[1337isnot1337]

Change and mathematical scope

This is a negative, exact-head audit of #4633 at b2998bacd50d57c301fe67c5392054440c42398c and its nine-file update b390d64abfcdba1a3816b86fc763cbc95d43c12c. The submitted numerical replays pass, but the claimed native supplier and outer saving do not follow. GitHub closed #4633 without merging during this audit; this work did not close it.

The current three-dimensional BIT quotient erases 41 rational entrance lines spanning rank eight, fails the retained rank-seven transport pair on all 111 ports, and still passes its verifier after all 2,184 positive MOVE charges are zeroed (reported κ = 0.25) or every frame basis is emptied. The complete updated scan also contradicts the claimed three-dimensional plateau and four-dimensional decline.

The report, source snapshots, independent counterexample verifiers and complete replay evidence are in research/pr4633-quotient-contract-audit. The proposed central-ledger entry records the negative review and observed unmerged closure separately from acceptance. No selected result or production verifier changes.

Task and overlap

User-assigned full audited replay, based on fbc9ea0249746acca4f504d4fda2b84e0e36cf0d. All 21 changed files and the subsequent nine-file delta are covered. This explicitly changes the bit-helper-contract card's scope to an audit.

Overlaps #4604 and Rohan Arun's concurrent port-contract comment. Added evidence includes rational-field counterexamples, verifier mutations, both-head full scans and exact native complex transport checks. The unchanged complex package from #4629 is included in the reviewed diff.

Validation

Python 3.12.15; serial replay with explicit timeouts and a sampled 1 GiB process-group RSS limit:

• Both original registered packages and the updated BIT package pass.
• All four artifact rebuilds match committed bytes.
• Full original scans: 255 BIT kernels and 511 complex kernels.
• Full updated submitted domain: 1,940 candidates; 438 of 455 triples accepted with 145 distinct numerical values, not the claimed 445 identical results.
• Original BIT rational-projector control passes.
• All 2,101 distinct complex climb/helper edges lower in both orientations with exact rank and absolute-phase recipe decoding.
• Source-bound independent counterexamples, meaningful mutation controls, committed-byte/source-history preflight, and focused central-ledger consistency checks pass.

Reproduce the short audit with:

python3.12 -B scripts/research_package.py --package research/pr4633-quotient-contract-audit

Full author replay commands, logs, resource receipts, per-candidate results and limits are retained in REPLAY.json and evidence/. The first source-control import required the documented retained dependency path; both the failure and successful retry are preserved. No unchanged selected-frontier cold replay, Lean proof, all-history suite or manual hosted workflow was run.

Credit

DaysSky's source bytes and notices are retained. Credit to DreamingOfClouds/eumemic and the PR589 lineage, PR4470 and its predecessors, Ryanson Jonathan (#4604), and Rohan Arun's concurrent review. Prepared for Cinna Davis with substantial OpenAI Codex assistance; Apache-2.0.

Open next questions

A revised construction needs native rational frame realizations, a proved per-port controller and rederived output invoice, geometry-authenticated MOVE pricing, and complete strict ordinary-BIT/outer admission. The audit establishes no corrected κ.