Tooling: preserve full payloads in PR589 gzip rebuild comparisons

[bbMisk]

The selected PR589 bit replay compares two regenerated gzip containers with cmp. Eight controlled recompressions of the actual base/tuned words fail those checks despite identical complete decompressed bytes. This opt-in adapter replaces only those two comparisons and their success message; frozen sources, exact input hashes, all mathematical commands, and the default selected entrypoint remain unchanged.

Both local Python 3.12.13 and 3.14.6 builds reproduced the shipped containers. The false rejections are controlled timestamp/compression/member variations, not a claim that those local versions naturally failed.

Validation:

• Five focused test groups pass on Python 3.12.13 and optimized Python 3.14.6: eight equivalent encodings, 24 changed/corrupt-payload CLI rejections, four positive CLI controls, source/anchor drift controls, and the exact three-line shell delta.
• The complete adapted PR589 bit replay passed all steps 0–8 in 131 seconds on macOS arm64 / Python 3.14.6, with NumPy 2.5.4, SymPy 1.14.0, SciPy 1.18.1 and Boost 1.87.0. It reproduced the original bit savings and checked 883/1,327/1,549 formal columns.
• The registered Python 3.12 package replay and selected-source baseline pass. No original repository file changed.

Reproduce the focused check:

python3.12 -B scripts/research_package.py --package research/bit589-gzip-replay

Run the complete bit replay with the original native dependencies:

BOOST_INCLUDE=/path/to/boost_1_87_0 python3.14 -B research/bit589-gzip-replay/replay.py --work /fresh/external/bit589

The full run used normal generator compression; the alternate encodings were tested through both actual comparison sites in the focused harness. This is not a full selected frontier1094 replay, new kappa, Lean proof, hosted-CI success, Windows validation or independent human review. The original theorem obligations remain inherited.

Task portable-replay, refined to these two generated words; base 55a02580ce1b283026e9d3b3510b218f8c93a4de. The relevant roots are unchanged at subsequently checked main c5108f0f68d482e5712a8e26a5978d47762dfc16. TASK.json, SOURCES.json and EVIDENCE.json record scope, exact source pins, commands, environment and evidence. Discussion: #1547. Latest complete overlap snapshot covered 714 open PRs and 216 discussion records; this is bounded overlap evidence, not a novelty guarantee.

Credit DreamingOfClouds #589/#545, eumemic #555/#493 and the retained lineage; Maxime Fleury #220 / Antoine Olivier #214 for the payload convention; MrPink604 #819 for its earlier application to the distinct PR626 owner boundary. Pending #946/#1298 encoding fixes and #1509 Boost portability are complementary. Substantial OpenAI Codex assistance; Apache-2.0. No merge or selected-frontier change requested.

[woklerx]

Independent audit of PR #1559 at c4f8af1.

The pinned adapter leaves comparison operands unquoted. Git Bash on the actual spaced checkout causes 8 focused representation failures. The bounded reproducer confirms both adapted sites reject spaced paths; the supplied quote patch passes both. Independent CLI fixtures accept 24 equivalent payload cases and reject 20 changed/corrupt cases under normal and optimized Python.

Evidence: eb08b99 / research/pr1559-path-payload-audit. Exact source Git-blob/SHA256 pins, checker, deterministic receipt and reproduction commands are retained in the artifact.

Only two adapted comparison sites are repaired. The inherited full native shell pipeline has other path assumptions and was not run or claimed Windows portable. No frozen source is edited. UPSTREAM-FIX.patch targets pending PR1559, not main.

Next: Apply the small patch to PR1559 and expand its test to quote operand paths explicitly; then decide whether the rest of the shell pipeline should promise paths with spaces.

Credit: Hanson Sun PR1559; DreamingOfClouds PR589/545, eumemic PR555/493; Maxime Fleury/Antoine Olivier gzip-payload convention. Substantial OpenAI Codex assistance; no independent human review or new kappa claim.

Exact committed audit artifact. Commit eb08b99300f3388bc91d077fd500a8aff9153711; base 0b1cccf86e2f5132ed286a26ef0a681a495740ea. Coordinated batch, round 13.

[Mister-Meeseeks]

Batch review at c4f8af1dcf9fc2317c893b6b250007ded5ef879f.

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

Reviewed scope: Opt-in adapter changes only two regenerated gzip comparisons to complete decompressed-byte equality, retaining immutable source hashes and every downstream command. Actual shell-site recompression/corruption controls pass; default selected replay and frozen sources remain unchanged.

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.