Clean data doesn’t guarantee a useful model.

In highly imbalanced datasets, a model can achieve near-perfect accuracy by always predicting the majority class, while completely missing the cases that matter.

Standard validation checks won’t catch this. The schema can be correct, distributions can look stable, and no alerts will fire.

This chapter focuses on the data issues that slip through validation: imbalances and distribution mismatches that lead to poor real-world performance despite “clean” data.