Eight leagues · Priced before kickoff · Graded after
Every fixture priced. Before kickoff, and kept.
A model that prices every fixture in eight leagues and shows its working. Fair odds published before kickoff, graded when the result is in, never edited in between. Not tips.
Closed beta. Access is by invitation.
Next up · Wed 16 Sept
Priced 14 Sept, 20:07
- Wed 16 SeptAtlético Madrid66.1%Osasuna13.5%HOME 1.51DRAW 4.90AWAY 7.42Wed 16 SeptAtlético Madrid66.1%Osasuna13.5%HOME 1.51DRAW 4.90AWAY 7.42
- Wed 16 SeptDeportivo La Coruna47.1%Sevilla25.0%HOME 2.12DRAW 3.59AWAY 4.00Wed 16 SeptDeportivo La Coruna47.1%Sevilla25.0%HOME 2.12DRAW 3.59AWAY 4.00
- Wed 16 SeptBarcelona80.8%Racing Santander7.2%HOME 1.24DRAW 8.33AWAY 13.81Wed 16 SeptBarcelona80.8%Racing Santander7.2%HOME 1.24DRAW 8.33AWAY 13.81
Three of 74 fixtures priced across eight leagues. Probabilities and fair odds are one statement; the full board is behind the gate.
01
Priced, not picked
A tipster gives you a selection. We give you a probability and its fair price for every outcome, so you can see what the model thinks a bet is worth before you decide whether it is.
02
Published, then kept
Each price is stamped when it is published and is never edited after kickoff. What you see on a graded fixture is exactly what stood before the match, not a version tidied up afterwards.
03
Graded in public
Every settled price goes into one record that asks a single question: when the model said something was 30% likely, did it happen three times in ten? No hit rate. Calibration is the harder claim, and it is the one we make.
The record, so far
1152 match-outcome predictions on 384 fixtures across eight leagues, settled since 14 Aug 2026.
| Priced | Predictions | Mean priced | Happened | Difference |
|---|---|---|---|---|
| 10 to 20% | 52 | 15.4% | 17.3% | +1.9 |
| 20 to 30% | 538 | 26.3% | 28.8% | +2.5 |
| 30 to 40% | 255 | 34.1% | 32.9% | -1.2 |
| 40 to 50% | 198 | 44.5% | 41.4% | -3.1 |
| 50 to 60% | 67 | 53.9% | 44.8% | -9.2 |
The bands with at least 30 predictions. 54 await settlement, which runs on Mondays. The full table and the reliability chart are on the record page, updated weekly.
No accuracy figure. Deliberately.
A single hit rate flatters whoever picks the sample. A model that only ever backs 1.20 favourites is accurate and useless. The honest test is whether the prices are right in aggregate, band by band, and that is what the record page shows, including the bands where we are off.
The model is a Dixon-Coles foundation over an exact joint distribution of scorelines, with isotonic calibration on the published markets.
Built by a former professional poker player who priced edges for a living and kept the receipts.
Closed beta. Access is by invitation.
Invitations are issued in small batches through the Premier League season. Enter a code to open an account; sign in if you already have one.
Prices are the model's fair odds, not offers to bet. PunditModel is not a bookmaker and takes no stakes. 18+. Be gamble aware.