Who goes down, who qualifies, who wins it?
Odds for relegation, continental places and the title, from a simulation of the rest of the season, thousands of times over. Live for the season in progress, and replayed matchweek by matchweek: this season so far, and finished seasons.
These are probabilities from a model, not predictions or promises.
How much to trust these
Each club's chance comes from simulating what is left of the season thousands of times, with the club's live Elo rating deciding the matches. Checked on past seasons, the odds beat a no-skill guess more clearly the further the season has gone: before it starts, a rating says little about relegation, and every outcome sharpens as the table fills.
The probabilities are the model's own, not calibrated: a calibration fitted only on earlier seasons was tried and did not improve them on the seasons it could be tested on (see "Model against reality"), whose tables show where the model is over-confident, for example for continental places. The live figures are plausible, not validated: the formats and the number of places of the last finished season are assumed to repeat, the matches still to be played use a stand-in schedule, and the ratings stop on the date shown for each league.
Replays that are not shown, and why
A replay is only shown when the model can be run on it with no information from after the date being predicted. These seasons are left out for that reason or because the data cannot support them:
- Peru Liga 1 2025: no club Elo rating within 120 days of the state dates (the latam export skips it)
- Chile Primera División 2024: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Ecuador Liga Pro 2025: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Paraguay División Profesional 2024: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Peru Liga 1 2024: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Russia Premier League 2019/2020: round 46: 2019/2020 is the first season of the data, so there is no earlier season to fit on
- Spain Segunda División 2024/2025: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Uruguay Primera División 2024: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- Venezuela Liga FUTVE 2024: round 46: the replay is strictly out of sample (the model is fitted only on seasons that finished before the replayed one), and this league has no earlier season with club Elo ratings to fit on
- China PR CSL 2025: round 47: the places a season is predicted with must be known before it starts (the count of an earlier season with complete outcome labels, or an official allocation), and this league has no earlier season with complete labels
- Egypt Premier League 2024/2025: round 47: the places a season is predicted with must be known before it starts (the count of an earlier season with complete outcome labels, or an official allocation), and this league has no earlier season with complete labels
- Tunisia Ligue 1 2024/2025: round 47: the places a season is predicted with must be known before it starts (the count of an earlier season with complete outcome labels, or an official allocation), and this league has no earlier season with complete labels
Every other finished season of every league in the list is replayed.
What the replay shows
Seasons replayed through the model's own Monte Carlo simulation (the number of simulations per matchweek is shown above the table), exactly as it would run live, using each club's live Elo rating on the day with the replayed season left out of the model's training so its own results cannot leak into its predictions. Matchweek 0 is the model's read before a ball was kicked; for a finished season the last matchweek is the settled table, and the Real result column shows what actually happened.
When an outcome shows an exact 100% or 0%, the real table and the real remaining fixtures guarantee it; a simulated near-certainty is shown as >99% or <1% and never as a bare 100% or 0%.
Where the table settles a place only on goal difference, or where a place went to a cup or European winner (for example an FA Cup winner's Europa League place), the last matchweek can differ from what really happened.
A season in progress appears as the last season of its league, marked "(in progress)": step through it matchweek by matchweek to see how the table and the odds moved. It has no real-result column, because there is no result yet.