What we publish
The core output is a three-way probability for a home win, draw and away win. Each official forecast is designed to carry a publication time, the kick-off known at publication, a model version and a stable match page. Once a final score is available, the same record can be settled without rewriting the original probability.
Locked probabilities
Home, draw and away values with their model and timestamp.
Visible outcomes
Final score, hit or miss and a probability-quality score.
Calibration evidence
Whether forecasts labelled 60% actually occur near 60%.
What “AI” means here
The forecast is produced by a statistical model trained and tested on historical football data. A separate grounded assistant, Kito, always offers a deterministic explanation of an already-published record; optional AI response planning is enabled only when its server-side model is configured. That model may choose only a full or compact view of server-authored evidence. Neither mode can alter probabilities, invent team news, recommend a stake or turn uncertainty into a guarantee.
This separation matters: model output is a versioned data product; conversational explanation is only an interface to that record.
Launch scope and current limits
The configured launch scope covers the Premier League, La Liga, Serie A, Nigeria Premier Football League and Kenya's FKF Premier League. Coverage still depends on a successful data feed, sufficient history and a model that passes its published acceptance gates. When those conditions are not met, the site shows a demo or unavailable state instead of presenting sample rows as official performance.
Football Proof AI does not accept bets, hold customer funds or guarantee profit. The probabilities are informational and should be judged as forecasts, not instructions.