From the original record to an understandable match view. Know where the data comes from, what is calculated and what is not yet available.
Football match data: API-Football / API-Sports. Public football views use provider-attributed fixtures and local historical records. The provider connection has been verified; access to current seasons and extended history depends on the account plan.
Awaiting Model Calibration
Research models are not enabled for production. No probabilities, scores or confidence values have been generated for this view.
AI language model does not invent match probabilities. Probabilities are produced by statistical models. The language-model explanation stage is not enabled. Research backtest scores are not production performance.
Stages 3–5 are a research pipeline, not a released prediction service.
Current provider-plan limitations restrict recent seasons and last-match queries. The local database is not a complete global history. Every team/league view labels its sample dates and season; missing records are not filled with generated data.
Matches requests only the selected date. Provider responses are cached: today's schedule for 10 minutes, future dates for one hour, historical dates for up to 30 days. Team, league and insight pages read the existing database and cache without triggering a full sync. Live means a provider-reported match status, not a real-time streaming guarantee. No automatic historical backfill is running.
Engineering validation on one league cannot establish general performance. Class imbalance, time separation, missing coverage and calibration must be assessed before release. Model development includes Log Loss, Brier Score and calibration/ECE, not accuracy alone. A three-hour post-kickoff safety window excludes potentially unfinished games from historical comparisons.
We distinguish provider-plan limitations, unsynchronized history, unreported records, pre-kickoff statistics, calibration, insufficient verified data and competition-specific coverage. An absent injury record does not prove all players are fit. A data-group count is not a confidence score. Results-based league tables may exclude official deductions and competition tie-breaks.