Skip to content

Evaluate selected PR1094 final-stage replay on native Windows - #2336

Merged
Mister-Meeseeks merged 9 commits into
CrocSwap:mainfrom
antunishdPursuit:research/selected1094-windows-successor
Oct 15, 2026
Merged

Mister-Meeseeks merged 9 commits into
CrocSwap:mainfrom
antunishdPursuit:research/selected1094-windows-successor

Conversation

@antunishdPursuit

@antunishdPursuit antunishdPursuit commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor

The selected PR1094 final-stage entry point rejects native Windows relative-path keys before creating output. This separately versioned, source-pinned successor applies only PR2316's two documented .as_posix() substitutions in memory and completes all eleven inherited strict stages, retaining every original assertion and source guard.

Every one of the 105 outputs is compared against the reviewed PR1094 archive. On Windows, 88 files (including all 22 gzip outputs) are byte-identical; the remaining differences are explicitly checked log line endings, invoice JSON key order, six relocated input-path keys, and corresponding verified receipt hash links and stage timings. The assembly is value-equivalent under those dependency checks, not byte-identical. No blanket hash or volatile-field exclusions are used.

The branch includes the existing historical CI compatibility repairs from #2055 through 92276c22f6f419bbefd870bf2bb373bc41b0c643, preserving their original commits and credit. Historical PR53 clone/test jobs previously read optimized live helpers instead of their unchanged source-pinned versions. The shared repair executes the original checks in authenticated temporary copies without weakening source or output guards. This dependency is independently reviewable; no GitHub PR has been merged.

Validation at 217ba48

  • Native Windows 11 / CPython 3.12.14: metadata/pins and complete registered successor replay pass, including all 11 stages, 105 output comparisons and 16 deliberate corruption rejections. The optimized parent retains assertions in the successor and strict subprocesses.
  • 43 targeted regression tests pass with UTF-8 enabled; full reconciliation and source/archive integrity pass.
  • Fork Ubuntu CI: all 64 jobs across eight workflows pass, including the complete 52-job matrix and scoped exact tooling/provenance audit.
  • Staged whitespace checks passed; working tree is clean after the integration commit.

Limits and next step

Ten authenticated archived PR1042 local inputs; selected final stage only. No cold ancestry, new kappa, full-theorem verification, frozen-source edits or selected-frontier change is claimed. The fork's configured Linux CI validates historical checks; the new package's registered Linux replay remains queued upstream. Additional full historical Windows attempts are platform-limited by the missing C++ compiler and POSIX relative-path keys in the frozen PR53 receipt; those historical checks pass in Ubuntu CI. Maintainer adoption of a portable selected entry point remains open.

Canonical finding and next questions: #2333

Predecessor #2316 at 1b732744b66cc3461559d52349ae07fa31d444f8 supplies the two-key diagnosis and requests this successor execution. Credit JosieElliston's macOS final-stage portability audit, Rohan Arun, Jeremy Kothe, Douglas Colkitt, Wentao Li and the retained source lineage. Base canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. Source pins, native receipt, precise comparison rules and actionable next questions are in the package.

Substantial OpenAI Codex assistance in investigation, implementation, execution, self-review and writing; no independent human review claimed. New work is Apache-2.0.

Keep the PR53 SOURCE.json, original acceptance suite, frozen certificate,
producer, selected frontier and updated live exclusion circuit unchanged.
Run both the clones CI group and the affected isolated unittest module in
temporary tracked-file snapshots with the exact predecessor script retained
in research/exclusion-embedding-masks/baseline.py, preserving all other live
inputs. Permit only the historically pinned byte hash or Wentao Li's exact
reviewed support-mask successor; reject unrelated source drift and corrupted
baseline bytes in negative controls. Retain all existing CI diff gates.

Predecessors: PR53 skip-clones by Chafik Boukhalfa and collaborators;
Wentao Li's accepted ExclusionCircuit support-mask optimizations CrocSwap#2001.
Prepared by Dennis (antunishdPursuit) with substantial OpenAI Codex
assistance. Apache-2.0; no new mathematical acceptance claim.
@Mister-Meeseeks
Mister-Meeseeks merged commit 5ad5a20 into CrocSwap:main Oct 15, 2026
65 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants