igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

Counter ASIC 3.0 status: sub-version 2 committed (07a809a7, id a788661687db4bb3, the fixpoint rule, seven fingerprints, packs sha256)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 13:03:19 +00:00
parent 6049b406ca
commit 068fa345af
1 changed files with 13 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
docs/plans/counter-asic-3-status.md
View file

@ -355,7 +355,19 @@ the project lead gave the go in advance for tonight: the shipper runs publish 1
| p25 | 1.28x | |
| the 53 passing | 0.9915x to 1.16x | no predicted source |
Three residual classes, all a constant delivered through a writer the rule admits: (1) saturation or zero preserved through rotl, rotr, load or mad (p31, p6, p23, p26, p34); (2) zero from mulhi (p45) and zero preserved by rotates; (3) the iteration boundary, the rule's writer state starting fresh at instruction 0 so an or at 63 feeds a load at 1 (p11). Sub-version 2's dataflow-freshness rule closes all three IF the freshness is computed as a fixpoint over the loop (the state after instruction 63 feeds instruction 0 of the next iteration), with the dynamic (c') count on load sources as the backstop; sent to the hash lane. "No lossy-sourced load by construction" is not true of sub-version 1 and stays held. Chip consequence on sub-version 1 by the 1.4 arithmetic: the hot set is still one item at one site, under 1 percent of rate for a chip caching it, so the ship is safe on the rate side; the auditor's flag is what sub-version 2 removes. F9's hot-set harness cannot be the second re-gate (its metric counts the era's designed half and quarter windows as hot; its chain-path 8.3 percent on sub-version 1 is the window model, verified on its worst seed); F8's census is the single re-gate instrument, re-run on sub-version 2 within the hour of its commit. 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 |
Three residual classes, all a constant delivered through a writer the rule admits: (1) saturation or zero preserved through rotl, rotr, load or mad (p31, p6, p23, p26, p34); (2) zero from mulhi (p45) and zero preserved by rotates; (3) the iteration boundary, the rule's writer state starting fresh at instruction 0 so an or at 63 feeds a load at 1 (p11). Sub-version 2's dataflow-freshness rule closes all three IF the freshness is computed as a fixpoint over the loop (the state after instruction 63 feeds instruction 0 of the next iteration), with the dynamic (c') count on load sources as the backstop; sent to the hash lane. "No lossy-sourced load by construction" is not true of sub-version 1 and stays held. Chip consequence on sub-version 1 by the 1.4 arithmetic: the hot set is still one item at one site, under 1 percent of rate for a chip caching it, so the ship is safe on the rate side; the auditor's flag is what sub-version 2 removes. F9's hot-set harness cannot be the second re-gate (its metric counts the era's designed half and quarter windows as hot; its chain-path 8.3 percent on sub-version 1 is the window model, verified on its worst seed); F8's census is the single re-gate instrument, re-run on sub-version 2 within the hour of its commit. SUB-VERSION 2 COMMITTED: ca3-v4-amend 07a809a7, 13:01Z (origin and build), the string with the attack-pass lane. The loop fixpoint is in as rule (a') in accept.rs (the freshness run to its fixpoint over base then shadow block; every load's source fresh in the steady state, else the candidate rejected and the next attempt drawn; it closes the iteration boundary the draw's fallback cannot see); (c') counts zero and all-ones alike (v == 0 or v == MAX) per load site over the 16,384 evaluations, rejected at 164 or more; both and the draw rule keyed on the class v4 shape on every draw path, so v2 and v3 do not move; mulhi never fresh. Epoch-0 id a788661687db4bb3 (the devnet seed's attempt 0 rejected by the new rules, attempt 1 accepted); must-differ c120d7963abdcd96 and 1a4230699a6b9c60 pinned; object byte 7 in recheck.rs. The seven fingerprints (Metal = Apple OpenCL on the M5 Max, 13:01:17 to 13:01:49Z, the control 90f794dd556f7a3b unchanged):
| Pack | Fingerprint |
|---|---|
| v4-devnet-epoch0 | e370fb2080b7dbb1 |
| era-0 | b7237555d31fc3cf |
| era-1 | b6b167fa15dfe2c9 |
| era-2 | 28bdf65eff33f2c4 |
| era-3 | e26d38c46f3f1b16 |
| era-4 | dd8fdf6ff4f59eed |
| era-5 | 8bf40f5cb858d835 |
Packs zip (eight packs, packs-ca3-v4-sub2) sha256 69c36772cd79e44e2ddd589466d9c64a94a13c9e970e9f27bd76feabb9b4581b. The suite re-runs through master's build-remote on box 2 (the worktree's own script predates --box; the first run died on the flag), line to follow; G1 on PC 2 under --cards-off after it. The sub-version-2 pairing waits on the node lane's re-pin to a788661687db4bb3 and byte 7. For the record, sub-version 1's pairing: igneum-pow 8c728ca3 against b7cc37e7 (8097d600's assert_ne in) 17 passed, 0 failed, rc 0, 12:21Z. 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