FBSA (Football Score Analytics, fbsa.app) is an independent research project in football probability modelling. We build, document and publicly evaluate statistical models that estimate scoreline and match-outcome probabilities for club and international football.
Our club model is a bivariate Poisson goal model with the Dixon-Coles low-score correction, blended with a gradient-boosting estimate of goal expectancy. A separate hybrid ensemble covers international matches. The system continuously benchmarks its probabilities against market-implied odds, and a self-learning loop re-estimates team strength after each round of matches. All dashboards on this site — club analytics, international analytics and the fully simulated paper trading tracker — are driven by the same production model.
Every model change must beat the current baseline on proper scoring rules (Brier, RPS) over a chronological out-of-sample window before it ships — parameters that only improve hit-rate are rejected. We also publish our failures: see our study of 6,478 matches against the closing line in the research notes, where the model lost to the market and we said so in full.
This is a data research site. We do not sell predictions, we do not accept gambling advertising, and nothing here is betting advice. Our audience is football analytics enthusiasts, data scientists, scouts and sports-technology professionals. Match data is derived from publicly available results.
Research questions, corrections and data-quality reports are welcome: contact@fbsa.app — see our Contact page for response standards.
FBSA Research