mirror of
https://github.com/saymrwulf/fips205-slhdsa-verified.git
synced 2026-09-04 20:03:44 +00:00
Round-8 review (GPT-5.6, register key `slh-provenance-contradictory`, HIGH),
raised in round 7 and unfixed since.
The machine-enforced subject is unambiguous and appears in PROVENANCE.json and
extract.sh:
fips205-source @ a3ce8e8644fe302019ed7ae271912333f1476de4
It appeared in ZERO markdown files. The prose instead named three other
revisions across four documents — README `797b4ef`, TRUSTED-BASE `3153988`,
ATTESTATION-BASIS `c945821` — so a reader could not tell which Rust the proofs
are about. Subject identity is part of the attestation object: a proof about a
model is not evidence for an unspecified source revision.
README.md current snapshot head -> a3ce8e8, with the lineage
bea1051 -> 797b4ef -> a3ce8e8 stated explicitly and
matching PROVENANCE.json's own upstream_deviation text
TRUSTED-BASE.md current snapshot head -> a3ce8e8
ATTESTATION-BASIS.md subject -> a3ce8e8, AND the "four model_integrity_sha256
hashes" claim corrected: only TWO (Types.lean,
Funs.lean) are regenerated by extract.sh; the two
*External.lean files are hand-maintained and separately
byte-pinned. Stating four invites a reproducer to expect
extraction to produce files it never touches.
RECORDED-RUN.md IS DELIBERATELY NOT REWRITTEN. It records a run that really
happened on 2026-07-24 against `797b4ef`. Editing the identity of a past run to
match today's pin would falsify the record rather than correct it. It now
carries a header saying so and naming the current pin.
Independent extraction remains UNVERIFIED — no third party has regenerated the
Lean from the Rust. That is unchanged by this commit and is stated where a
reader will meet it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|---|---|---|
| .. | ||
| gen/SlhVerify | ||
| Proofs | ||
| AUDIT-MANIFEST.txt | ||
| check-selftest.sh | ||
| check.sh | ||
| drill.sh | ||
| extract.sh | ||
| lean-guard | ||
| PROVENANCE.json | ||
| RECORDED-RUN.md | ||