Standalone study repository, migrated from the model-selection article series.
- Historical-evidence paper (PDF)
- Paper source and evidence mapping
- Maintenance audit
- Latest detailed research article — preserved before the readable blog replacement, including October corrections
- Original detailed article
- Executable companion
- Source provenance
- Byte-identical published original
- Migration preservation audit
- Recovered original prediction arrays
The original measurements and code are imported. The historical-evidence manuscript and PDF are complete; independent scientific reproduction remains pending. The generic harness configuration is unconfigured and must not be used to claim verified results. Follow the companion README for the original runnable workflow. No new experiment, human approval, or independent reproduction is claimed by this migration.
The repository holds the detailed research article, methods and paper. The blog provides the accessible explanation; its replacement is prepared separately from this preservation commit. Original third-party licences and notices remain applicable; no blanket relicensing is applied.
Run make setup followed by make test for the maintained companion checks. These
synthetic tests do not reproduce the scientific panel. Optional tests clearly skip when
pinned data or completed-run files are absent.
make paper requires Python 3.11 (managed by uv), pdfLaTeX and BibTeX. It verifies the
preserved evidence, regenerates numerical tables and the historical claim ledger, and
builds paper/build/main.pdf. The checked-in PDF and receipt record a clean source
snapshot and the actual TeX engine versions. This formatting build does not download
data, generate embeddings, fit models or rerun the exploratory analysis. Original
publication archives remain unchanged; use companion/README.md for maintained code.