CLAUDE.md

1# ML Evaluation Memory
2 
3Apply this inside `ml/evaluation/`.
4 
5- Report validation setup, data splits, metrics, and known limitations.
6- Prefer clinically cautious interpretation of model quality.
7- Track false positives, false negatives, calibration, and subgroup concerns when relevant.
8- Evaluation artifacts should support review, not overstate readiness.
9- Summaries should state what the model can and cannot support.
10