Verify

../

nonagon_review: Retains the accepted exact finite review, distinct grading, original subjects and reported runs, with documentary and interface changes identified.

---

The [[distance_problems/sallerk_2026_convex_nonagon_relations/evidence/verify/nonagon_review|whole-claim review]] retains the refutation-failed verdict for one fixed strictly convex nine-point set with maximum distance multiplicity three at every vertex and the three Er87b relations. Its distinct grading, source-reading limits, subject revision and reported executions are retained with the report.

verify_quartic.py uses the retained reviewer's quartic arithmetic and all 21 named obligations through the current shared harness. It is the filing author's interface adaptation. Local normal and optimized Python 3.12.13 replays each passed all 21 named checks. The exact original program is reviewed_quartic.py; both modes also passed its 21 obligations. These integration replays are separate from the retained independent mathematical review.

The adapter's docstring identifies it as the current adaptation and names nonagon_review.md for the mathematical review. The observed reproduction record is this page and the evidence account below.

The snapshot's unchanged docstring refers to an adjacent REVIEW_REPORT.md; that report's filed page is nonagon_review.md.

The retained source transcription is a provenance-redacted derivative, not a byte-identical copy of the original reviewed context, which is not retained. Only original lines 17–18 changed: a private commit/cache locator was replaced by the public post anchor and a saved-thread hash label, so its provenance does not require that private location; the hash value on line 19 was replaced by a removal marker on 2026-10-02. Every other byte, including the complete mathematical quotation, is unchanged.

The review's context list names the exact reviewed transcription, which is not retained; the retained transcription is the provenance-redacted derivative at ../assets/reviewed_source.md described above. Documentary source-locator wording in the assessment is normalized. Location references identify the exact reviewed context, not private dependencies of the current pages. Except for that locator wording, the retained assessment is unchanged. All other frozen attachments and the exact mathematical witness remain unchanged; the author checker's later interface change is recorded in the evidence account. This limited documentary normalization does not alter the immutable reviewed subject or extend its verdict to changed mathematical content.

Both original-checker replays reported eight bisections and cap 400. The adapter replays emitted named harness checks and ALL CHECKS PASS (21 checks), without bisection diagnostics; the outputs are not line-for-line comparable. The 400-bisection cap and its exhaustion path remain unchanged.

A later author execution of nine specified interface cases in both modes produced all 18 required refusals: exit 1, all named diagnostics, no success banner and empty stderr. Only the perturbed-C1 pair changed mathematical obligations; the other cases concern input binding and failure propagation. The zero-cap case returned the unresolved-sign refusal before any bisection, not a measurement of the real 400-step cap. These observations are distinct from the retained historical controls and independent mathematical verdict.

The evidence account gives the ordinary-clone commands, exact input, domain and failure behavior. No E0097 resolution, minimum-size result, uniqueness, alternate-branch classification, degree assertion or numerical tier follows from this review.