Francesco Acerbi
Inter
Model vs baseline
The baseline knows his position and his expected minutes, nothing else. We publish the comparison whichever side wins. CRPS is lower-is-better, skill is the percentage improvement.
| Stat | Per match | Model CRPS | Baseline CRPS | Skill |
|---|---|---|---|---|
| Shots • | 0.78 | 0.4307 | 0.4631 | +7.0% ±5.5 |
| Shots on target • | 0.11 | 0.1116 | 0.0996 | -12.1% ±10.4 |
| Anytime scorer | 0 goals | 0.00032 | 0.00178 | +82.1% |
The scorer row is Brier score, not CRPS, because it is a yes or no market. It is reported but kept out of the predictability score. Our models expected 0.15 goals from him across these 9 starts.
Season by season
One entry per covered season. This is our accuracy over time, not his form.
| Season | Club | Starts | Score |
|---|---|---|---|
| 2022-23 | Inter | 9 | -2.5 |
| 2024-25 | Inter | 8 | -4.1 |
How to read this page
Everything here measures our forecasts, not the player. A score of -2.5 says how much closer our shot distributions landed than an informed benchmark that already knows the two strongest drivers of shooting volume: the player's position and how long he stays on the pitch. That benchmark is hard to beat consistently, and roughly a third of qualifying players do not. It is not a rating of how good he is, and a negative score means our model has work to do on him, not that he played badly.
Read it with the standard error beside it. The skill is a ratio of two small numbers, so it is much noisier for players who rarely shoot than for a striker with three or four attempts a match. Only started matches count, because a prediction made for a starting eleven is not a prediction about a substitute. Records are scored once the confirmed line-ups are announced ahead of kick-off.
A season is served from a single source. Seasons marked as a backtest were replayed by a model frozen before them, so the whole season is out of sample. Live seasons are reconciled after each matchday. See the full board, the Champions League title odds, and the methodology page.