soft3/audit/chain/chain conformance — kernel reference run 2026-10-16.md

cyber soft3 audit chain conformance audit cyber
π 0.0% 802 words · 5 min

Receipt of the first run of chain-conformance.md on the consensus kernel (crate/src/chain/, branch feat/chain-kernel, revision 1243ef2 plus the engine split committed with this file). Per the audit doctrine: revision, command, observed result; the requirements stay in the specs.

commands

cd soft3/crate
cargo test                                   # 168 lib + 11 bin tests, 0 failed (212 s debug)
cargo test --lib chain::                     # 66 chain tests (19 engine scenarios)
cargo build --release                        # 0 soft3 warnings (3 pre-existing in bbg's vendored fjall)
soft3 conformance check chain                # chain.snap: 9 entries, no drift

Stand: ~/cyber/.stands/N1 — soft3 feat/chain-kernel over main 8525b17 (wallet #184, genesis #183), siblings at the make.sh integration base (bbg 1d39975 0.3.0, cybergraph feat/durable-applications, lens release/0.2, nox release/0.4, zheng feat/public-certificate-v3, mudra refactor/spell, hemera/strata/tade/tru/tok/radio/foculus defaults). Genesis file: ~/cyber/.cache/genesis/pussy.tade (3,249,582 bytes) from soft3 genesis build.

results by section

section result where
§1 hemera anchors green — empty, "boot", "pussy", 168 zero bytes chain::sigs::tests
§2 compact target green — 7 canonical, 6 rejections chain::header::tests
§3 work green — 2^228/2^232/2^236/1, 3-block sum 50,331,645 chain::u256::tests, chain::pow::tests
§4 ASERT green — all 6 rows with e, shifts, frac, factor, bits (row 5 pins floor toward −∞ and the cubic; row 6 the clamp); T-spacing holds bits; T/2 for 600 s halves the target; 100× slowdown back at T within 70 min chain::daa::tests
§5 header green — bytes, id bdfbbff9…, 167/169 bytes, version 2, wrong chain; mined header d1eec40d… valid at 2^232, rejected at 2^228, nonce −1 fails chain::header::tests, chain::pow::tests
§6 transactions green — coinbase bytes/txid/leaf/node, transfer statement c7e4a223…, signed transfer verifies and fails on every byte flip / 101 / 103 bytes / other chain; cyberlink 257 B, valence 2 and p == q rejected; claim account 28ff5c6d…, legacy owners cca6f468… and d5e54c49…; encoding rejections every kind chain::sigs::tests, chain::tx::tests, chain::tree::tests, chain::tests::rules
§7 identities green — neuron_of, tokens, topic 1f003a48… chain::sigs::tests
§8 state green, blessed — empty root, conformance genesis (file, root, block 0), after block 1 and after a cyberlink in the engine, maturity 100 rejected / 101 accepted, conservation at every height of a 200-block random chain through a 3-block reorg conformance/chain.snap, chain::tests::{rules,simulate}
§9 consensus green — heaviest chain, first-seen ties, reorg from tip−120 rejected / tip−119 taken, checkpoint vetoes a heavier fork, foreign signer and past-sunset checkpoints dropped, PoW error before the body, wrong state root marks the header invalid, time == MTP rejected / MTP + 1 accepted, now + 121 s held then accepted; three stores converge chain::tests::forks
§10 genesis green, blessed — pussy particle 2a83e54f…, state root f2296834…, block-0 id b863ce28… reproduced by the kernel's own path (Genesis::state → genesis::state::load); bostrom stays red (lens commitment limit, audit/genesis-2026-10-15.md) chain::genesis::tests, conformance/chain.snap
§11 wire not this kernel — PR #187 (chain/net) —
orphan pool not this kernel — the net layer's pool —

Blessed entries written by soft3 conformance bless chain (9): the three genesis lines unchanged from the hand-written file, plus chain::tx@v1 signed transfer / cyberlink fingerprints and chain::state@v1 empty root, conformance genesis file / root / block 0.

caveats, stated plainly

  • Nonce placement: bbg at the integration base has no NeuronRecord.nonce; the kernel keeps the nonce in a side map beside the state (feature bbg-nonce off) — every rule holds, the nonce is not yet under state_root. bbg branch feat/chain-dims (97c6272, 56 tests green against bbg master's siblings) carries the field as the fourth neurons leaf; when it lands with DIMENSION_VERSION 3, flip the feature default and re-bless chain.snap in the same PR (every root moves).
  • Conservation is checked from per-token running sums the state backend maintains through its own trait (Σ balances == supply for every token, native supply == S0 + h·SUBSIDY), not by tok: tok's MintLedger owns settle-mint conservation over its own balance map and cannot sum bbg balances; chain.md §3.4's "tok owns the arithmetic" is not met literally.
  • Chainwork of block 0 is 0 (genesis is not mined); chainwork(h) sums the work of blocks 1..h.
  • The 200-block simulation and the 130-block reorg scenarios run at a test target of 2^255 through a cfg(test)-only override of POW_LIMIT; no release path has that hook.
  • cargo test of the crate takes ~210 s in debug: the pussy genesis root (35 s) and the fork scenarios dominate.
  • The pool (chain.md §11, in flight on specs/pool-formats) will move the genesis p params and every blessed root; rule set 1 rejects every kind ≥ 4 already and the activation table names no rule set 2 yet.

Graph