TRANSPARENT BY DESIGN
How Ressq football predictions are produced
Ressq combines versioned statistical models with current football context. The product separates a probability, an explanation and an optional 18+ price comparison so users can see what each claim means.
01
Inputs and point-in-time evidence
Ressq uses football results and events, team and player performance, competition context, venue, current squads, availability, likely or confirmed lineups and price data only where the product explicitly compares a model estimate with a market. The system is designed to use evidence available at the prediction timestamp.
02
Outcome probabilities
Match models estimate a distribution across outcomes rather than one certain result. Home/Draw/Away figures should be read together and are normalised as a probability distribution.
03
World Power
World Power is a comparable team-strength signal adjusted for opponent and competition context. It helps describe relative strength but does not determine a match by itself.
04
Player and lineup context
Predicted lineups require a complete, unique eligible XI for both teams. Players marked unavailable are excluded. Confirmed lineups supersede predictions, and incomplete evidence fails closed instead of creating a false squad.
05
Simulations
Versioned Monte Carlo simulations sample many plausible match or season paths. Results are distributions with expected values and uncertainty, not promises of an exact score, rank or player total.
06
Language generation
Governed language generation explains model values and supported fixture evidence. It must preserve the numeric output, avoid guarantee language and fall back to deterministic copy when grounding or provider checks fail.
07
Versioning and pre-kickoff changes
Model versions, data cutoffs and timestamps identify how an output was produced. Predictions may change before kickoff when meaningful evidence changes; the final pre-kickoff snapshot is the evaluation record.
08
Limitations
Football is noisy. Red cards, injuries, tactical choices, officiating, weather and low-frequency events can defeat a well-calibrated estimate. Missing data, small samples and changing competitions can also reduce confidence.
Evaluation
Ressq evaluates completed predictions over defined samples. Headline outcome rates are paired with sample size; probability models also use calibration and Brier-style scoring where appropriate.
Open prediction performance →