Why this exists in ML If you use predicted probabilities for decisions, calibration is often more important than AUC.