- Preserve the separation between
src/(package),tests/(developer checks),examples/demo/(reviewer-facing smoke test), andpaper/(manuscript reproduction). - Do not commit raw or controlled data.
- Run
ruff format,ruff check, andpytestbefore committing Python changes. - Keep figure calculations in scripts and export figure-facing source tables.
For work on CRM_R1 or revision Priorities 1-5, first read:
paper/revision/CRM_R1/ANALYSIS_PLAN.mdpaper/revision/CRM_R1/config/priority1_protocol.json- The priority-specific protocol under
paper/revision/CRM_R1/config/ paper/revision/CRM_R1/AGENTS.mdpaper/scripts/README.mdpaper/FIGURE_MAP.csv
Do not calculate the Priority 1 held-out 72 h endpoint while the protocol status is
DRAFT_NOT_FROZEN. Status FROZEN is still insufficient until
paper/revision/CRM_R1/scripts/17_check_priority1_freeze.py passes against the immutable external
freeze manifest.
For Priorities 3 and 4, preserve the frozen Priority 2 claim census and randomized review IDs. Never expose audit status, method membership, stability, or context-review outputs in a grading or rater packet. PubMed abstracts and raw XML are private working artifacts and must not be committed.