Detect future information leaking into features, labels, or evaluation. Use when any pipeline step might expose data not yet available at prediction time.
