How Calibrated Sports Models Work

A raw model might say "70% home win" but only win 55% of the time at that confidence level. Calibration measures whether stated probabilities match real outcomes — the foundation of honest sports analytics.

Key metrics we publish

• Hit rate — percentage of resolved predictions that were correct.
• Brier score — measures probability accuracy (lower is better; 0 = perfect).
• Calibration curve — when we say 60%, do outcomes happen ~60% of the time?

Why we recalibrate daily

Team form, injuries and league dynamics shift. Probascore runs a calibration monitor per market and adjusts display confidence when models drift. You see the numbers — we don't hide bad weeks.

Ensemble + Poisson

Football uses a four-signal ensemble (momentum, Poisson xG, head-to-head, defence differential) before calibration. Other sports use tailored pipelines (e.g. total points for basketball/NHL).

Open the backtest dashboard or try Pro for full model breakdown on Premier.