21. Machine learning evaluation

Create honest train, validation, and test splits; select metrics; detect leakage; tune models; estimate uncertainty; and compare against meaningful baselines.