Football prediction accuracy: how our AI model performs
This is our honest scorecard. Every prediction gets graded against what actually happened on the pitch. The one question that matters: when we say a result has a 60% chance, does it really come in about 60% of the time? Here is the answer, measured on 56,049 real matches (Premier League, Bundesliga, La Liga, Serie A, Ligue 1, Eredivisie, Primeira Liga, Süper Lig, Championship, Super League Greece, Pro League, 2. Bundesliga, League Two, Ligue 2, Segunda División, Danish Superliga, Ekstraklasa, Liga I, Allsvenskan, Eliteserien).
- 56,049 matches graded
- Premier League, Bundesliga, La Liga, Serie A, Ligue 1, Eredivisie, Primeira Liga, Süper Lig, Championship, Super League Greece, Pro League, 2. Bundesliga, League Two, Ligue 2, Segunda División, Danish Superliga, Ekstraklasa, Liga I, Allsvenskan, Eliteserien
Yes. When we state a chance, reality lands within about 1.06 percentage points of it.That is market level honesty: our probabilities are as truthful as the betting market's own price.
Promised vs delivered
Each row is a group of predictions. The light bar is the chance we promised. The solid green bar is how often it actually happened. When the two line up, our percentages are honest.
Read it like this: in the 30-40% band we said those results were a roughly one in three shot, and reality agreed almost exactly. That close match, across every band, is the whole point of this site.
What this is, and what it is not
We would rather lose your click than your trust, so here it is, plainly.
What this is
- Honest probabilities you can actually trust, proven above.
- A strict No-bet discipline: most rounds we tell you to sit out.
- A public, time stamped log that updates every single day.
- A risk and sanity tool, built to protect a bankroll.
What this is not
- A tip sheet or a guaranteed win machine.
- A claim that we beat the sharp closing market. We do not.
- A reason to bet money you cannot afford to lose.
- Financial advice. The numbers inform, you decide.
Daily log129 checks on record
| Date (UTC) | Matches | Accuracy | Top pick wins | CLV |
|---|---|---|---|---|
| 2026-10-30 11:22:15 | 56,049 | 0.0106 | 47.8% | -3.18% |
| 2026-10-29 22:59:59 | 56,049 | 0.0106 | 47.8% | -3.18% |
| 2026-10-28 22:51:34 | 56,038 | 0.0106 | 47.7% | -3.18% |
| 2026-10-27 19:18:17 | 56,038 | 0.0106 | 47.7% | -3.18% |
| 2026-10-26 22:53:52 | 55,895 | 0.0107 | 47.7% | -3.17% |
| 2026-10-25 21:22:27 | 55,895 | 0.0107 | 47.7% | -3.17% |
| 2026-10-24 19:52:47 | 55,895 | 0.0107 | 47.7% | -3.17% |
| 2026-10-23 21:35:49 | 55,895 | 0.0107 | 47.7% | -3.17% |
| 2026-10-22 12:17:43 | 55,755 | 0.0108 | 47.7% | -3.16% |
| 2026-10-21 22:02:01 | 55,755 | 0.0108 | 47.7% | -3.16% |
| 2026-10-20 21:53:51 | 55,755 | 0.0108 | 47.7% | -3.16% |
| 2026-10-18 11:40:39 | 55,755 | 0.0108 | 47.7% | -3.16% |
| 2026-10-17 20:56:01 | 55,755 | 0.0108 | 47.7% | -3.16% |
| 2026-10-16 21:37:51 | 55,755 | 0.0096 | 47.7% | -3.16% |
| 2026-10-15 21:44:39 | 55,755 | 0.0096 | 47.7% | -3.16% |
| 2026-10-13 14:35:32 | 55,743 | 0.0096 | 47.7% | -3.16% |
Questions, answered straight
Plain definitions for every term on this page. No jargon left unexplained.
What is CLV, or closing line value?
The closing line is the final price the sharpest bookmakers settle on right before kickoff. Years of data show it is the most accurate price in football. CLV simply compares our price to that closing price. Positive means we were ahead of the market, near zero means we matched it. We publish it honestly, and we do not beat it.
What is calibration error?
It measures whether our percentages are truthful. If everything we call a 60% chance happens about 60% of the time, the calibration error is near zero. Lower is better. Ours sits at market level, which is about as honest as a probability gets.
What does promised vs delivered mean?
We take every prediction, group them by the chance we gave them, then check how often they actually happened. The graphic higher up shows the two side by side. When they match, our numbers are honest. That is the single most important thing on this page.
What is the top-pick hit rate?
In every match our model has one most likely outcome. This is how often that single pick comes in. It lands around half the time, because favourites win about half their matches. It is a sign of reliability, not a betting edge.
What is the Brier score?
An accuracy score for yes or no predictions, like Over 2.5 goals. Zero is perfect, 0.25 is a coin flip, lower is better. We show it so you can see the goal markets get graded too, not just the match result.
What is the sharp closing market?
The final odds from the sharpest bookmakers at kickoff. They absorb all the late money and information, which makes them very hard to beat. We are upfront that no model, including ours, beats them on public data.
What is a No-bet?
When no outcome offers enough value, we say No bet and sit the round out. That happens most rounds. Knowing when not to bet protects a bankroll more than any tip, so we treat the discipline as the real product.
What is the champion model?
The current best model, the one that produces the numbers you see (right now: the ensemble). It only keeps the title by beating every rival in honest, out of sample backtests. If a challenger wins, it takes over.
Why not just tell me what to bet?
Because honest probabilities plus knowing when to stay out is worth more than a tip sheet. We would rather lose your click than your bankroll. The numbers inform, you decide.
How often does this update?
Every day. The daily log above is the timestamped record of every check, so you can watch the numbers hold up over time instead of taking our word for it.
