This is a preservation record, not a new mathematical result or a change of research assignment. The current prime-pair continuation remains FRONTIER_2026_09_12.md, with its independent review. Older attempts and reviews below retain their original scope and dates.
Recovered scientific records
| Source | Preserved material | Status of this import |
|---|---|---|
PR #196, 46fdb433752a | frontier/2026-09-06/factorial_direct_bn/, factorial_full_cost/, two tests | Exact original experiment files, including negative results |
PR #199, 96ce564c89a3 | frontier/2026-09-06/certificate_route_test/BASELINE_REVIEW.md, review data and test | Review of the named combined-weight baseline only |
PR #200, 85a42e7c0c7d | frontier/2026-09-06/joint_correction_candidate/, original ZIP, review and test | Review of this candidate only, not the later adaptive chain |
PR #209, 895127103890 | MINOR_LEVEL_SETS.md and checker/output | Historical proof attempt and finite checks |
PR #210, 9d307cc092ef | ARITHMETIC_FOURTH.md and checker/output | Historical proof attempt and finite checks |
PR #211, 338430d457f3 | CENTERED_DISPERSION.md and checker/output | Historical proof attempt and finite checks |
| Original recorded file patches, September 8 | FOUR_FORM_SIEVE_ATTEMPT.md, checker and historical output | Recovered uncommitted attempt, not a completed upper bound |
Original research patches 727b3325, 2eb13ce9 | PAGE_ZERO_ASYMPTOTIC.md, probe and output | Both patches restored, including the explicit author-only review correction |
PR #226, 6d139c37528b | Davenport-Heilbronn hunt, mission, scripts and outputs | Interrupted hunt preserved, no final verdict awarded |
The source bytes are pinned in artifacts/recovery_2026_09_13/manifest.json, checked by tests/test_research_recovery_archive.py. Passing that check establishes preservation, not correctness of an asymptotic argument. Original result files must not be overwritten by reruns.
Recovery details and boundaries
The four-form note and checker were recovered from the original recorded add-file patches, followed by the recorded note correction. The checker hash is a0cd7978ce97e29df94edf85bbad4287bcee9c7771b9829aadeb8b15002789ea. The final note hash is b525cd74a55251432c087d584168fef9db55d135134731ed687a0376c93757a9. artifacts/four_form_sieve/historical-result.json is the captured original output. Its note hash refers to the version before that final prose correction; it is not a fresh execution against the corrected note.
PAGE_ZERO's second patch says the argument has not been independently read. Restoring the files does not remove that qualification. Likewise, the later adaptive packages are not covered by the two earlier reviews restored here.
A bounded numerical replay also found two over-rounded descriptions in PAGE_ZERO section 6. The saved data reach approximately 0.000327542 where the prose says "at most 3e-4", and 0.000006008405 where it says "6e-6". The source bytes are retained as historical evidence; those literal upper descriptions should not be reused. The replay checks numbers, not the proof.
The short-interval branch claude/email-check-4vxjvb at 9425a7b remains a separate source record. Its scientific drafts and its unfinished literature monitor were not indiscriminately merged. The older support arm in PR #148 also remains separate pending review of its shared guard changes. Their Git histories were retained, not deleted. Neither is a new assignment from this reconciliation.
The original manuscript artifacts, raw conversations, loose worktree files, all-ref Git bundle and recovery manifests are preserved privately. They are not public mathematical endorsements. Two generated Work ZIPs still require direct export: the quotient-geometry experiment and final continuation packet. The visible download failed with ERR_BLOCKED_BY_CLIENT; this record does not pretend that those bytes were recovered. Existing cumulative archives and their hash checks remain in place.
Where to resume
Read the September 12 frontier for prime-pair/RH work. Use the factorial subdirectories for certificate provenance, not as substitute RH bounds. Treat the zero-proportion/short-interval line and the scalar-window manuscript as distinct research tracks. This reconciliation does not choose the next attack, rerun large experiments, or promote an unreviewed claim.