Skip to content

test: add the BZMathlib conformance module and proof-track probe suite - #9504

Open
kim-em wants to merge 2 commits into
mainfrom
wave/bz-mathlib-phase34
Open

test: add the BZMathlib conformance module and proof-track probe suite#9504
kim-em wants to merge 2 commits into
mainfrom
wave/bz-mathlib-phase34

Conversation

@kim-em

@kim-em kim-em commented Aug 24, 2026

Copy link
Copy Markdown
Owner

This PR add the Phase-3 and Phase-4 evidence code for HexBerlekampZassenhausMathlib, which owns an executable tactic/checker surface and therefore falls outside the correspondence-only conformance exemption. The new conformance/HexBerlekampZassenhausMathlib/Conformance.lean (Oracle: none — the surface is proof-emitting and kernel-certified, with the untrusted compiled search python-flint-checked in the computational sibling; Mode: always) covers irreducibility, factor_poly, and both budgeted ! variants across typical, edge, and adversarial cases including the A4 multi-prime certificate, the cyclotomic X^6 - 1 factor pinning, multiplicity attribution, the over-budget dense-size error path, and genuine kernel-fallback replays on Swinnerton-Dyer inputs, with five axiom pins at the clean triple. The Phase-4 proof track lands as ten matched-import build-only probes (Baseline, three factor ladders plus a repeated-factor partner, three irreducibility rungs, two kernel-replay rungs) split into a CI probe lib and a scientific arm, with scripts/bench/bz_mathlib_sweep.py and its unit test mirroring the freshest sweep-runner model, the structural proof_probes: registration the Mathlib-free bench lint requires, and new ## Conformance and ## Phase-4 proof evidence SPEC sections. The scaffolding token's flagged gaps are confirmed stale: #2200's associates ambiguity is resolved by sign-normalized multiset uniqueness in FactorSoundness.lean and #2201's Mignotte wiring is live in FactorBound.lean, both proven and axiom-pinned. Timed measurements and the done_through bumps follow in the serialized measurement PR.

🤖 Prepared with Claude Code

Kim Morrison and others added 2 commits August 24, 2026 02:56
…actic surface

HexBerlekampZassenhausMathlib owns an executable runtime (the
factor_poly/irreducibility elaborators and their kernel-decide bang
fallbacks), so per SPEC/testing.md it takes the ordinary Phase-3
criteria rather than the correspondence-only exemption. Add
conformance/HexBerlekampZassenhausMathlib/Conformance.lean exercising
all four tactic entry points on committed Polynomial ℤ and Hex.ZPoly
fixtures across the certificate languages (single-prime witness,
Eisenstein handover, multi-prime degree obstruction, kernel fallback),
pinning hand-derived factor lists and counts, the decline diagnostics
on reducible/zero/unit/over-budget inputs, and the emitted proofs'
axiom cones; wire it into the HexConformance globs and record the
runtime-contract conformance section in the library SPEC.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This adds the Phase-4 proof-track evidence code for the
HexBerlekampZassenhausMathlib tactic surface: build-only fresh modules
under bench/HexBerlekampZassenhausMathlib/ProofProbe/ exercising
factor_poly on distinct-quadratic products over ℤ at degrees 4/8/12
(with a degree-8 multiplicity-attribution pair), irreducibility on the
Eisenstein binomials X^n - 2 at degrees 4/8/16, and the kernel-decide
fallback irreducibility! on the Swinnerton-Dyer minimal polynomials at
degrees 4/8, each with the matched import-all baseline; the
HexBerlekampZassenhausMathlibProofProbe (CI) and
HexBerlekampZassenhausMathlibProofProbeScientific (release) lean_libs;
the external sweep runner scripts/bench/bz_mathlib_sweep.py with
manifest tests; the libraries.yml proof_probes declaration
(done_through unchanged; measurements are serialized externally); and
the SPEC 'Phase-4 proof evidence' section. Progress notes in
progress/2026-08-24T02-57-08Z.md.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@kim-em
kim-em enabled auto-merge (squash) August 24, 2026 03:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant