Track record and method

We publish how every forecast is made, and we grade each one honestly once the match is settled. No cherry-picking, no hidden losses, no guarantees. Statistical estimates for analysis and entertainment only.

Live accuracy

Recent graded forecasts

How the forecasts are made

Each forecast blends several well-established football models, then reports the result as probabilities with a confidence level, never as a certainty:

• Poisson goals model - estimates how many goals each side is likely to score.
• Dixon-Coles adjustment - corrects the Poisson model for low-scoring scorelines and recent form, which raw Poisson handles poorly.
• Expected goals (xG) and form blend - weights recent underlying performance, not just results, so a team riding luck is not overrated.
• Ensemble and confidence - the models are combined and we publish a confidence band for every pick, because the uncertainty in football is large and real.

How we grade ourselves

When a match finishes, we record the actual result against the forecast and update the numbers above. We use two honest measures: the outcome hit rate (how often our most likely outcome was correct) and the Brier score (how well-calibrated our probabilities were, where lower is better). Past performance is not a reliable indicator of future performance, and any single weekend can go against the model.