Evaluation Metrics
Phase 9 supports confusion matrices, accuracy, balanced accuracy, per-class precision/recall/F1/support, macro and micro F1, calibration bins, expected calibration error, Brier score, ROC/precision-recall reporting boundaries, IoU, Dice, track agreement, processing time, FPS, and cache hit rate.
Reports always include sample count, class distribution, coverage, abstentions, data origin, and warnings. Unsupported or underpowered calculations return a typed error or an explicit unavailable result.