Match Analytics AI
MatchesTeamsLeaguesInsightsProAccount
Research workspace

Match Analytics AI

Football Analytics & Match Intelligence
Verified records. Transparent methods. Clear limits.

Data & MethodologyAboutContactPricingTermsPrivacyRefund PolicyResponsible Use
MatchesTeamsLeaguesInsightsProAccount
Transparency

Data & Methodology

From the original record to an understandable match view. Know where the data comes from, what is calculated and what is not yet available.

Data Sources

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.

Stored fixtures
431
Stored teams
112
Target leagues
20

Modeling

EloPoissonDixon-ColesEnsembleCalibration

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.

From data to explanation

  1. 01Raw Data
  2. 02Feature Engineering
  3. 03Statistical Models
  4. 04Probabilities / Ratings
  5. 05AI Explanation
  6. 06User Interface

Stages 3–5 are a research pipeline, not a released prediction service.

Data Coverage

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.

Update Frequency

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.

Model Limitations

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.

Model History →Model Performance →

Missing Data Handling

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.