Model track record
We publish how the model actually does. Every forecast is a pre-match snapshot, settled against the real result, with nothing edited after the fact. Below is the match-outcome accuracy and the Brier score, segmented by competition, type and month.
Matches scored
90
Outcome accuracy
68.9%
Brier · skill vs naive
0.481 +23.1%
By competition
| Segment | Matches | Accuracy | Brier | Skill |
|---|---|---|---|---|
| FIFA World Cup 2026 | 90 | 68.9% | 0.481 | +23.1% |
By competition type
| Segment | Matches | Accuracy | Brier | Skill |
|---|---|---|---|---|
| International | 90 | 68.9% | 0.481 | +23.1% |
By month
| Segment | Matches | Accuracy | Brier | Skill |
|---|---|---|---|---|
| June 2026 | 78 | 65.4% | 0.499 | +22.2% |
| July 2026 | 12 | 91.7% | 0.361 | +10.3% |
How to read this. Accuracy is the share of matches where the model's most-likely outcome (win, draw or loss) was correct. The Brier score measures how well-calibrated the whole probability vector was (0 is perfect, lower is better). Skill compares the model's Brier to a naive baseline that just predicts the base rate; a positive number means the model beat it.
Coverage is deepest for the ongoing World Cup today; other competitions segment in automatically as their forecasts settle. This is a live record, updated as matches finish.