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 (draft v2, unapproved) incorporates corrections from an
automated AI methods review of draft v1 and has not been executed. 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, 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 the maintained companion's offline tests. Data-dependent tests skip
until the companion workflow has produced its inputs; a green offline check is not a
scientific reproduction. Push and pull-request CI also validates archived manuscript tables.
The historical archives are unchanged. See the review record
for findings, fixes and remaining reproduction limits.