Model estimates

Probability describes which player the model favours for this matchup. It does not describe certainty that the prediction is correct. A player with the stronger historical record can still lose the model’s matchup-specific estimate because recent form, head-to-head context, round performance, schedule effects, and other features also matter. Round scores are point estimates, not guaranteed outcomes or calibrated confidence intervals. Stored evaluation uses a historical held-out sample; production refreshes reuse the saved model without retraining or rerunning that evaluation.