I explained differences between target leakage and train/test contamination, recommended strict temporal splits, feature provenance checks, cross-validation strategies, and using a separate holdout set. I gave an example where a future-derived feature would leak and how I removed it and reran validation.