issued 49.1 percent of the cap, B 24.5, C 28.7; in year 1 B pays 50 percent and C 63 percent of A's subsidy; a miner joining in year 10 gets 6.25 percent of the year-1 block subsidy under A, 25 under B and 21 under C; at the 1x budget (the TV-04 multiple, the 1-percent-of-cap yardstick) fees must exceed the subsidy from year 13 under A and year 21 under B and C; the meaning: a longer same-cap schedule halves early security income in native units for later entrants and later fee dependence; the founder picks (BLOCKED row B0); the litepaper's emission sentences describe the curve without the ramp (the code 3,900.5M by year 12, the litepaper 3,938M). THE THREE MONETARY CONFLICTS ANSWERED (the node lane to the TV-01 lane, 09:37 UK, with the line numbers): (1) two separate 80/20 rules ship today: the subsidy (80 percent to the producer, 20 percent to the proving pool, which pays provers per accepted proof record) and the tip (80 percent to the miner alone, 20 percent to developer registrations); the litepaper's "80 percent to the miner and provers" at site/litepaper.html 614/622 the superseded sentence; the design's 90-percent-of-proving-fee route only on the unmerged proving-payment branch, its activation param absent from 5d53a591 and 27f54124, no activation scheduled; (2) the 1 percent tail at params.rs 2014 is igneum-testnet-1's own approved genesis schedule (ledger-decisions.md 110, E22), not mainnet's; kept, named testnet-only; mainnet (1858) and devnet-4 (2372) carry CURRENT with Tail::Cap; binding the testnet to D01 a founder's genesis change, not a 2.0.3 commit; (3) the year-five tail vote superseded by D01, the litepaper source saying so at line 630; devnet-4 and mainnet run the same per-second curve, the 36,961,000.39 IGN devnet-4's own sum at 1 bps/18 decimals, the 4 billion cap mainnet's at 10 bps. THE PCs (the relay lane, 09:37 to 09:38 UK): both relay agents alive, holding their lock, not polling since 09:07 (PC 1) and 09:04 (PC 2), the 8 October wait class (waiting on the app process tree a task relaunched); the "died with the engine" read at 09:16 the wrong cause, said so; both nodes on the chain verifying peers' proofs and locking checkpoints 266 to 270; the "process to end" line to the founder cancelled; both PCs BLOCKED for everything of the lane's; the bundle hardening landing (agent-watchdog c40103e6 + e5fd3f86: the agent as a logon task with a five-minute watchdog, re-registration when missing, a stale lock cleared by pid on every install); the software path the public edition's own OTA channel: the shipper's answer, no 2.0.2 rebuild, the next public entry 2.0.3 whose apply helper ends the old app's process tree and proves the new engine answers api/state before it exits (two commits on release-2.0.3, the sha by 13:00 UK), its Windows install running install-agent.ps1 from the bundle on each PC; the house PCs on the public channel and root, the lab install through the freed agents after; the one decision for main: the rule-33 rule (the 2.0.3 entry only on a PASS block to a moving tip) or its lift for the house PCs. THE HAND, SEED AND LIGHT-READER ONTO THE HUBS' CHAIN (the build-server lane, 09:47 UK, under /run/dn4-restart.pid): the explorer indexer reads igneum-dn4-hand itself (IGNEUM_EVM_RPC 127.0.0.1:26870, igneum-observer-dn4 the observer.mjs reader on it), so the hand and its seed go onto the hubs' chain (fresh datadirs with the old aside, the 5d53a591 kit's igneumd, the three hubs as the only peers, --nodnsseed, the same ports, the units re-enabled, the within-ten read-back every minute); the light-reader the same way (the coordinator's conditions: its old datadir aside never deleted, igneum-light-service repointed in the same step); the indexer and the reader restart once the hand answers.
+THE MINI'S REBOOT LEG PASS AND THE MINI MINING ON THE MOVING CHAIN (the shipper, clocks by date): the restart by its own command at 09:45:02 UK; ssh back at 09:46:14 (71 s); the igneum console session logged in by itself (the FileVault pre-boot carried by the founder's setup; no hand); the app relaunched by the login session at 09:45:59; the supervisor under launchd at 09:46:04; the relay agent registered at 09:46:16 (a first try failed on the network not yet up, launchd respawned it); the node back at its kept tip and at 09:47:20 synced on the moving chain, DAA 49,960, 6 peers, epoch seed 284fb765, mining, a block accepted at 09:47:16; the tip feed (network.igneum.mini-tip, plist f881eb2a) and the workers-page collector (network.igneum.mini-collector, plist 1e9db456) moved under launchd (they had been nohup scripts and did not survive the reboot); the three legs PASS (the agent kill 1 s, the update path 95 s, the reboot 71 s to ssh and about 2 minutes to mining), the reboot line moving from NOT RUN to PASS in the next landing (the first two landed as a3289d19). The feed: lp-4090-04's chain at EVM 26,025, DAA 49,948 at 09:47:02 UK. The entries back in play under the rule (a PASS block to a moving tip): the HiveOS entry canary from c7f43510 armed behind the one-tip read; the Windows entry waiting on PC 2's Setup upload through the relay (its agent wedged) and the 2.0.3 updater fix (the helper proving the new engine answers api/state; two commits on release-2.0.3, the sha by 13:00); the decision to main: the house PCs' only software path the 2.0.3 public entry, which publishes under rule 33 only on PC 2's PASS block or on main's word lifting the rule for the house PCs; the default the rule stands. THE FLEET LANE'S FOUR TAKEN (09:49 UK): every line from the Mac's clock from here; fk-x0-5090 renting at the gate since 09:47 for the hash lane, -lgc never permitted on a rented pod; fk-x0-7600 and fk-x6-arc cannot exist (Vast's on-demand offers 0 for every AMD name queried, the 400-offer scan at 09:36 no AMD or Intel card of any kind, RunPod none), the house rigs the only path for those cells; the Metal end line's non-relay recorded as its fault; wave two (the 40 node boxes) running from 09:48, lp-06's reset onto lp-04/05's chain at 09:50 by the default, the seeds with the build-server lane, lp-4090-43 last, RESET END and the tracker's within-ten count next, then 27f54124's rule-33 canary. THE HASH LANE'S X1 PIN CONFIRMED AND THE X0 BUNDLE (09:50 UK): the X1 candidate's pin confirmed (the merge-base's 5f4d6dc6 reproduced untouched, the candidate's b9b5d0d1, the class string with +flat, generator 6, no gate edits), the kit lane cutting its GPU kit by 10:30; the X0 bundle at build-1:/srv/artefacts/tas/x0-pins/x0-bundle.tgz (sha 864470a7914b9c2b: v4-devnet-epoch0, the l8off kit's hl-v6-all and the signing object's hl-v6-all-cs, the gen-6 Linux worker 804a6f7f, the run script by the recipe; the 7 October numerator's d7a413c7 the Windows exe of the installed app, so the row names the pod's worker and reads both packs through it, the pair being the figure), stock only on a rented pod, said in the row; the 7600 rows and the AMD kit reruns NOT RUN "the house rigs unreachable, no rented AMD card", X6's Arc cell the same, X6's 7600 cell likewise unless main unblocks a rig. THE TOKEN VALUE RE-CLOCKS (main, 09:5x UK, the founder's speed rule): TV-01's complete spec.json by 11:30 UK; TV-02's final agreement rows by 11:30; TV-04's second cut by 12:30 with the panel's role costs from the fleet's measured operator costs and the maintainers' committed figures or BLOCKED with the exact missing number; D02 landed as the founder's decision row; the coordinator's ruling on TV-04's coverage: the USD costs as cost inputs, the budgets in native units, no USD-per-IGN, break-even or purchasing-power figure anywhere (VR-04), each coverage cell a parameter table over the one unknown V (the USD value of one BU) with the symbolic threshold V* = the cost in USD over the budget in BU labelled as a quotient of two inputs, the minimum required capacity (the hashrate and the shards per year) from the node lane or BLOCKED by name. X4's EQUIVALENCE (the k lane, 09:48 UK): the three reg64c forms (the direct closed-sum tree, the generated 8-group-sum form recomputing only the groups written since the last load, the prefix form = review B's F05) EQUIVALENT to the frozen fold on CPU at branch 1p5x-x4 b51d8c04b, the known-failed form refused first (the evidence on build-9 /srv/builds/igneum-wt-1p5x-x4/out/differential/VERDICT: the wrong form disagreeing on all 8 reg64c windows and all 32 fuzz seeds); direct, generated and prefix agreeing with the same-work references on 12 windows of 4,096 nonces (contexts 01 and 02, D and D1, the 1572864 boundary both sides) and with the fold on 32 random epoch/era/prehash seeds x 1,024; context 03 (class v5) the no-op control; the fold export byte-identical to hl-v6-all.tgz 91314310, the prefix export byte-identical to hl-v6-all-prefix.tgz d4b07747 in all 14 files; the meaning: a specialist or a GPU may use any of the three forms with no consensus change; the first paired GPU row running on fk-x4-4090 at stock (8 runs A B C D D C B A); the knee NOT RUN (the container refuses -lgc and -pl); the chip side from the adversary lane's constants: the generated form and the running total 7 nJ per hash (5 to 10) against 56 nJ for the literal fold, because the shadow writes only r0..r7, so a specialist's a is under 1 and the ratio moves up, not down.
+
 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) |