[mihir-s-05]

PR #4992 was archived with independent scientific review and source-bound replay still open. This answers its review handoff with a written proof review and a separately implemented complete-word oracle. No correctness gap was found in the stated sharp-cap argument.

The new generator chooses the four positions of the repeated label and a perfect matching on the remaining eight positions. Its completeness argument covers all 51,975 coordinate-normalized multiplicity words before any geometric filtering. Literal checks yield 3,381 simple walks, 828 universal walks and 4,968 weight-four/five functional cases: the maximum excess is 11, and no universal (8,4) split occurs. These bounds and counts are credited prior results; the added evidence is the different oracle, semantic controls and explicit review of the repeated-projection implication and support classification.

Proof review and completeness argument · exact receipt · canonical investigation

Validation

Windows 11, CPython 3.12.0, standard library only:

python3.12 -B scripts/research_package.py --package research/f7-five-label-independent-audit
python3.12 -B scripts/check_research_commit.py --package research/f7-five-label-independent-audit

• Registered normal and optimized replays pass. Eight actual code mutations are rejected with stored-receipt comparison disabled; an unmodified positive control passes.
• The predecessor's own registered normal/optimized replay also passes (23,816 support sets, 127,400 set pairs and 13,664 universal graphs).
• All seven package files match the published commit. Official export integrity and fresh Git-free, read-only optimized verification/mutations pass; all eight exported files remain unchanged. Export SHA256: d7e66dc735d26f8612d7e1fee6ccd5a8c6bc47b90d9a6a95fdef9ad656e881d5.
• The reviewed README's Git blob matches both exact author head 12a0105150f5f44a19a2199db8506d1aa44b009a and canonical-main base f259dabc7d70e0db8d6d36bfda888f008bbfecc1. Package provenance depends only on that main ancestor. The optional full-history preflight is unrun in the shallow clone.

Scope, coordination and credit

Seven new files in one isolated package. No engine, frozen predecessor, central acceptance ledger, selected frontier or complete kappa calculation changes. No seven-label census, supplier admission, selected cold replay, formal proof-kernel acceptance or independent human review is claimed. Selected kappa remains 60403/20000000.

Main/guidance were unchanged at the 07:16 UTC submission refresh. Main advanced during publication to 4c0a359d88f22f05189d51f48387750b801a45d4; a fresh fetch found no changes to guidance, selection, audited sources, this package, or the shared replay/publication workflow files. The original scientific baseline is retained. Remote overlap covered 329 open PRs and 4,150 Discussion records with complete pagination; local coverage was 47 records with 6,665 tracked notes omitted. The refresh identified a concurrent six-plus-six path-join audit, which is acknowledged and uses a different oracle. PR5001 engine integration is separately claimed; this package does not take that work.

Credit Ali Romman (#4992), Mihir Srivastava (#4907), Nick Petalas (#4894), unpaidmoon (#4148), Nicholas Stone (#2973), James Reed (#3096) and Ross Richardson (#943). Prepared by Mihir Srivastava with substantial OpenAI Codex assistance. The account also authored predecessor #4907; this is disclosed as a fresh solo agent/code audit of the successor, not independent human acceptance. Apache-2.0. Open next questions follow the existing engine-integration claim.