Issues

• Wire writes 8 bytes per tensor value but the decoder refuses anything outside i32 -- top 4 bytes are provably sign extension, 2.00x payload

  audit
  mgillr/acfa-rs#127 · Filter by author mgillr opened on Aug 21, 2026

• entry.rs::leaf() folds a MALLEABLE Ed25519 signature -- any future state-root-derived challenge is grindable at 1/(1-phi) trials

  audit
  mgillr/acfa-rs#126 · Filter by author mgillr opened on Aug 21, 2026

• Lemma 12 NEVER certifies at scale -- certified=false at every d from 1e3 to 1.024e6, bits_short RISES with d

  audit
  mgillr/acfa-rs#125 · Filter by author mgillr opened on Aug 21, 2026

• layer-1 MAX_COORDINATE_OPS has NO runtime override -- no shipped build can touch a large-model instance at any setting

  audit
  mgillr/acfa-rs#124 · Filter by author mgillr opened on Aug 21, 2026

• enc_tensor allocates per coordinate -- 6.46x available from an allocation-free encoder (memoisation half killed: sound memo costs what it saves)

  audit
  mgillr/acfa-rs#123 · Filter by author mgillr opened on Aug 21, 2026
  1

• A redacted receipt carrying a FORGED claimed_output_root verifies Ok -- nothing checks it on that path

  audit
  mgillr/acfa-rs#122 · Filter by author mgillr opened on Aug 21, 2026

• CRITICAL trap on the streaming path: l1_max is max-over-pairs of sum-over-coords, and the wrong chunking forges Lemma 12 certificates

  audit
  mgillr/acfa-rs#120 · Filter by author mgillr opened on Aug 21, 2026

• SCALE: the wall is memory, not work -- and exact integer arithmetic makes chunked accumulation bit-identical, so 1B params fits in 80 MB

  audit
  mgillr/acfa-rs#119 · Filter by author mgillr opened on Aug 21, 2026

• stress_max panics on a malformed numeric argument -- rust-04 at a site the argv guards do not cover

  audit
  mgillr/acfa-rs#118 · Filter by author mgillr opened on Aug 21, 2026

• APERTURE: 0 of 49 audit findings came from executing the system -- ten execution surfaces have never been mined

  audit
  mgillr/acfa-rs#117 · Filter by author mgillr opened on Aug 21, 2026

• War-room seat G cannot be woken: launchd delivers and archives but cannot re-invoke a session, so the cursor never advances

  audit
  mgillr/acfa-rs#116 · Filter by author mgillr opened on Aug 21, 2026

• cli_reject.rs uses fixed $TMPDIR paths -- two concurrent cargo test runs collide and produce a spurious red

  audit
  mgillr/acfa-rs#115 · Filter by author mgillr opened on Aug 21, 2026

• encode() refuses nothing -- the #105 guarantee holds on encode_checked only

  audit
  mgillr/acfa-rs#114 · Filter by author mgillr opened on Aug 21, 2026

• vectors_v1_compat.json provenance is not reproducible in-tree: its generator ships as a .rs.txt that does not compile here

  audit
  mgillr/acfa-rs#113 · Filter by author mgillr opened on Aug 21, 2026

• #98's WireError exhaustiveness is unmeasured: acfa-verify's own exhaustive match masks the probe

  audit
  mgillr/acfa-rs#112 · Filter by author mgillr opened on Aug 21, 2026

• The MAGIC / EMITTED_PREIMAGE const tie is one-directional: EMITTED_PREIMAGE=V1 with MAGIC=V2 compiles

  audit
  mgillr/acfa-rs#111 · Filter by author mgillr opened on Aug 21, 2026

• reference/tests/ is not wired into CI -- 8 tests that nothing runs

  audit
  mgillr/acfa-rs#110 · Filter by author mgillr opened on Aug 21, 2026

• The recorded #87 small-order universe is wrong: 14 members with a 10/4 split, not 13 and 10/3

  audit
  mgillr/acfa-rs#109 · Filter by author mgillr opened on Aug 21, 2026

• reference/ has no guard on the #87 door: is_usable_pubkey is deletable from verify() at 8 passed / 0 failed

  audit
  mgillr/acfa-rs#108 · Filter by author mgillr opened on Aug 21, 2026

• The small-order key fixture tests 4 encodings; the true set is 13, including 3 that must fail-closed on decode

  audit
  mgillr/acfa-rs#107 · Filter by author mgillr opened on Aug 21, 2026

• HIGH: admit() is context-blind -- state.rs filters round, conviction, PKI and signature, never ctx

  audit
  mgillr/acfa-rs#104 · Filter by author mgillr opened on Aug 21, 2026

• The float-absorption result has no CI gate at all: xarch_absorb is referenced 0 times in ci.yml

  audit
  mgillr/acfa-rs#103 · Filter by author mgillr opened on Aug 21, 2026

• LOW: demos/run.sh always reports exit 0 and its verdicts are hardcoded prose; not referenced by CI

  audit
  mgillr/acfa-rs#99 · Filter by author mgillr opened on Aug 20, 2026

• MEDIUM: claim-caveat-check.py self-test passes while 5 of 6 pairings go unevaluated -- fl-06 re-entering through the tool built to stop it

  audit
  mgillr/acfa-rs#97 · Filter by author mgillr opened on Aug 20, 2026

• HIGH: layer2-finality -- five guards in halt.rs and certificate.rs survive deletion of the behaviour they name

  audit
  mgillr/acfa-rs#95 · Filter by author mgillr opened on Aug 20, 2026