Our track record
Every finished match is checked, every night: how often our picks landed against the chance we gave them, and our forecast of the result beside the market's closing prices.
Our picks, graded
650 of 899 picks landed (72%). Our numbers gave them 75% on average.
The soundest pick of every match since October 18, 2026, chosen as on our Predictions today page from the estimate saved before kick-off, and checked against the score after 90 minutes. The page itself started on 3 November 2026; the picks before then apply the same rule to the estimates saved at the time.
Last 7 days: 53 of 75 landed (71%); our numbers gave them 75%.
By the chance we gave
| We gave | Picks | Landed | Our chance, on average |
|---|---|---|---|
| 50–60% | 4 | 50% | 53% |
| 60–70% | 144 | 65% | 67% |
| 70–80% | 751 | 74% | 76% |
By kind of pick
| Pick | Picks | Landed | Our chance, on average |
|---|---|---|---|
| Double chance | 292 | 71% | 74% |
| Over 1.5 goals | 272 | 77% | 77% |
| Under 3.5 goals | 196 | 73% | 76% |
| Over 2.5 goals | 51 | 69% | 70% |
| Under 2.5 goals | 19 | 42% | 70% |
| A team to win | 57 | 77% | 72% |
| Both teams to score, or not | 12 | 33% | 62% |
In chats: 144 of 194 picks given before kick-off landed (74%); our numbers gave them 71%. Of 38 combos, 10 landed in full (26%); our numbers gave them 28% on average.
If our numbers are honest, picks we give 70% should land about 7 times in 10. When they land less often than the chance we gave, our numbers were too sure; more often, too cautious. This page shows which, either way.
These are probabilities, not promises: a 70% pick still fails three times in ten, and a good week or a bad one is a small sample.
Our result forecast against the market
Matches from October 4, 2026 to November 21, 2026, model ratings/1, ratings/2, ratings/2.2, ratings/2.3, ratings/2.4, ratings/2.5.
| Matches | Count | Brier (Daro / Market) | Log loss (Daro / Market) | Hit rate (Daro / Market) |
|---|---|---|---|---|
| All scored matches | 1135 | 0.586 / 0.560 | 0.984 / 0.945 | 54% / 56% |
| Saved at kick-off | 856 | 0.587 / 0.559 | 0.986 / 0.945 | 53% / 56% |
| Last 30 days | 816 | 0.585 / 0.560 | 0.982 / 0.945 | 54% / 56% |
233 more matches were scored without the market, which had no closing price for them.
By competition (10 matches or more)
| Matches | Count | Brier (Daro / Market) | Log loss (Daro / Market) | Hit rate (Daro / Market) |
|---|---|---|---|---|
| Primera División | 116 | 0.620 / 0.589 | 1.028 / 0.981 | 50% / 51% |
| Serie A | 105 | 0.527 / 0.491 | 0.906 / 0.854 | 65% / 70% |
| Major League Soccer | 92 | 0.615 / 0.600 | 1.026 / 1.001 | 51% / 55% |
| FA Cup | 78 | 0.643 / 0.597 | 1.070 / 0.994 | 49% / 51% |
| Primera A | 66 | 0.565 / 0.527 | 0.955 / 0.897 | 56% / 58% |
| Liga Profesional Argentina | 64 | 0.615 / 0.582 | 1.027 / 0.973 | 45% / 47% |
| UEFA Champions League | 54 | 0.511 / 0.474 | 0.882 / 0.833 | 65% / 69% |
| UEFA Europa Conference League | 54 | 0.548 / 0.507 | 0.921 / 0.864 | 57% / 63% |
| UEFA Europa League | 54 | 0.584 / 0.562 | 0.983 / 0.953 | 52% / 57% |
| Liga MX | 50 | 0.649 / 0.621 | 1.071 / 1.029 | 44% / 54% |
| Copa del Rey | 50 | 0.589 / 0.530 | 0.979 / 0.889 | 50% / 58% |
| La Liga | 43 | 0.572 / 0.544 | 0.967 / 0.922 | 58% / 56% |
| Premier League | 41 | 0.574 / 0.570 | 0.967 / 0.962 | 49% / 51% |
| Bundesliga | 37 | 0.576 / 0.587 | 0.972 / 0.989 | 57% / 60% |
| Ligue 1 | 37 | 0.529 / 0.523 | 0.900 / 0.891 | 62% / 62% |
| Liga Pro | 33 | 0.705 / 0.752 | 1.159 / 1.235 | 30% / 36% |
| Primeira Liga | 27 | 0.513 / 0.523 | 0.889 / 0.915 | 63% / 59% |
| Africa Cup of Nations Qualifiers | 18 | 0.698 / 0.731 | 1.139 / 1.173 | 44% / 33% |
| DFB Pokal | 16 | 0.414 / 0.415 | 0.738 / 0.736 | 75% / 69% |
| International Friendlies | 16 | 0.549 / 0.445 | 0.930 / 0.766 | 50% / 63% |
| Copa Uruguay | 14 | 0.534 / 0.504 | 0.912 / 0.867 | 57% / 64% |
| CONCACAF Nations League | 12 | 0.481 / 0.457 | 0.829 / 0.774 | 75% / 67% |
| Coupe de France | 11 | 0.463 / 0.406 | 0.809 / 0.730 | 73% / 64% |
How to read it
The Brier score measures a forecast of the three outcomes (home, draw, away): 0 is perfect, and saying a third each for every match scores 0.667. Log loss punishes confident misses harder. For both, lower is better.
Hit rate is how often the outcome we thought likeliest happened. It ignores how sure we were, so the Brier score is the fairer measure.
The market is the bookmakers' closing prices with their margin removed: the hardest benchmark there is, because it already knows line-ups and news.
Saved at kick-off, or rebuilt
Since 18 October 2026, our estimate is saved just before each kick-off and never edited. For matches before that, the estimate was rebuilt afterwards from the matches played before kick-off, with today's method: close to what was shown, but not a record. Both are counted, and the saved ones are shown on their own.
The backtest
Before release, the model was tested week by week on 9,190 matches (August 2025 to October 2026): a Brier score of 0.608 for the result, against 0.642 for guessing each league's average.