Track record
How our predictions perform
Statistical football predictions are only worth anything if the record is open. Every prediction we publish is checked against the final score, and the results live here, good and bad. Figures below come from a walk-forward test on 17 leagues: each prediction was made using only the matches that had already been played, exactly as the model works in production.
WaveBets Blend
Our headline prediction. The Dixon-Coles model combined with the bookmakers' prices, published as the site's probabilities.
Best overall accuracy: beat the bookmakers' calibration in all 17 leagues tested.
Dixon-Coles
Our core statistical model. Attack and defence ratings per team, fitted on match history with time decay and a low-score correction.
More accurate than a plain Poisson model in most leagues.
Elo ratings
A single strength number per team, updated after every result. Simple, transparent, and a useful sanity check on the main model.
Tracks form quickly but is less precise on goals markets.
Bookmakers
The combined bookmaker price with the margin removed. Not our prediction, shown so you can judge ours against it.
The benchmark every other model is measured against.
Read the numbers honestly
Our probabilities are more accurate than the bookmakers' in most leagues we tested. That does not mean every pick wins, and it does not mean a profit is guaranteed: after the bookmakers' margin, our test found no pick type that wins reliably enough to promise one. What we publish is the most accurate forecast we can build, shown next to the odds so you can make your own call. Where the models win and where they lose, the record is public.