Cross-Validation

Cross-validation that matches the data

A defensible model-selection workflow: choose splits from the data-generating process, keep preprocessing inside each fold, and separate tuning from evaluation.