REVIEW PREPARATION ARTIFACT — LOCAL DRAFT

This ZIP is a local preparation copy. It has not been externally submitted,
distributed, peer reviewed, or accepted. The submitted manuscript is supplied
separately; original manuscript source/PDF are intentionally omitted here.

Included: implementation, configurations where separate files exist, tests,
dependency locks, protocol and claim-to-evidence documentation, full and pilot
results, verification records, and previously generated tables/figures.
Git metadata, CI workflows, software caches, and unrelated or untracked files
are excluded. All numerical results, source code, protocols, and test logic
are preserved without change. No new experiments were run while packaging.
Generated table fragments remain under paper/generated/ because the analysis
scripts use that output directory; they are not a manuscript submission.

CONTENTS.json lists files in this review copy. Detailed original/export hashes
and copy transformations are retained outside this ZIP in local audit records.
Original experiment manifests retain their scientific provenance checksums;
those checksums refer to the unchanged experiment sources and result data.

NAMED REVIEW COPY
The original project MIT copyright notice and third-party notices are intact.
Reviewer-facing author names are intentionally retained in this named copy.

REPRODUCTION
Use Python 3.14 for exact AST/source-hash reproduction. Recorded execution used
Python 3.14.7; analysis used Python 3.12.13, NumPy 2.5.3, Matplotlib 3.11.2.
From this directory, create/activate an environment and run:
  python -m pip install -r requirements-lock.txt
  python scripts/prepare.py
  python -m unittest discover -s tests -v
  python scripts/verify.py
  python scripts/run.py --pilot --output reproduction/pilot
  python scripts/run.py --output reproduction/full
  python scripts/verify.py --reproduction reproduction/full --execute-sample

prepare.py requires Git and network access to fetch the pinned public QuixBugs
commit 4257f44b0ff1181dedaedee6a447e133219fcebf. The external checkout is not
bundled. data/provenance.json and third_party/ retain provenance and notices.
The verifier without --execute-sample audits stored results and replays policy
selections without requiring the external checkout. To regenerate figures and
tables from retained results in this disposable review copy:
  python scripts/analyze.py
This last command rewrites derived summaries/tables/figures in the review copy.

Scientific scope: reference-oracle-assisted mutation diagnostic, not an
assessment of AI-generated repository patches. Historical aborted runs are
retained so protocol deviations remain inspectable.
