Independently audit encoded-bank crossing and dimension proofs

[DLamdata]

This independently audits PR #2872 at cc5c2398505447a321fb9f9826163825e3b269fd. No substantive gap was found in its crossing criterion, weighted producer-prefix dimension theorem or nonunitary dimension-four odd-cycle obstruction under the stated separated-label, identical-paired-word, arbitrary-state contract. The addition is independent proof review and different executable evidence, not a new representation theorem, supplier or exponent.

Only research/encoded-bank-criterion-audit/ is added. The package credits Tekalign/@tekuuu (#2872), Gareth Davidson/@bitplane (#2850), Dennis/@antunishdPursuit (#2859), and the concurrent #2871 work cited by the source. Coordination: #2842 (comment).

Validation on CPython 3.12.14/Linux:

• All 974,528 canonical paired-word/commutation-graph cases through five columns pass, with 128 literal-rewrite cross-checks and meaningful omitted-edge/domain controls.
• A separate rewrite oracle passes all 162,014 labeled cases through four columns. Exact rational RREF checks the unrestricted nonunitary commutant fixture and hypothesis controls.
• All three final registered commands, optimized relocated/read-only replay, both receipt-corruption controls and commit-byte preflight pass. All three pinned predecessor command bodies pass; its 13-file tree was Git-authenticated and unchanged.
• Independent verifier and skeptical-review agents found no substantive defect; two wording ambiguities were corrected. Substantial OpenAI Codex assistance is disclosed; no independent human or formal-kernel review is claimed.

python3.12 scripts/research_package.py --package research/encoded-bank-criterion-audit

Base 67b122d33d6c599a49f65d50afa2f3fe8d790d02; publication 3091cfbd61005b3e5daa7dbfcf3377597924552d. Canonical checkpoint 74a07147f3951842897dff8b0d70a3415692e6e2 changes launcher cache isolation: this unchanged package also passed all three commands through that exact launcher. Read-only merge-tree compatibility passes; the scientific baseline remains pinned. Final overlap paginated 495 open PRs, 2,077 closed/merged PRs and 1,517 Discussion records, plus 2,543 local records; this is bounded coverage, not a novelty guarantee.

Native caller costs, physical recursion, hosting/precision, the full selected cold replay and full multiplication theorem remain outside scope. The selected record is unchanged. Further work needs a paid caller that changes the audited contract; this does not justify another unchanged-encoding search. New work is Apache-2.0.

Handoff checkpoint: canonical c7df0bb890cae0cc393cdd32dad19c3facc80992 is 24 commits beyond the scientific base; the contribution is one commit ahead. Guidance, selected record and launcher are unchanged from the tested 74a0714 checkpoint. Read-only merge-tree compatibility passes; the 13-file contribution remains unchanged.