Track record
Every match is forecast before kick-off and checked against the real result after full time. Hits and misses, all shown: 1884 games from Sat 10 Oct to Mon 16 Nov. Updated every night.
1012 of 1884 games
10% of games
338 of 479
(3+ goals or not)
Day by day
Share of results called right each day (days with 5+ games). Small days swing a lot; the total above is the fairer guide.
Do the percentages hold up?
Grouped by how strongly we rated the favourite. The bar is how often the favourite actually won; the thin line is what we said. Close together means the percentages can be trusted.
How we compare with the bookmakers
On the 766 league games where we have the bookmakers' closing prices, our favourite and theirs got 29 fewer results right (415 v 444 of 766). Getting the favourite right is only half of it, so we also score how close the percentages were (lower is better): the bookmakers' were a little closer to what happened. On 766 games the bookmakers are ahead on this measure; we show it either way.
| xGenius | Bookmakers | |
|---|---|---|
| Favourite won | 415 of 766 (54%) | 444 of 766 (58%) |
| Brier scoreHow far the home, draw and away percentages were from what happened. Lower is better. | 0.584 | 0.562 |
| Log lossLike the Brier score, but punishes confident misses harder. Lower is better. | 0.983 | 0.950 |
| Over or under 2.5 goals, BrierLower is better. | 0.245 | 0.236 |
Bookmakers' prices are the closing averages from football-data.co.uk (the pre-match average where a closing price is missing), with the bookmakers' margin taken out, used only as a yardstick. They arrive twice a week (Tuesday and Friday), so this section was last updated Fri 13 Nov; the rest of the page updates every night. Stats, not tips: nothing here is a reason to bet.
How forecasts are recorded
Every night we publish forecasts for the next two weeks from that night's league tables. When a game finishes, it's scored against the forecast made from the data published the night before it was played, so before kick-off; results are never used to change a forecast. The record starts on Sat 10 Oct: the forecasts ran every night from then, a few days before the public launch on Tue 13 Oct. Postponed and abandoned games are left out.
Latest results
Results by day: yesterday’s games, forecast v result →