Forecast, not a tip. Probabilities are model outputs and carry uncertainty. Do not risk money on them.
Methodology
A Poisson goals model, blended and calibrated — and how to read what it prints.
Each forecast comes from a Maher-style Poisson goals model. It estimates each side's attacking and defensive strength, and home advantage, from results weighted towards the recent ones, blended with expected goals where a competition has them.
Its output is calibrated with isotonic regression, fitted separately for each kind of evidence a forecast rests on, so a “60%” happens close to 60% of the time. Calibration quality is tracked with the Brier score and expected calibration error (see the weekly recap).
That applies where we have enough graded matches to fit and check a curve. Competitions without that history — cross-league ties, most recently — still receive a forecast from the same model, and it is marked as one rather than described as calibrated. The reliability page lists exactly which markets have been measured and which have not.
Every percentage is a model output, not a certainty. The largest of the three 1X2 numbers is the model's most likely outcome, and an upset is expected some of the time, not a failure of the model.
The Top Forecast is the one market the model selected for a match before kickoff. It is settled on its own market at 90 minutes, separately from the 1X2 numbers.
Forecasts cover the competitions listed on the Leagues page. They are for information only, and are refreshed as results arrive, until kickoff.