PBS and PLL are superior evaluation metrics for probabilistic classifiers, fixing flaws in Brier Score (MSE) and Log Loss (Cross-Entropy). Strictly proper, consistent, and better for model selection, early stopping, and checkpointing.
