Football analysis methodology
Understand historical inputs, probabilities, evidence quality and the limits of football analysis.
What does the platform analyse?
Historical results support pre-match probabilities, which are compared with verified prices. Mathematical value and readiness to act are assessed separately.
How is future information excluded?
Results must be available before both the prediction cutoff and kickoff. Date-only results use a conservative two-day availability margin. Each team needs at least five historical matches for an estimate; insufficient evidence blocks recommendations.
Where do form and head-to-head records come from?
Form uses the latest five available matches in the recorded league, newest first. Head-to-head records cover the year before kickoff and must pass the publication cutoff. Missing records do not mean the teams have never met.
Are probability and confidence the same?
No. Probability is a model estimate, not a guarantee. Evidence quality describes completeness and verification, rather than the chance of winning.
Why can a fixture have no recommendation?
Prices may be missing or stale, history may be insufficient, or official lineups may be unverified. The platform must not invent prices or statistics to produce a recommendation.
What is AI used for?
When enabled by an operator, AI explains context from verified evidence. It cannot change the frozen model equations or probabilities. Opening a public page does not automatically trigger a paid AI request.
How is an improvement demonstrated?
Evaluation must use data excluded from model development and measure probability error, calibration, sample size, and performance by league and time period. Historical results and software tests alone do not establish future predictive accuracy.
Can research silently enter production?
No. Research, challenger and shadow outputs are separated from production decisions. Changing the frozen production model requires evaluation evidence and approval.