Football models & analysis
Prediction Analytics
How the model spreads its calls, and how confidently it makes them.
- Predictions
- 245,347
+2 edge over the baseline · 154,597 graded predictions
What the model predicts, against what happens
Left: the model’s selected outcomes. Right: the actual outcomes for the same evaluated matches. This compares selection frequencies; it does not measure probability calibration.
Model predicts
Actually happened
Confidence spread
How many calls the model makes at each confidence level. Whether those levels are actually right is on the track record.
Risk profile
Every figure on this page describes the model running in production today. Predictions from retired models are excluded.
This page describes how the model behaves, not how often it is right. For that, see the verified track record.
View our track recordHow the models work
Explore the methodology behind our match predictions.
Methodology →Historical performance
Review past predictions and how they performed.
View our track record → Prediction Analytics →Model performance analytics
How the model behaves rather than what it predicts: accuracy over time, calibration by confidence band, and performance split by competition and market. Calibration is the measure that matters most, because a model that says seventy percent should be right about seventy percent of the time.
Figures update as results are processed, so a matchday appears here within hours of the final whistle. Competitions with small samples are noisier than the headline number suggests, which is why the sample size is shown next to each figure.