We check our work
Model calibration and skill, open to all. Including the failures.
To keep ourselves honest: we publish model calibration metrics with full transparency, just like we do with the pick record.
Methodology
Brier score measures mean squared error between the predicted probability and the actual outcome (0 = perfect, 1 = worst). Lower is better.
The naive baseline uses each market's empirical win rate as a constant forecast: the weakest possible reference. Beating the naive baseline is the minimum expected of any model.
This page is a technical audit of model predictions. It does not constitute betting advice. Past predictions do not guarantee future results.