Igneum: the GPU-mined zkEVM L1. Node, miner, app, spec and site.
Find a file
igneum-labs 9e0bb8e5fb Counter ASIC 4.0 research: the per-load shadow's duplicate reads, found and fixed (the acceptance test judges the order the class executes)
The finding (tests/ca4_trace.rs, the first export traced at 22:16 UTC): the per-load class derived 10,728 distinct items of 12,288 over three units (the class v4 shape 12,286), 1,482 same-iteration duplicate lanes at sites 8, 10 and 15; the mechanism is not the sub-block's last writer alone: a lossy base writer (mul, mulhi, or) followed by 27 passes of a 16-instruction map collapses a load's source register to 1 to 17 distinct values in 32 lanes before the next load (the census's 29 failing seeds of 64 at 22:16 UTC name mulhi, mul, or, rotl and load as the last base writers). The fix, two layers: (1) generator: a per-load sub-block instruction that writes the NEXT load's source register is redrawn from the injecting families when its op is mul, mulhi or or (consumes a draw; the class's own stream); (2) accept.rs: the dynamic test steps the per-load sub-blocks inside run_unit (alu_step, the same arithmetic as the arms) so saturation and distinctness are judged on the register file the loads read from, and a new rejection DuplicateLanes (the per-load class only) refuses a candidate whose load reads one address in two lanes of a unit; a rejected candidate redraws the attempt. After: the trace reads 12,287 of 12,288 with 0 duplicate lanes on the genesis seed; the 64-seed census on a second dataset reads 1 duplicate pair in 16,384 rows (seed ca4-census/49, the chance floor of a 2^24 index space, about 0.5 expected; the class v4 shape's own 2 of 12,288 are the same floor). verify.rs gains trace_load_indices (the diagnostic). Suite on igneum-build-2: 64 + 2 + 7 + 4 + 19 + 2 + 7 passed, 0 failed (RESULT rc=0, 22:23 UTC). The pinned packs unchanged; mx8+shl256x27's accepted attempt and id move, so the pack is re-exported.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-10-07 22:24:02 +00:00
.claude/agents Igneum: design docs, Metal lottery-hash prototype, CUDA test pack, finality simulation 2026-10-03 15:06:01 +00:00
.github/workflows The overlap sweep joins the gate: one check line in tools/ci/pre-push.sh (self-test, then the built site; IGNEUM_OVERLAP_APPS=1 adds the miner and wallet UIs), CI installs Playwright outside the tree before the gate, the README row; the measurer caches visibility and clip chains per element so a long page reads in seconds (7 October 2026, 16:1x UK) 2026-10-07 13:57:30 +00:00
app Reproducible builds: SOURCE_DATE_EPOCH from the commit's author time, TZ=UTC and one fixed target path in every build path; self-test 2026-10-06 20:09:23 +00:00
brand site-ui-5, deploy 1: the site in the apps' look (7 October 2026, 15:0x UK): one design layer on the app's tokens, the slim bar, the home fold on the shared scene, /download, OS and GPU marks 2026-10-07 13:46:23 +00:00
docs Counter ASIC 4.0 research: 20.2 the verifier rows (AVX2 tile path 0.047 us per tile per unit, mm1430 10.14 ms loaded; the per-load placement derives 11 percent fewer distinct items per warp: a uniformity fault as drawn, with its fix) 2026-10-07 22:08:58 +00:00
igneum-census Counter ASIC 3.0 item 8: the latency-shadow knob (LoadClass +sh<S>x<R>), its packs and the PC 2 playbook 2026-10-06 07:58:35 +00:00
igneum-pow Counter ASIC 4.0 research: the per-load shadow's duplicate reads, found and fixed (the acceptance test judges the order the class executes) 2026-10-07 22:24:02 +00:00
infra Merge remote-tracking branch 'origin/master' into peer-directory 2026-10-07 14:18:50 +00:00
packaging Reproducible builds: SOURCE_DATE_EPOCH from the commit's author time, TZ=UTC and one fixed target path in every build path; self-test 2026-10-06 20:09:23 +00:00
pool Pool v0 rebased onto master and the 0.3.14 fork: the program class and era seed ride with every seeds and job line; the re-check test for class v3 and class v4 2026-10-06 18:49:13 +00:00
proto-cuda Counter ASIC 4.0 research: two experimental program classes behind the pack (no chain class, no consensus change) 2026-10-07 21:53:54 +00:00
proto-metal Counter ASIC 3.0 gates (node): the program id carries the class. A class v4 program is generator 4 wherever it is made: the CLI's --era path stamps the generator from the class (era_generator_of: 4 on V4_CLASS, 3 otherwise; ProgramClass::of_load_class), show honours --program-class and --era-hex; the shadow block marks class v4 in packcheck, packfile.h and the Metal worker (a generator 3 pack with IGNEUM_SHADOW_INSTRS is refused as a v4 program stamped v3, a generator 4 pack without it is refused; generator 2 ladder packs unchanged); the seven gate packs re-exported (generator 4, id c120d7963abdcd96 against the v3 control's 73bcbfe8ccf988f1, every other line byte-identical); class-v4.mjs asserts every v4 epoch's id against the CLI's same-seed v3 and v4 ids (--id-check-against v4 is the assertion's failed case); v2 and v3 ids byte-identical (60 + 7 + 4 + 19 + 7, the pinned packs) 2026-10-06 16:24:02 +00:00
proto-newpow Horizon: lane 8 (new-proof-of-work) lands: three schemes, two prototypes measured on rented 4090s 2026-10-06 20:18:39 +00:00
proto-opencl Merge origin/master into ca3-coord: Counter ASIC 3.0 complete (every gate green, P2 green, P1 written); the drive-ref check skips single-quoted here-strings and the copied-sources check reads code lines only (master's CI red on 471832b); main's decisions and the close in the status file 2026-10-06 18:05:54 +00:00
proto-vdf proto-vdf: Wesolowski VDF between the certified checkpoint and the program seed 2026-10-03 16:39:01 +00:00
proving exec 0.3.14 app side on release-0.3.14: the prover exports from one block below and seeds from the account dump; nothing claimed below the restart; the harnesses 2026-10-06 16:59:18 +00:00
relay Console: the job-channel ping (MF-11, 7 October 2026). Every 0.3.21 app names itself on its wake long-poll (machine, version, last job); /wake records one row per machine in relay_wake_seen before it holds; the Machines tab reads "job channel polled N ago, last job X" and, after 15 minutes without a poll, "job channel silent since <time>, last job X" in red, whatever the uploads say (PC 2 lost power at 10:46Z and nothing said so until a person read the intake). Tests in relay/test/wake.test.mjs. 2026-10-07 14:10:59 +00:00
scene Chain scene 2.0.4: the box's height follows the lanes (onSize, autoHeight, the lane geometry in stats) (7 October 2026, 16:3x UK) 2026-10-07 13:58:33 +00:00
sim Mission item 12: the peer-directory fast-time harness (ten listing nodes announcing their loopback p2p address, every node advertising an unroutable external ip so gossip is dead and the directory is the one live source, a fresh node with one genesis peer watched for 8 outbound), the eclipse simulation (the draw as implemented, without replacement, 34 percent attacker over m keys), and peer_directory_activation_daa in the 60x file 2026-10-07 13:14:00 +00:00
site Merge site-ui-5 ce8ad8ca into master (gate: green on ce8ad8ca, recorded by tools/ci/pre-push.sh; the full gate runs in CI on this merge) 2026-10-07 14:09:28 +00:00
tools Counter ASIC 4.0 research: the identity that forbids a zero-premium chip lever, the ranked designs, and the SM-sparse worker variant 2026-10-07 20:55:27 +00:00
.gitignore gitignore: infra/cross/out-workers-box/ (the workers tools/workers-remote.sh fetches from the box) 2026-10-06 19:11:39 +00:00
.vercelignore Commit identity rule and a deploy ignore file 2026-10-03 17:33:46 +00:00
CLAUDE.md Igneum: design docs, Metal lottery-hash prototype, CUDA test pack, finality simulation 2026-10-03 15:06:01 +00:00
rust-toolchain.toml rust-toolchain.toml: one pin for every side (1.99.0, the two cross targets); the mismatch refusal reads it 2026-10-06 23:44:20 +00:00