How accurate are our football predictions?
Since 23 July we have published 1,517 settled tips. On average the model claimed 74.7%, and 77.1% landed. Everything below is worked out from the predictions as they stood at kick-off, losers included.
Every tip we published
The tips on the daily predictions page, day by day, as the page published them. Claimed is the model's average probability; landed is how many came in. A gap near zero means the confidence was honest. Rows under 20tips are greyed: too few to read much into. The page's list can change during the day as bookmaker prices move, so this is the list as it stood at kick-off.
Previous models: 23 July to 26 October
| Tips | Claimed | Landed | Gap | |
|---|---|---|---|---|
| Home or Draw | 455 | 83.7% | 83.7% | +0.0 |
| Over 2.5 | 442 | 67.2% | 67.0% | -0.3 |
| Home Win | 328 | 71.4% | 76.2% | +4.8 |
| Away or Draw | 183 | 78.8% | 84.2% | +5.3 |
| Away Win | 95 | 68.3% | 81.1% | +12.7 |
| Over 3.5 | 8 | 90.4% | 87.5% | -2.9 |
| Tips | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 50–55% | 1 | 53.2% | 0.0% | -53.2 |
| 55–60% | 62 | 58.6% | 50.0% | -8.6 |
| 60–65% | 119 | 62.2% | 75.6% | +13.4 |
| 65–70% | 321 | 66.6% | 70.7% | +4.1 |
| 70–75% | 218 | 72.4% | 70.6% | -1.7 |
| 75–80% | 221 | 77.5% | 82.8% | +5.3 |
| 80–85% | 425 | 82.1% | 83.1% | +0.9 |
| 85–90% | 133 | 86.7% | 88.7% | +2.0 |
| 90–95% | 6 | 91.2% | 66.7% | -24.5 |
| 95–100% | 5 | 98.0% | 100.0% | +2.0 |
Current models: 27 October to 27 October
| Tips | Claimed | Landed | Gap | |
|---|---|---|---|---|
| Home Win | 3 | 76.2% | 66.7% | -9.5 |
| Away or Draw | 2 | 79.5% | 100.0% | +20.5 |
| Home or Draw | 1 | 80.3% | 100.0% | +19.7 |
| Tips | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 70–75% | 2 | 73.3% | 50.0% | -23.3 |
| 75–80% | 1 | 77.7% | 100.0% | +22.3 |
| 80–85% | 3 | 81.2% | 100.0% | +18.8 |
The league models were retrained and went live on 27 October, so their record is kept separate rather than pooled with the old models'.
Every probability, not just the tips
Tips are the confident end. This is every probability the model stored for a league match since 23 July, grouped in 5-point bands: when it said 40 to 45%, how often did it happen? Bands with fewer than 20 matches are greyed.
| Matches | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 5–10% | 1 | 8.2% | 0.0% | -8.2 |
| 10–15% | 31 | 14.7% | 3.2% | -11.5 |
| 15–20% | 177 | 18.1% | 7.9% | -10.2 |
| 20–25% | 285 | 22.7% | 17.5% | -5.2 |
| 25–30% | 765 | 28.1% | 27.5% | -0.6 |
| 30–35% | 1352 | 32.5% | 34.7% | +2.2 |
| 35–40% | 1940 | 37.7% | 36.3% | -1.4 |
| 40–45% | 2584 | 42.5% | 42.7% | +0.2 |
| 45–50% | 1997 | 47.4% | 47.3% | -0.1 |
| 50–55% | 1442 | 52.4% | 51.0% | -1.4 |
| 55–60% | 999 | 57.2% | 56.3% | -0.9 |
| 60–65% | 444 | 62.3% | 64.2% | +1.9 |
| 65–70% | 293 | 67.4% | 70.0% | +2.6 |
| 70–75% | 170 | 72.1% | 72.9% | +0.8 |
| 75–80% | 94 | 77.6% | 89.4% | +11.8 |
| 80–85% | 51 | 81.2% | 84.3% | +3.1 |
| 85–90% | 2 | 88.2% | 100.0% | +11.8 |
| Matches | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 5–10% | 75 | 9.0% | 9.3% | +0.3 |
| 10–15% | 205 | 12.9% | 13.2% | +0.3 |
| 15–20% | 716 | 17.9% | 19.0% | +1.1 |
| 20–25% | 3313 | 23.3% | 23.0% | -0.3 |
| 25–30% | 6813 | 27.1% | 26.2% | -0.9 |
| 30–35% | 1471 | 31.5% | 28.6% | -2.9 |
| 35–40% | 34 | 35.8% | 35.3% | -0.5 |
| Matches | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 0–5% | 2 | 4.5% | 0.0% | -4.5 |
| 5–10% | 38 | 9.2% | 0.0% | -9.2 |
| 10–15% | 514 | 12.9% | 8.8% | -4.1 |
| 15–20% | 1268 | 17.8% | 18.9% | +1.1 |
| 20–25% | 2246 | 22.7% | 23.9% | +1.2 |
| 25–30% | 2841 | 27.5% | 28.0% | +0.5 |
| 30–35% | 2444 | 32.2% | 34.0% | +1.8 |
| 35–40% | 1418 | 37.4% | 38.2% | +0.8 |
| 40–45% | 890 | 42.4% | 41.5% | -0.9 |
| 45–50% | 449 | 47.0% | 53.0% | +6.0 |
| 50–55% | 196 | 52.2% | 53.6% | +1.4 |
| 55–60% | 118 | 57.3% | 75.4% | +18.1 |
| 60–65% | 110 | 62.6% | 67.3% | +4.7 |
| 65–70% | 52 | 67.3% | 82.7% | +15.4 |
| 70–75% | 40 | 72.0% | 82.5% | +10.5 |
| 80–85% | 1 | 82.3% | 100.0% | +17.7 |
| Matches | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 0–5% | 1 | 2.0% | 0.0% | -2.0 |
| 20–25% | 7 | 21.2% | 71.4% | +50.2 |
| 25–30% | 1 | 29.3% | 0.0% | -29.3 |
| 30–35% | 368 | 32.7% | 35.6% | +2.9 |
| 35–40% | 775 | 37.4% | 41.3% | +3.9 |
| 40–45% | 2410 | 43.3% | 46.8% | +3.5 |
| 45–50% | 3660 | 48.3% | 51.9% | +3.6 |
| 50–55% | 2842 | 53.4% | 58.8% | +5.4 |
| 55–60% | 1201 | 58.3% | 62.9% | +4.6 |
| 60–65% | 397 | 61.3% | 66.8% | +5.5 |
| 65–70% | 780 | 65.5% | 63.3% | -2.2 |
| 70–75% | 93 | 72.4% | 66.7% | -5.7 |
| 75–80% | 80 | 77.2% | 72.5% | -4.7 |
| 90–95% | 11 | 90.9% | 100.0% | +9.1 |
| 95–100% | 1 | 98.0% | 100.0% | +2.0 |
| Matches | Claimed | Landed | Gap | |
|---|---|---|---|---|
| 20–25% | 1 | 22.6% | 0.0% | -22.6 |
| 30–35% | 13 | 32.5% | 53.8% | +21.3 |
| 35–40% | 59 | 37.6% | 33.9% | -3.7 |
| 40–45% | 726 | 43.4% | 46.3% | +2.9 |
| 45–50% | 2294 | 47.6% | 50.6% | +3.0 |
| 50–55% | 5275 | 52.8% | 54.9% | +2.1 |
| 55–60% | 2960 | 57.2% | 59.1% | +1.9 |
| 60–65% | 1155 | 61.7% | 60.7% | -1.0 |
| 65–70% | 124 | 66.4% | 72.6% | +6.2 |
| 70–75% | 19 | 71.2% | 57.9% | -13.3 |
| 75–80% | 1 | 76.2% | 100.0% | +23.8 |
How to judge any football prediction site
Ask for four things. The sample size: a 90% hit rate over twenty tips says almost nothing. The claimed probability next to the result: a hit rate means little without knowing how sure the site said it was. The whole record: a winning month chosen after the fact is marketing. And the prices: anyone can hit 90% backing 1.05 shots.
This page covers the first three. The price is on every tip on the daily predictions page, the track record has every Slip of the Day with its result, and the methodology explains how the probabilities are made.
Frequently asked questions
- What does it mean for a prediction to be accurate?
- For a probability, accurate means calibrated: of all the times the model says 70%, about 70% should come in. A model that only ever predicts the obvious favourite can have a high hit rate and still be badly calibrated. That's why this page shows what was claimed next to what landed, not a hit rate on its own.
- Why don't the tips land 90% of the time?
- Because a 75% tip should lose one time in four. A hit rate means little without the prices: backing very short favourites gets a high hit rate and still loses money. What matters is whether the claimed probability matches what lands.
- Are these numbers edited after the match?
- No. Every prediction is frozen at kick-off, and has been since 23 July. Earlier predictions were re-scored when the models were retrained, so they aren't counted here.