How CupChances Works

The simulation

For every active competition, our engine plays out the rest of the season up to 1,000,000 times. Each simulated season uses our match-prediction model to decide a result for every remaining fixture, applies the competition's own tiebreaker rules, and records where each team finished.

A team's "chances" — to win the title, get promoted, qualify for Europe, or get relegated — is simply the share of those million simulated seasons where that outcome happened. It's a genuine probability distribution over how the rest of the season could realistically unfold, not an editorial guess.

Predicting each match

Before we can simulate a season, we need a realistic prediction for each individual match. We fit a Poisson-based expected-goals model per team, using their actual scoring and conceding record — and, where a team's own sample size is thin, we pool historical results across every competition we track, not just the one being calculated, so a new promotion or a small league still gets a grounded prediction rather than a coin flip.

For knockout competitions, seeding and round-by-round advancement probabilities come from the same underlying match model, adapted to each competition's actual bracket format.

Team strength ratings

Alongside match-level predictions, we maintain a PageRank-style strength rating per country: a team's rating accounts for the strength of the opponents it has played, not just its results in isolation — beating a strong side counts for more than beating a weak one. This feeds into cross-competition predictions where two teams from different leagues meet.

How often this updates

During live matches, scores are checked roughly every 10 minutes. A goal, and every other outcome-relevant event, triggers the full simulation to run again for that competition — the probabilities you see are typically only a few minutes behind the actual match.

Data sources & limits

Match data and team information come from public sports data providers. We correct data errors as quickly as we can once they're identified, but a simulation is only as good as what it's fed — treat these numbers as a rigorous estimate, not a certainty.