Model evaluation framework — cross-validation strategies, metric selection by task type, baseline requirements, ablation design, and train/val/test discipline. Use for all ML experiments before reporting results.
