Serie A model performance
Every published prediction is stored before kick-off and scored after the match. We report proper scoring rules, not cherry-picked win rates. The diagrams below are reliability plots from the frozen walk-forward samples. Favourite-correct on the live table is a noisy extra.
Italy. Model unversioned. Ledger updated 22 Sept 2026, 10:14 UTC. Dated quote: scored ledger note.
Serie A live sample
Published forecasts after kick-off. Over 2.5 and BTTS are Yes/No leans at 50%. Favourite-correct here is a noisy extra, not the quality number.
- Matches in view
- 0
- Favourite-correct
- —
- Prediction hits
- —
No scored matches for this view yet. Upcoming fixtures stay on the predictions pages until kick-off.
Serie A performance questions
- What is a reliability diagram?
- When the model assigns 70% to the favourite, those matches should be won about 70% of the time. The chart plots predicted favourite probability against how often that favourite won. Closer to the diagonal is better. ECE is the average gap.
- Why not lead with favourite-correct?
- A 50% live round can still be well calibrated, and a lucky round can look sharp. Log loss, RPS and this diagram measure whether the probabilities were honest. Favourite-correct stays in the table, last.
- Can I filter by league and round?
- Yes. Each league and each scored round has its own URL, so the table, title and stats are specific to that slice.