Accuracy, in the open
Every KnowScore prediction is graded after the match. This page shows the numbers from several angles, including the places where the AI is weaker than the betting market. We publish the weak spots too.
1. Accuracy by confidence
Each prediction carries a confidence score from 0 to 100. Picks the AI is more sure about have been right more often.
2. Do the percentages mean what they say?
If a forecast is well calibrated, picks rated around 60% should be right around 60% of the time. Here is how the AI's top pick did, by the probability it gave. Matches where two outcomes tied for the top spot are left out.
| Probability we gave | Picks | We said (average) | It was right |
|---|---|---|---|
| under 40% | 219 | 37.2% | 33.3% |
| 40–49% | 431 | 44.3% | 45.7% |
| 50–59% | 347 | 54.3% | 64% |
| 60–69% | 235 | 63.9% | 70.6% |
| 70%+ | 168 | 76.8% | 81.5% |
For example, picks we rated 50–59% (on average 54.3%) were right 64% of the time. For picks under 50% the two columns are close. In the middle of the range the picks have come in more often than the percentage we stated, so our numbers have tended to be on the cautious side.
3. The AI against the betting market
For matches where we stored bookmaker odds (since 16 September 2026), we compare the AI's top pick with the favourite implied by the bookmakers' average odds, with the bookmakers' margin removed. "Always pick the home team" is a simple baseline.
| Pick | Right | Matches | Hit rate |
|---|---|---|---|
| KnowScore AI | 393 | 694 | 56.6% |
| Bookmaker favourite | 408 | 694 | 58.8% |
| Always the home team | 323 | 694 | 46.5% |
Over 694 matches since 16 September 2026, the bookmaker favourite was right 58.8% of the time and the AI's pick 56.6%. The market was slightly ahead. Betting markets pool a huge amount of information, so they are hard to beat, and KnowScore does not claim to beat them. What the app adds is the reasoning behind each number: injuries, lineups, form and squad value, in one place.
The same comparison on the full probabilities (Brier score, where 0 is perfect, lower is better and always guessing a third each scores 0.667): AI 0.568, bookmakers 0.540.
4. When the AI and the market agree, and when they don't
In 84.2% of matches the AI's top pick was also the bookmakers' favourite. In the other 15.8% the two disagreed.
In the matches where they disagreed, the bookmakers gave their favourite 41.5% on average, against 57.2% when the AI agreed, so these are mostly closer contests. In those matches the bookmaker favourite was right more often than the AI's pick. When both point the same way, the pick has been right more often than the AI's overall average. 5 matches where the AI had no single top outcome are not counted here.
5. Week by week
Single weeks swing a lot with about 100 matches each. Judge the trend over months, not days.
| Week of | Matches | AI right | Bookmaker favourite right |
|---|---|---|---|
| 14 Sept (from 16 Sep) | 107 | 49.5% | 47.7% |
| 21 Sept | 57 | 59.6% | 57.9% |
| 28 Sept | 79 | 51.9% | 60.8% |
| 5 Oct | 108 | 57.4% | 58.3% |
| 12 Oct | 103 | 64.1% | 65% |
| 19 Oct | 120 | 52.5% | 53.3% |
| 26 Oct | 92 | 63% | 68.5% |
| 2 Nov | 28 | 57.1% | 67.9% |
6. By competition
All graded predictions, competitions with at least 30 so far. Friendlies and the World Cup come from the earlier months.
| Competition | Predictions | Right |
|---|---|---|
| Friendlies | 163 | 57.1% |
| Champions League | 162 | 64.2% |
| La Liga | 109 | 50.5% |
| World Cup | 104 | 69.2% |
| Nations League | 104 | 55.8% |
| Serie A | 100 | 59% |
| Eredivisie | 99 | 56.6% |
| Premier League | 90 | 48.9% |
| Primeira Liga | 90 | 55.6% |
| Süper Lig | 90 | 48.9% |
| Jupiler Pro League | 90 | 60% |
| Ligue 1 | 81 | 54.3% |
How we grade
- A pick is right when the result matches the outcome with the highest predicted probability: home win, draw or away win. If two outcomes tie for the top spot, either one counts. In knockout ties the team that advances is graded.
- The prediction graded is the final one made before kick-off, so it includes the effect of official lineups.
- Youth teams are excluded. Grading is automatic: nobody chooses which matches count.
- The model has been adjusted since launch (for example on 17 October 2026), so the overall figures include earlier versions. The AI-against-market comparisons only cover matches since 16 September 2026, when we began storing bookmaker odds.
These figures describe how often predictions were right. They are not betting advice and say nothing about profit. KnowScore does not publish profit or return figures and does not accept bets.