Those who publish predictions show only the ones they got right. We do the opposite: every AI Analysis generated on GoalCast is verified automatically at the final whistle and the outcome, right or wrong, stays on record forever. This page shows all the numbers, with no selection and no exceptions; the data updates by itself as matches finish.
| Market | Correct outcomes | Accuracy |
|---|---|---|
| Double chance | 1020 of 1337 | 76% |
| Over/Under 2.5 | 797 of 1338 | 60% |
| Both teams to score (GG/NG) | 716 of 1258 | 57% |
| Full-time result (1X2) | 658 of 1243 | 53% |
| Suggested combos | 2085 of 5039 | 41% |
Likely scorers: 588 hit out of 5342 indicated (the hardest market, we say it plainly).
A model is calibrated when its percentages match real frequencies: among predictions given at 60-69%, about two out of three must come true. For every declared confidence band and every market, the table shows the share of predictions that actually hit. If the observed share falls within the band, the model is honest; if it falls below, it is over-promising; if above, it is too cautious.
Data updated on Sunday, 22 November 2026 (last automatic verification) over 1337 verified analyses, including those with confidence below 60% that are not published: calibration measures the model, not the selection. Bands with fewer than 20 predictions are not shown; the table is recomputed at every build.
| Declared confidence | Full-time result (1X2) | Double chance | Over/Under | Both teams to score |
|---|---|---|---|---|
| below 50% | 35% (63 of 179) | — | — | 75% (15 of 20) |
| 50-59% | 46% (200 of 439) | 58% (22 of 38) | 50% (271 of 539) | 51% (343 of 670) |
| 60-69% | 56% (228 of 410) | 70% (281 of 399) | 60% (289 of 483) | 63% (291 of 465) |
| 70-79% | 73% (114 of 156) | 76% (441 of 577) | 66% (106 of 161) | 64% (63 of 98) |
| 80-89% | 89% (42 of 47) | 83% (210 of 254) | 89% (49 of 55) | — |
| 90-100% | — | 96% (64 of 67) | — | — |
In short — Full-time result (1X2): average declared confidence 60%, correct outcomes 53% (−7 points); Double chance: average declared confidence 74%, correct outcomes 76% (+2 points); Over/Under: average declared confidence 62%, correct outcomes 58% (−4 points); Both teams to score: average declared confidence 59%, correct outcomes 57% (−2 points). The gap is the difference between correct outcomes and average confidence: negative when the model promises more than it delivers, positive when it is more cautious than necessary.
The same checks, split by competition (only leagues with at least 5 verified analyses). The percentage is the share of correct outcomes over the four main markets: 1X2, double chance, Over/Under and both teams to score.
| Competition | Verified analyses | Correct outcomes |
|---|---|---|
| Serie A | 136 | 71% (349 of 494) |
| UEFA Champions League | 95 | 69% (251 of 363) |
| La Liga | 85 | 64% (197 of 309) |
| Premier League | 70 | 60% (150 of 251) |
| Serie B | 64 | 58% (128 of 219) |
| Ligue 1 | 57 | 64% (130 of 202) |
| UEFA Europa League | 50 | 49% (98 of 200) |
| UEFA Europa Conference League | 49 | 59% (115 of 196) |
| World Cup | 49 | 72% (142 of 196) |
| Serie C - Girone A | 44 | 55% (96 of 176) |
| Serie C - Girone C | 36 | 50% (72 of 144) |
| Serie C - Girone B | 35 | 59% (83 of 140) |
| Bundesliga | 28 | 71% (79 of 112) |
| UEFA Nations League | 27 | 77% (80 of 104) |
| Coppa Italia Serie C | 20 | 60% (48 of 80) |
| Allsvenskan | 16 | 58% (37 of 64) |
| Challenge Cup | 13 | 63% (33 of 52) |
| Coppa Italia | 12 | 50% (24 of 48) |
| Championship | 10 | 55% (22 of 40) |
| K League 1 | 10 | 53% (21 of 40) |
| Eliteserien | 9 | 56% (20 of 36) |
| Liga MX | 9 | 69% (25 of 36) |
| Premier League | 9 | 69% (25 of 36) |
| Primeira Liga | 9 | 61% (22 of 36) |
| Pro League | 9 | 61% (22 of 36) |
When a match ends, an automatic process compares every prediction of the analysis (1X2, Over/Under, both teams to score, double chance, combos and scorers) with the real result and records the outcome in the database. There is no human intervention: nobody can choose what to count, and no prediction can be changed or deleted after publication.
The percentages include all the analyses generated on the platform, including those requested by users: there is no selected subset. Probabilities remain probabilities: a prediction with 60% confidence is expected to be wrong 4 times out of 10. The purpose of this page is not to promise winnings, but to let you judge the model with the same data we use.