# Reproduction notes

What can be reproduced from this Space alone: `code/pl_pilot/make_pl_pilot_figs.py --agg results/pl_pilot_aggregate.json --out figs/` regenerates figures 8–10 from the published aggregates.

What needs the private run folder (not published): `build_pl_pilot_public.py --src <run_dir> --out <dir>` rebuilds the aggregates from `results_main37.json`, `en_control_check.json`, `preflight_report.json` and the post-hoc `postanalysis.json` and fails if cross-file checks disagree; `pl_pilot_run.py` / `pl_pilot_lib.py` are the frozen measurement code (sha256 `4aa082d9…`, `2a96543c…`, matching the manifest); the Polish corpus is private, so a new run needs your own Polish realisation of the EvalAwareBench templates (check the licences first).

`pl_pilot_postanalysis.public.py` is the post-hoc script with six task identifiers redacted.

Run id `plpilot-04c874127fd7-20261017T183028Z-f212`; manifest `04c874127fd71d61b66c1e011855d21ac57c03a140ac8e85b4daec3685d8be87`. No new measurement was made for this page.