A forecast that reruns itself. Every visit pulls the latest scores, re-rates every team on what just happened, re-simulates the bracket, and grades its own past picks. Nobody touches it between rounds.
Twenty thousand simulations of the games still to play, recomputed on our server each round.
Walk-forward scoring. For every knockout game already played, we take the pick the model would have made beforehand and check it against the result. No hindsight, recomputed on every load.
Each surviving team's chance of lifting the trophy. Ratings shift after every match, so these move with form. Knocked-out sides fall to zero on their own.
A green highlight marks a confirmed result. A live game shows each side's pre-game pick next to a live win chance that shifts with the score and the clock, refreshed every minute.
The games still to come, with a live countdown to kickoff, the model's pick, and its most likely scoreline.
How far the model expects each team to go.
| Team | Win QF | Reach Final | Title |
|---|
Tournament top scorers, tallied live from the feed. The players most likely to swing the games still to come.
Every completed knockout result, grouped by round, most recent round open by default. This same history feeds the model's ratings, so it shapes the forecast.
How the model's knockout picks have held up, round by round. The line tracks cumulative accuracy through each stage and extends as the quarterfinals and beyond are played.
Open any surviving team to see what is driving its number: the rating breakdown, path odds, next match, scorer depth, recent form with the exact rating move each result caused, and how much a key absence would cost. Every figure is computed on our server; the page only shows the result.
Every team starts from a pre-tournament rating. The page then walks through every completed match in order and adjusts both teams after each result, weighting the move by the margin of victory. A team on a run climbs, a team stumbling slides. Those live ratings drive a win-probability curve built for how random one-off soccer really is, and the remaining bracket runs twenty thousand times on our server, so the method stays private.
Accuracy is measured walk-forward. Before a game is scored, the model only knows what happened before it, so the track record is what the model actually would have said, not a number picked with hindsight. Scores come from a public sports feed, refreshed on every load. Confirmed winners lock in, eliminated teams drop to zero, and everything rebuilds around whoever is left. Nobody edits this by hand between rounds.