Redacted excerpt; full evidence withheld. $ pwd; ls -la; id -u; python3 --version; cat README.md 2>/dev/null | head -200; git status 2>&1 | head -3 exit: 0 pydantic/pull/13596) | MIT; core notice retained | | mypy-paramspec-unpack-crash | [mypy #22123](https://github.com/python/mypy/pull/22123) | MIT; typeshed notices retained | | qcew-county-shift-share | [BLS QCEW](https://www.bls.gov/cew/downloadable-data-files.htm) | US government public domain | The code-task PR authors and commit messages were reviewed for AI authorship markers; none were found. Undisclosed AI use cannot be ruled out. Data answers are generated deterministically by `build_reference.py`, not written by a model. Synthetic fixtures under `tests/` test the runner and are excluded from the benchmark pool. One pilot task from a third-party library was removed to keep the licence trail clean. Its historical results remain labelled `removed-pilot-task`; this tree cannot rerun it. Code snapshots and their upstream contact metadata are downloaded only after explicit preparation. Source and gold come from the pinned human-authored upstream commits, not generated code. The released data copies generalize public entity labels that overlap private terms; record [excerpt truncated]