Why this exists in ML Most real tasks have asymmetric error costs; precision/recall surface those tradeoffs.