RadarPremier League · Surprises against the model
2025-26 ↺ — reconstructed walk-forward estimates: each one calculated only from earlier matches and frozen on match day at 00:00 UTC (the site did not exist then). All leagues → How good is the model? →
364 matches with an estimate out of 380 played (teams with fewer than 8 matches in the model's window have no estimate).
The biggest upsets 2025-26 ↺
The matches in which the result happened that the model gave the smallest chance before kick-off. Example: if the model gave the away win 5% and the away side won, the match is near the top of the list. A 5% chance happens, on average, once every 20 matches.
- ↺model: 1 69% · X 20% · 2 10.7% · the actual result had 10.7% (≈ 1 in 9.32)
- ↺model: 1 68% · X 21% · 2 10.8% · the actual result had 10.8% (≈ 1 in 9.24)
- ↺model: 1 68% · X 19% · 2 13.4% · the actual result had 13.4% (≈ 1 in 7.49)
- ↺model: 1 67% · X 19% · 2 14.1% · the actual result had 14.1% (≈ 1 in 7.09)
- ↺model: 1 68% · X 18% · 2 14.5% · the actual result had 14.5% (≈ 1 in 6.89)
- ↺Liverpool FC 1–1 Burnley FCmodel: 1 77% · X 14.9% · 2 8% · the actual result had 14.9% (≈ 1 in 6.69)
- ↺model: 1 63% · X 22% · 2 15.0% · the actual result had 15.0% (≈ 1 in 6.66)
- ↺Wolverhampton Wanderers FC 2–1 Liverpool FCmodel: 1 15.3% · X 22% · 2 63% · the actual result had 15.3% (≈ 1 in 6.52)
- ↺Crystal Palace FC 2–3 Burnley FCmodel: 1 59% · X 25% · 2 16.2% · the actual result had 16.2% (≈ 1 in 6.16)
- ↺Chelsea FC 1–1 Burnley FCmodel: 1 73% · X 17.3% · 2 10% · the actual result had 17.3% (≈ 1 in 5.77)
- ↺model: 1 17.7% · X 22% · 2 60% · the actual result had 17.7% (≈ 1 in 5.64)
- ↺model: 1 66% · X 17.8% · 2 16% · the actual result had 17.8% (≈ 1 in 5.61)
- ↺model: 1 67% · X 19.1% · 2 14% · the actual result had 19.1% (≈ 1 in 5.23)
- ↺Tottenham Hotspur FC 1–2 West Ham United FCmodel: 1 58% · X 22% · 2 19.5% · the actual result had 19.5% (≈ 1 in 5.13)
- ↺model: 1 19.5% · X 24% · 2 57% · the actual result had 19.5% (≈ 1 in 5.12)
- ↺Wolverhampton Wanderers FC 2–0 Aston Villa FCmodel: 1 19.6% · X 26% · 2 54% · the actual result had 19.6% (≈ 1 in 5.11)
- ↺model: 1 64% · X 19.6% · 2 17% · the actual result had 19.6% (≈ 1 in 5.11)
- ↺Wolverhampton Wanderers FC 2–2 Arsenal FCmodel: 1 8% · X 19.9% · 2 72% · the actual result had 19.9% (≈ 1 in 5.04)
- ↺West Ham United FC 1–1 Manchester City FCmodel: 1 13% · X 20.0% · 2 67% · the actual result had 20.0% (≈ 1 in 5.01)
- ↺model: 1 18% · X 20.2% · 2 61% · the actual result had 20.2% (≈ 1 in 4.95)
How often low-chance results happen n = 364
All the possible results (home win, draw, away win) the model gave a small chance to, grouped into bands. For each band: how often they happened, next to what the model estimated on average. A well-calibrated model has the two numbers close together.
| the model's chance | results | happened | actual | estimated on average |
|---|---|---|---|---|
| 0% – 10% | 13 | 0 | 0.0%95%: 0–23% | 7.4% |
| 10% – 20% | 119 | 19 | 16.0%95%: 10–24% | 16.3% |
| 20% – 30% | 455 | 127 | 27.9%95%: 24–32% | 25.2% |
Results under 20% that came: 19; expected on average 20.3 (if the model is right, between 13 and 28, 95%).
Teams and upsets 2025-26 ↺
For each team: the results under 20% that went its way (a win or a draw) and against it (a defeat), next to how many were expected on average. “Expected” = the sum of those results' chances in its matches.
| Team | matches | in favour: actual | expected | against: actual | expected |
|---|---|---|---|---|---|
| Manchester United FC | 37 | 3 | 0.9 | 0 | 0.2 |
| Liverpool FC | 38 | 3 | 2.0 | 3 | 2.2 |
| Aston Villa FC | 37 | 1 | 0.3 | 1 | 0.9 |
| Tottenham Hotspur FC | 37 | 2 | 1.3 | 1 | 0.4 |
| Leeds United FC | 30 | 2 | 1.4 | 0 | 0.2 |
| Manchester City FC | 38 | 2 | 1.5 | 3 | 3.2 |
| AFC Bournemouth | 37 | 1 | 0.6 | 0 | 0.2 |
| Nottingham Forest FC | 37 | 1 | 0.9 | 0 | 0.2 |
| Chelsea FC | 38 | 1 | 0.9 | 1 | 1.1 |
| Brighton & Hove Albion FC | 38 | 1 | 1.0 | 0 | 0.3 |
| Sunderland AFC | 30 | 0 | 0.1 | 0 | 0.2 |
| Wolverhampton Wanderers FC | 36 | 3 | 3.3 | 0 | 0.0 |
| Arsenal FC | 37 | 1 | 1.3 | 2 | 3.9 |
| West Ham United FC | 37 | 2 | 2.5 | 0 | 0.0 |
| Fulham FC | 37 | 0 | 0.6 | 0 | 0.2 |
| Crystal Palace FC | 37 | 0 | 0.6 | 1 | 0.4 |
| Newcastle United FC | 37 | 0 | 0.7 | 0 | 0.5 |
| Everton FC | 37 | 0 | 0.8 | 0 | 0.2 |
| Brentford FC | 37 | 0 | 0.8 | 0 | 0.3 |
| Burnley FC | 36 | 3 | 4.6 | 0 | 0.0 |
This season · estimates frozen in real time ❄
The matches of the current season played after the model's estimate was frozen, within the last hour before kick-off, in which a result under 20% came.
20 matches played with a frozen estimate ❄; in 1 of them a result under 20% came.
- ❄model: 1 58% · X 23% · 2 19.4% · the actual result had 19.4% (≈ 1 in 5.14)
How to read it
How to read the figures on this page and what they do not say.
- A match's upset = the chance the model gave, before kick-off, to the result that happened (home win, draw or away win). The smaller it is, the less the model expected the result.
- A result with a 20% chance happens, on average, once every 5 matches. Upsets are normal: the right question is whether they come as often as the model says — the “How often” table answers it.
- Estimates marked ↺ are reconstructed: calculated only from earlier matches, but the site did not publish them at the time. Those marked ❄ were frozen and published within the last hour before kick-off.
- The figures say what happened against the estimate, not why.
← Premier League: standings, results
Updated: 25 Oct 2026, 07:20 UTC