DONE: full-payload PR589 gzip rebuild comparisons (PR1559)

[bbMisk]

DONE — PR #1559, commit c4f8af1dcf9fc2317c893b6b250007ded5ef879f.

Established: eight controlled recompressions of the PR555 base and PR589 tuned words fail the original two shell cmp checks while preserving every decompressed byte. The opt-in adapter changes those two comparison sites and their success message; original sources, immutable hashes, and all later commands remain unchanged.

Verification: five focused groups pass on Python 3.12.13 and optimized Python 3.14.6, including eight representation controls, 24 changed/corrupt-payload CLI rejections, source drift and exact adapter-delta controls. The complete adapted bit replay passed all steps 0–8 in 131 seconds on macOS arm64 / Python 3.14.6 / Boost 1.87.0, including 883/1,327/1,549 formal columns and the original exact bit savings. The registered package replay and selected source baseline also pass.

Limits: local Python 3.12.13 and 3.14.6 regeneration matched the shipped compressed bytes; this is a controlled false-rejection demonstration, not a natural local version failure. The complete replay used normal generator compression. No full selected frontier1094 replay, new kappa, Lean, hosted CI, Windows or independent human review is claimed. Default integration remains a maintainer decision.

Credit DreamingOfClouds #589/#545, eumemic #555/#493 and retained lineage; Maxime Fleury #220 / Antoine Olivier #214 and MrPink604 #819 for the predecessor payload convention and distinct owner-checker adapter. Prepared for Hanson Sun with substantial OpenAI Codex assistance. Scope released; no merge or selected-frontier change.

---

Original claim:

CLAIM

Question: do the two compressed-byte cmp checks in selected PR589 tune/verify.sh step 1 reject regenerated words with identical full decompressed payloads, and can an opt-in adapter fix exactly those comparisons?

Base: 55a0258 (current main at session start). Task portable-replay, selector seed 325e378b7be425b7dd26bee91e4a8ff7, refined scope.

Method/domain: regenerate the PR555 base and PR589 tuned words under local Python 3.12 and 3.14; preserve original package manifests and all mathematical checks; compare full decompressed bytes with recompression, payload-change and corrupt-stream controls. Bound to those two cmp sites and one session.

Expected artifact: a source-pinned opt-in replay adapter and an independently runnable focused regression package, if the failure reproduces. No selected-frontier change or new kappa.

Overlap: live snapshot at 2026-10-14 02:40 UTC covered 714 open PRs and 207 Discussion records, complete pagination. Prior payload convention/adapter: Maxime Fleury #220, Antoine Olivier #214 and #819; bit lineage: DreamingOfClouds #589 and eumemic #555/#493. Pending #946/#1298 concern encodings and #1509 concerns the Boost dependency, not these two comparisons. No novelty guarantee.

Claim expires: 2026-10-14T04:43:25+00:00. Prepared for Hanson Sun (@bbMisk) with substantial OpenAI Codex assistance. Apache-2.0.