Plan 6.13: the 0.3.25 node pin c9ad753a, its gates, digest 2066aa57, the open items for the next line

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/plans/counter-asic-3-node.md

@@ -326,6 +326,9 @@ the project lead's launch-shape proving hold (24 prove-only 4090s, 150 transacti
 | d5b68fae | a transaction is admitted at the larger of the executor's next DAA and the chain's virtual DAA plus one (the relay sink through a closure over the consensus manager; eth_chainId, net_version, eth_sendRawTransaction and the status read the same height), and a chain-id mismatch between the network's own two ids is a refusal with its reason, never a misbehaviour strike: at 12:05Z build-1's seed and node1 read "misbehaving peer: invalid evm transaction …: wrong chain id: expected 4463, got Some(4464)" from peers re-executing from genesis (the fleet's 21 re-walks; every node at a move by the stamp rule), whose pools named the old id for the length of their walk and dropped every peer relaying a transaction; the known-failed test is the re-walk's shape. Beside it, not a bug: the 21 stale nodes (a divergent execution state, so a divergent class v5 dataset) refuse every honest v5 block as invalid proof of work, strike the relayer and ban build-1's seed for an hour; the re-walks cure it; a node that cannot check a v5 header for want of the epoch's state holds off (PowCacheQueueFull, outside the strike rule) |
 | 6ccaf9e9 | the pairing finding and rule 19 (12:29Z to 12:41Z): the fast-time pair on d5b68fae read FAIL on an attempt-3 seed (epoch 9: the miners drew 65b57e3b847d362e, the freeze CLI ebf64b32e2d84c5b); the 0.3.25 line's release worktree carried an untracked igneum-pow copy from before the class v5 freeze (content hash e01ea128fab1: accept.rs without the (c''') per-site floor MIN_DISTINCT_RATIO_V5 and its HotItemSite refusal; generator.rs and memhard.rs older), every 0.3.25 gate artefact linked it through the `.cargo/config.toml` paths override, none shipped; the 0.3.24 line's worktree and the live network were on the freeze (1c420786's igneum-pow/src = 29106189ca1e by the content-hash method, cbc5bd0aa10585c8576e71e37a8ee47a045ae51754e9ddf749d0c21e6a535f88 by the shell's per-file method); builds.jsonl's "pairs_with" names the parent repo's HEAD, never the override copy, so it never said which tree was linked. The copy replaced by the freeze's tree on the Mac and both boxes at 12:34Z. Main's rule 19, the build-time half: consensus/pow/build.rs fingerprints the linked tree, fails the build unless it equals packaging/pow-freeze.txt (IGNEUM_POW_FREEZE_CHECK=0 for a research build), embeds IGNEUM_POW_FINGERPRINT in every binary's strings and start lines; the handshake field on the next commit. The Devnet 3 ceiling re-cut to 90,000 (the publish limit 82,800, about 15:53Z) under the shipper's slip rule; the digest moves |
 | 3ffcf83b (release-0.3.24-node) | the testnet v5-at-0 re-cut, landed by the default rule 30 minutes after the clean crossing: TESTNET_PARAMS program_class_v5_activation_daa 0, program_class_signal CLASS_SIGNAL_V5; the identity test moved; the testnet digest moves |
+| c9ad753a (THE 0.3.25 NODE PIN, named 13:08Z) | 6ccaf9e9 plus igneum-miner embedding the full IGNEUM_POW_FINGERPRINT line (the linker had dropped the unreferenced static). Every gate green at 13:01:14Z: build (igneumd 68526b25…, igneum-miner d4f4c98d…, the fingerprint cbc5bd0a… read back in both), miner 29, core 177, exec 54, pow 19, consensus 134, p2p-flows 38; the Devnet 3 canary with digest 2066aa57505e5ecbd585d061364abb0032d5b5b29cc41c54f4b38cb81c2ba6eb, byte 6, the 0.3.24 pin refused both ways; the testnet canary b2e856ed on this line. On the same node code and object (6ccaf9e9): the fast-time SUMMARY PASS at 12:57:35Z (cross-0325-6ccaf9e9: 12 of 12 ids equal to the freeze CLI's) and the v5 lane's read-back PASS on epochs 8 to 10 (the FAIL's attempt-3 seed at the exact freeze draws 1f1cf82877f46ee6: neither id of the FAIL was the freeze's). The ceiling lands at 90,000 about 17:53Z; the publish limit 82,800 about 15:53Z. The move's minute: the last FETCHED plus ten, about 13:30Z to 13:40Z; the minute's gate the census at 28,462 after the re-walks, the unban per box, the first lock |
+
+Open for the line after 0.3.25: the handshake carries the igneum-pow fingerprint and a node refuses a peer with a differing copy (its own mixed-version gate); `igneum-miner program-id` (the seeded read for the harness's attempt-3 line); the snapshot loader re-numbers a resumed range against the DAG; the claim floor falls back to depth when finality is paused; bans cleared with the state class, not by the clock; node1's tie-break at 09:42Z and the stale nodes' per-node genesis roots, read from their logs.
 
 The crossing at 68,400 (11:48Z to 11:57:40Z, 8 October 2026; clean on the honest side: the first class v5 block at 11:57:40Z, 341 blocks in the next four minutes, seven stale-pack attempts refused, no honest block refused, no state-wait or catch-up fault on build-1's nodes): the chain stalled at DAA 68,403 with no epoch-19 block because every fleet mining box ran the hive package's igneum-worker-cuda, which runs pack generators 2, 3 and 4 and refuses class v5's generator 5 ("WORKER FAULT seed mismatch", 0 MH/s from the first epoch-19 template); the kit's class v5 worker had been benched on 24 cards that morning and never placed on any miner's --worker path. The nodes read 0 PoW rejections and handed the right template (the 0.3.24 miner's CPU id-read on build-1 named epoch 19 class v5 program id 3d375a55029e7e60, equal to the kit lane's). A second requirement read the same minute: a miner not told its node's EVM port (--exec-rpc) cannot fetch the epoch's execution state and never prepares class v5. Lessons: a worker that cannot run the object's next class refuses at the pack prepare, loudly, hours before the boundary (the plug, tune, play rule); no hive tar ships without the class the object names; every miner loop carries --exec-rpc.