Serie A Model Accuracy and Prediction Record

This record uses only Serie A matches. Historical validation and forecasts recorded before kickoff are separate.

The independent historical test covered 585 matches: the model's most likely result happened 53.0% of the time, with log loss 0.985.

Historical rolling fits assume results are available three hours after kickoff. Date-only archive results conservatively become available three hours after the end of their match date in UTC. Parameter selection, calibration and the final test are separate chronological periods. Validation through 1 Oct 2026, 00:00 UTC; this is not a guarantee of future accuracy.

Recorded before kickoff

The live record will appear as forecasts reach the pre-24-hour recording window. No historical matches receive invented pre-match predictions.

Sources and limitations

Fixtures, results and provider standings: football-data.org. Completed score history since 2020-21: OpenFootball and football-data.org. Forecasts use only results available before their publication time.

Older archive records without a verified timezone use the end of the match date in UTC. Recent seasons use exact football-data.org timestamps. This keeps historical results out of training until their conservative availability time.

No verified injury or confirmed-lineup adjustment is included for Serie A. Promotion playoffs and promotion probabilities are outside this model. Probabilities are estimates, not guarantees. Check this league's model record.