Audit: bind retained PR589 supplier identity and COPY deficit

[Arham-Tahir64]

Change and mathematical scope

PR #1277 leaves a concrete next question: can the retained PR589 supplier identity be bound when an assembly preserves its accepted numbers but names a false commit/blob? This adds an opt-in source-binding guard, confined to research/retained-bit-supplier-binding/. The finding and broader ancestry-equality repair remain Sagar Thakrar’s; this is the complementary executable guard.

The guard checks the exact PR589 record and compressed word across the contributor’s original research/e8-bit-p111/ path, the integration commit, and main’s renamed research/e8-bit-p111-tuned/ path. It independently follows D/A frame assignments and ranks the nine COPY frames over Q: nine rank-seven totals, decoder constant 63, and 64*28416 - 1808352 = 32*(8*111 - 9*63) = 10272. All four committed PR741/799/1042/1094 assemblies and the archived bit result pass.

--assembly and --bit-result guard newly emitted outputs against that exact source; an optional receipt hashes the exact byte snapshots checked. False provenance, a same-blob different real commit, compensating COPY ranks and a mass-preserving histogram change are rejected. The implementation imports no supplier/frame/moment checker and runs under -I -O -B.

No new κ, circuit, paid-profile change or production adoption. Existing verifiers, frozen inputs and the selected record are unchanged. The frame trace measures COPY ranks only; it does not establish scalar/frame legality or H1–H5. Recorded profile/budget fields remain inputs. No full cold/native/scalar/phase replay, certified moment, all-size/compiler, hosting, precision/tape, exact-recovery or analytic proof is claimed. PR1277’s original acceptance fixture is credited, not freshly replayed here. Both this guard and its broader ancestry-equality repair would be needed in a successor wrapper.

Task and overlap

• Base current main: bce54fb3f895cbe5cd4c0eb477c3afeed100190f; contribution: 1356813dddd0e03ea138743a799c2aece15f7d44.
• Card supplier-boundary-audit, refined to source identity/COPY deficit; nonce 78116c731f9a59be32e8cd1544265921, adversarial-audit track, priority revision 2. TASK.json records scope, domain and stop rule.
• PR589: a5afc8c6169ea61ba7c51dfd3cade02cc7b13b46; integration: b3eb785ac5b4b1ca2beea0f80db078a505626a90; EXPECTED blob: e293ddc557742637f9f679f083a1a282d87b1b13. SOURCES.json pins 11 live inputs and two historical provenance records.
• Nearest overlap: #1277, unchanged head 02d279515ff1fc468ab99a19d89461f1b6391507, owns the finding and next question; #1307 treats bit projectors; #1312 treats complex-owner X2. No pending PR is a code dependency.
• Refreshed public API snapshot at 2026-10-13 22:42:31 UTC: 815 open PRs including drafts, complete PR pagination; repository helper searched titles/bodies and local top-level notes/ledger. Start-here, active listing and live handoff Discussion were read. Complete Discussion pagination was unavailable; browser was signed out and the GitHub connector has no Discussion-write tool, so the claim is local. Arbitrary diffs/closed PRs are not exhaustively searched; no novelty proof is claimed.

Validation

Windows 11, CPython 3.12.14, standard library plus Git:

python3 -B scripts/research_package.py --package research/retained-bit-supplier-binding --check-only
python3 -B scripts/research_package.py --package research/retained-bit-supplier-binding
python3 -B scripts/research_baseline.py
git diff --cached --check

All pass. Registered replay includes the deterministic receipt plus 10 test groups / 23 rejection fixtures, actual CLI rejection and relocated-input/source-corruption checks. The card’s historical owner-boundary positive control also passes. The final package replay was rerun after the receipt-byte-snapshot change.

The broader test_research_*.py suite ran 20 tests and reported 3 Windows errors in unchanged tests: missing os.killpg on the deliberate timeout path, plus temporary-directory cleanup errors. Process-local core.autocrlf=false and workspace scratch directories were used. This is recorded as a failed broader suite, not a pass; package checks succeed separately.

Credit

Arham Tahir, with substantial OpenAI Codex assistance for research, implementation, testing and writing; no subagents or independent human review. Credit Sagar Thakrar / #1277, DreamingOfClouds / #589, eumemic / #555, the #545 rank-mass identity, Douglas Colkitt’s integration/bindings, Rohan Arun, Jeremy Kothe, Hao Yan, nightangelflowerwin-ops / #865, Lucas Steinmetz / #838 and the retained supplier lineage. Existing notices remain intact; new work is Apache-2.0.

Open next questions

README.md and TASK.json give the next task: combine this guard with #1277’s ancestry equality in a successor wrapper, exercise the false-source/timing variants, then cold-replay. Maintainer acceptance and versioned source/coverage pins block production adoption.

[Mister-Meeseeks]

100-PR batch review at 1356813dddd0e03ea138743a799c2aece15f7d44.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Retained bit-supplier source binding, independent nine rank-seven COPY recount, and adversarial provenance guard. Pins and controls strengthen coverage of the existing 10,272 deficit; no selected supplier modification or new κ.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.