mirror of
https://github.com/saymrwulf/fips205-slhdsa-verified.git
synced 2026-09-03 19:53:49 +00:00
mrwulf
851e976450
I wrote the README and TRUSTED-BASE to describe six bindings, then added Phase
3c an hour later and did not update them. That is exactly the drift GPT-5.6's
round-8 review is about, committed by me immediately after fixing it. Recorded
plainly rather than quietly corrected.
README six -> seven, with Phase 3c stated in full: both allowlists,
both directions (UNCLASSIFIED / STALE), why the instrument
surface carries cones, and the accounting identity as set
containment with its actual numbers — kernel 300 = corpus 265
+ instrument 35, residual none
check.sh header lists Phase 3c
The failure mode is worth naming because it is cheap and recurring: a phase is
added, the button is re-run, the button is green, and nothing anywhere fails
because a document is stale. Nothing in this repository can catch it — the
gates check the corpus, not the prose. Only reading the diff catches it, which
is why the round-8 briefs asked for exactly that.
Button re-run after the edits: green, accounting over 300 kernel constants.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|---|---|---|
| .. | ||
| gen/SlhVerify | ||
| Proofs | ||
| AUDIT-MANIFEST.txt | ||
| check-selftest.sh | ||
| check.sh | ||
| drill.sh | ||
| driver-allowlist.txt | ||
| extract.sh | ||
| inventory-allowlist.txt | ||
| inventory_gate.sh | ||
| lean-guard | ||
| model-correspondence.py | ||
| MODEL-CORRESPONDENCE.txt | ||
| PROVENANCE.json | ||
| RECORDED-RUN.md | ||