Football Matter Data & Analysis Methodology
A score is an observation; a rating is an interpretation. This guide explains the difference, the inputs our code uses and the limits of the available data.
Live scores, fixtures and final results
Our backend integrates API-Football (API-Sports) and stores fixtures, teams, events and statistics in PostgreSQL. Live website components typically request updates every 30 seconds; that is a display polling interval, not a promise of provider delivery or end-to-end freshness. Background dispatch and reconciliation refresh stored data when configured. Interrupted, postponed and finished matches require different handling. Provider corrections can change a previously displayed result.
Reading football statistics
Possession describes a share of the ball, not the quality of chances created. Shots include attempts that do not test the goalkeeper; shots on target are a narrower provider-defined category. Cards and event timings may be corrected. Lineups are provider records, not confirmation of future availability. Compare like competitions and seasons, and distinguish a missing value from zero. Our xG display uses expected_goals data when supplied; we do not create xG from a scoreline.
League tables and team form
Tables use provider standings where available, or derived tables rebuilt from recorded completed fixtures for configured competitions. Derived ordering by points, goal difference and goals scored approximates official rules and may not capture deductions or every tiebreak. Points and rank depend on competition rules, deductions and group format. Recent form represents wins, draws and losses from up to five completed fixtures in the selected scope. Five results are a small sample: opponent strength, home/away balance and competition mix matter. A form sequence is not a prediction. Empty pre-season standings may cause the upcoming-match rating to use the latest season with played matches.
Match previews and model estimates
Preview inputs can include stored recent results, table position, head-to-head records and provider predictions. Squad availability and tactical observations require separate supporting evidence. Our live probability display adjusts a pre-match estimate using score difference, red cards and elapsed time. It is a heuristic, not a calibrated guarantee, an official forecast or betting advice. Missing evidence must remain missing rather than being filled with plausible details.
Player data and ratings
Player season and match ratings come from provider records where available; averages must be interpreted with their appearances, season and competition scope. They are not our own observed scouting grades. Legacy mock player rankings have been retired. When verified ranking inputs are unavailable, those endpoints return no ranked entries. Missing data does not imply a poor performance.
Upcoming-match interest rating
The MatchRating field is a deterministic selection aid, not an AI prediction of excitement. The original formula adds up to 40 competition-tier points and up to 60 points from the average relative table position of the teams. An optional v2 uses configured competition weights plus up to 45 table-strength points and neutral treatment for group formats. Missing usable standings receive a neutral strength fallback. Ratings omit future lineups and cannot measure how entertaining a match will actually be.
Event balance and player trends
The match event-balance graph starts at 50/50, shifts 20 points for a recorded goal and penalises a recorded card by 8 points, or 14 for a red card, with display bounds. It is an event-based heuristic, not measured territory, pressure or tracking data. Unknown-team events do not invent a direction. The pitch animation uses illustrative movement or sample tracking replay, not tracked positions for the displayed fixture. We do not publish invented percentage changes in player performance when verified time-window statistics are unavailable.
AI football insights
Ask AI first looks for cached or precomputed responses, then database tools and query routes, with Groq and Gemini analysis fallbacks when needed. Generated explanation may overstate what the data supports. Live hot takes are automated commentary, not human-reviewed match reports. Check factual claims against primary records. A model output, cached answer or numerical probability is not independent verification.
News, editorial references and corrections
Our news feed includes BBC football RSS summaries with originating links; those summaries are distinct from original editorial analysis. Blog and externally managed articles have separate publication systems. Corrections in the stored data become visible on subsequent fetches, subject to the caching of each feature. Editorial publication checks exclude disabled articles from public routes. Send a URL and evidence to editorial@footballmatter.com when you find a discrepancy.