Our predictions draw on dozens of data sources — from 49,000 historical international match results to individual player statistics, manager profiles, and team playing styles. This page is the complete inventory: every dataset we use, what it contributes, and how rigorously we've tested its impact.
What shapes a match
What data goes into the model
Resumen + muestra · documento completo de 3,281 palabras
Resumen
Muestra
A note on importance claims
Most data inputs in this project have not been rigorously feature-importance-ranked. The team-strength models (Dixon-Coles, Hierarchical Poisson, Elo) have been compared head-to-head on a 940-match holdout window with Brier, log-loss, and ECE reported in documentation/methodology.md. Beyond that core comparison, ablations on individual context features have not been run. Where backtest evidence exists, this page cites it. Where it does not, the page says so. No importance numbers on this page are fabricated. A reader who notices an invented "Brier improvement of N points" would correctly stop trusting the rest of the document; if the number isn't here, it's because the ablation hasn't been done.
Documento completo
Pro Pass
¿Quieres el documento completo?
What data goes into the model tiene 3,281 palabras. El Pass desbloquea este documento y todas las notas de investigación completas, más las probabilidades por partido, la comparación de cuatro modelos y el análisis táctico por partido.
24h self-service refund·No subscription, no auto-renewal·Access through 31 Dec 2026. See refund policy.