How PlayMaker AI calculates predictions
1. Data sources
PlayMaker AI uses provider-backed fixture, score, standings, and market data from football-data.org, TheSportsDB, The Odds API, and optional API-Football coverage. OpenRouter and Ollama may explain model outputs, but they do not replace the statistical prediction engine.
2. Prediction inputs
The model reads fixture metadata, team names, scores, match status, standings, recent completed results, goal difference, goals scored and conceded, expected goals, bookmaker odds, market freshness, and provider data quality. Missing or stale inputs reduce model, data, market, and live confidence separately before the final confidence is displayed.
3. Probability model
Football probabilities are built from team-strength ratings, historical scoring profiles, Elo-style adjustments, and a Poisson goal distribution. Live matches add current score and match-state weighting so projected final scores cannot contradict the verified live score. Calibration then shrinks extreme probabilities when sample size or data quality is limited.
4. Bookmaker comparison
Bookmaker odds are converted into implied probabilities and adjusted for overround where possible. PlayMaker AI keeps model probability separate from bookmaker probability, then compares them afterward. Value is only shown when odds are fresh and the model clears confidence, sample-depth, and data-quality guardrails. If odds are absent, the market state explains whether the fixture is unmapped, not open, live-market unavailable, stale, or disabled instead of pretending value exists.
5. Exact scorelines
Exact scorelines are estimated from the same expected-goals and Poisson score distribution used for 1X2, totals, and both-teams-to-score probabilities. They are shown only when confidence, reliability, data-quality, volatility, and live-state gates clear. Otherwise the app displays why the exact scoreline is held.
6. Confidence, reliability, trust, and risk
Probability answers what outcome the model thinks is likely. Confidence measures evidence strength. Reliability reflects how proven the model is in similar situations. Trust determines whether a pick should be displayed as elite, strong, watchlist, caution, data-limited, live watchlist, or pass. Risk increases when volatility, missing data, stale markets, live uncertainty, or model disagreement are high.
7. Why predictions may be held
Predictions can be held when provider data is missing, teams cannot be mapped confidently, historical samples are too thin, market odds are stale or incomplete, confidence is low, or volatility is too high. When a match is live, verified score and match state can produce a live-context prediction even if historical depth is limited.
8. Responsible use
PlayMaker AI predictions are informational probabilities only. They are not betting, financial, legal, or professional advice, and they do not guarantee outcomes. Odds, scores, injuries, lineups, and probabilities can change quickly.