DONE: PR1075 complete cube-XOR differential replay

[RossRichardson]

Status: DONE; exact-head scoped incorporation is complete. Updated 2026-10-16 08:36 UTC; priority revision 5. Claim released. Hosted CI remains queued without a conclusion.

PR1075 follow-up receipt, current head 4ba229532a76516e5f6f5befc1d6d9dc4834f6f9. Full author-checked differential verification passes at the reviewed head (118.575 s) and integrated head (105.254 s), each covering 4,324 complete rows / 8,596,927,616 binary coefficient positions, both complete receipts and all 17 tests. The new head also passes nine inherited cube tests, all 38 effective source pins and selected-record integrity. Separate AI-assisted audit found no substantive issue; no independent human review is claimed.

The original manifest, core, fixtures and benchmark evidence remain unchanged. A separately authenticated three-file transitive-import overlay makes the historical replay compatible with canonical main ffbd90dd.... No new kappa, t=6 exclusion, supplier or formal-theorem claim follows.

PR1075 was merged at 2026-10-16 07:53:33 UTC via 9f3adbaeb4df6c934edfaf324d0cc3f77f8b7214. Canonical main a1bfbc677fdcbc2a27a9b3bf6c78cac8fd6da9e2 records reviewed head 4ba229532a76516e5f6f5befc1d6d9dc4834f6f9 as incorporated, with all four scoped review-completion fields true. All 14 contribution blobs match that reviewed head. The retained maintainer receipt covers the complete 4,324-row / 8,596,927,616-coefficient comparison and 17 controls; it is separate local evidence, not a hosted-CI result.

The current-head Linux CI is still queued / no conclusion; all 62 checks at this head remain queued at this status check. No new review request or code defect was found, so no duplicate CI run or source revision was made. Reproduce in a fresh checkout without bytecode caches: python3.12 -I -O -B research/cube-xor-replay/verify.py (about two minutes locally, 15-minute cap).

Delta, exact pins, credit and separate loader-hardening question. Original cube work: Douglas R. Colkitt. Adapter/follow-up: Ross Richardson with substantial OpenAI Codex assistance; Apache-2.0.

Initial continuation status (superseded; preserved for provenance)

Status: FINDING / local continuation running, 2026-10-16 07:15 UTC; priority revision 5.

Question: finish the complete bounded cube-XOR differential/certificate comparison requested in PR #1075's review, after the review's 180-second cap. Scientific pin: d84d3b950127ff9252e5c059a629bb654500192a; current guidance main: ffbd90dd90dbe0055464f3a1411b25a3a4fdc6b8.

The existing focused Linux CI job is successful. Its log explicitly contains both differential cases, 17 passing tests, and PASS_CUBE_XOR_VERIFICATION rows=4324 coefficients=8596927616 t6_rank=3088 t6_excluded=false. It checked the PR merge ref 74be1199fb224e0ed24f3a9713e771f976066721 (head d84d3b950... into abdc6a700...) on Ubuntu 24.04.5 / CPython 3.12.15.

I am also running the unchanged python3.12 -I -O -B research/cube-xor-replay/verify.py at the exact PR head in a fresh checkout on macOS, with a 900-second cap and retained logs. A separate AI-assisted source/coverage audit checks the mask identity and complete row counts. This is a continuation of the outstanding check, not a new construction or benchmark; no new kappa, t=6 exclusion, supplier admission, formal proof, or independent human review is claimed.

Next step: finish the local replay, post the exact receipt to the PR and release this continuation. Advisory continuation expires 2026-10-16 09:15 UTC. Overlap refresh covered 489 open PRs and 1,531 Discussion records with complete pagination and no warnings at 07:13-07:14 UTC; no separate cube-XOR investigation thread was found. Original cube family, rank screen and certificate: Douglas R. Colkitt. Adapter and follow-up: Ross Richardson with substantial OpenAI Codex assistance; Apache-2.0.

[RossRichardson]

FINDING / scope refinement, 2026-10-16 07:18 UTC. The unchanged reviewed head now passes the complete local CPython 3.12.12 command in 118.575 seconds: both differential cases, all 17 tests, and the full 4,324-row / 8,596,927,616-coefficient marker. Local log SHA256: 2673a033caa449e61e7013040d09aa3a84db0ddfb74c83ac6a3239c87954de3d.

The prepublication compatibility check found three intentional source-pin rejections on current main: audit_fused_block.py, exclusion_circuit.py, and shared_point_circuit.py changed since the historical baseline. These are transitive imports; their changed functions are outside the cube replay's executed path. I am preparing a narrow integration refresh that preserves the original manifest, core/certificate/fixture pins and benchmark evidence, records the three current-main dependency bytes separately, and reruns the same complete command after integrating main. No scientific input, row cap, rank or exclusion claim changes. The normal-import loader still assumes a fresh process and absent bytecode caches; this continuation explicitly verifies that execution boundary. The original-head review request is resolved by the retained exact local and hosted receipts; current-main compatibility is the additional bounded follow-up.

[RossRichardson]

DONE / claim released. PR1075 comment contains the full current-head receipt at 4ba229532a76516e5f6f5befc1d6d9dc4834f6f9: both cases, all 17 tests, 4,324 rows / 8,596,927,616 coefficients, 105.254 s. Original-head continuation also passed in 118.575 s. The narrow compatibility refresh preserves the historical scientific inputs and evidence. Focused CI: queued / no conclusion; 62 checks observed at the new head. The top summary now records this completed finite result and the distinct hosted-CI/maintainer handoff. No new kappa or supplier admission is claimed.