Standalone study repository, migrated from the model-selection article series.
The original measurements and code are imported. A LaTeX manuscript reporting these existing results is
in paper/main.tex, with the compiled PDF at paper/build/main.pdf.
Rebuild it with make paper-imported (system Python and local TeX Live only; it formats archived metrics
and compiles, and runs no experiment, data download or environment creation). Provenance for the manuscript
is in evidence/paper-migration/: claims ledger, content-coverage map, build
receipt, independent AI review and status.
Independent reproduction is pending: protocol.md records the imported-study status; the new harness execution remains unapproved. The generic harness configuration must not be used to claim verified results. Follow the companion
README for the original runnable workflow. No new experiment, human approval, human review or independent
reproduction is claimed by this migration.
The repository will hold the detailed methods and paper; the blog will provide a shorter accessible explanation. Original third-party licences and notices remain applicable; no blanket relicensing is applied.
Run make test for companion regression tests and archive integrity checks. PR CI runs these checks; it does not retrain models or establish scientific reproduction. The repository audit records scope and limitations. Maintained companion code includes safeguards added after the historical run; the original downloadable code and results archives remain unchanged.