Log Loss

A scoring rule for probability forecasts that heavily penalizes confident wrong predictions. Lower is better; it is the primary metric used to gate whether a machine-learning model is good enough to promote.

• Brier Score — A forecast-accuracy score — the average squared error between predicted probabilities and outcomes. Lower is better.
• Calibration — How well a model's stated probabilities match reality — a well-calibrated 60% pick wins about 60% of the time. Shown on a reliability curve.