Teams
Model strength on one axis, social heat on the otherFixtures
FPI's call against the result, match by matchThe engine,
in the open.
World-leading prediction is not a black box. Every number FPI publishes is traceable to real, free, public data, run through a calibrated statistical engine, and graded against the sharpest tests in the sport. Here is exactly how.
Every fact, from a free public source
Every feed is free and public. The statistical core is fit on match evidence alone. The social read sits on top as a small, capped adjustment that had to earn its place out of sample, and the market is held aside as the benchmark every call is graded against.
How good the predictions are
Graded prospectively against the Pinnacle closing line, the toughest fair test in sports forecasting. This is the live, running scoreboard, updated on every result.
One clean pipeline that runs itself
Each source has its own worker on a timer, event-gated so it writes only when there is genuinely new data. A single event-driven loop chains the whole engine the moment a match ends, and a health watchdog flags any stale feed within minutes.
How a single probability is built
Fit team strength
A Dixon-Coles bivariate-Poisson model is fit on the full history of internationals, each match time-weighted by a power-law decay so recent form counts for more without discarding the past. A prior from Elo and shrunk expected goals anchors thin-data teams.
Turn strength into scorelines
Each team’s attack meets the other’s defence to set two scoring rates; a low-score correction captures that draws happen more than pure chance predicts. Home advantage counts only when a genuine host plays at home.
Read off the odds
Summing the scoreline grid gives home / draw / away for a group match, and a single who-advances number for a knockout tie, with extra time and penalties folded in exactly.
Carry the doubt
The whole fit is repeated across hundreds of bootstrap re-samples, so every published number is a posterior average with an honest confidence range, never a single false-precise point.
The tournament, computed exactly
From the live bracket the engine computes each team’s exact chance of reaching every stage by an exact dynamic program over the draw, evaluated on every posterior sample. Zero Monte-Carlo noise in the headline; a 60,000-tournament fused simulation runs alongside as a cross-check, and the two must agree before anything publishes.
The lines we will not cross
Rigorous statistics, zero guesswork
Every probability comes from a calibrated, peer-grade model with years of anti-overfitting discipline behind it. No hunches, no house edge, no thumb on the scale.
We listen to the whole world
Millions of real conversations across every market, listened to and weighted. On this build the social read nudges match odds inside a tested, capped band, and never touches the title race or the bracket.
Graded against the sharpest line, in the open
The Pinnacle closing line is the toughest fair test in football forecasting, so it is the one we publish ourselves against, match after match. The scoreboard above is the running record, not a highlight reel, and it reads the same whether the gap flatters us or not.
Every number, provably real
No figure is ever invented. Each one traces to a real, dated public source, and when two sources disagree we surface the conflict instead of hiding it.
Social intelligence
Fan-perception radar, share of voice, and sentimentAttribute radar
Every team's fan-perception shape, sorted by overallShare of voice
Daily mentions, de-sampled to true scale, top 12 over 30 daysNet sentiment
Positive minus negative share of mentions, all 48