How it works

Every player-season is scored on eight things — finishing, creation, progression, duels and defending, aerial play, goalkeeping, discipline and errors — measured against other players in the same position in the same season. That is what lets a 2009 midfielder be compared with a 2023 one: both are judged against their own contemporaries rather than against a raw number that drifts with the era.

Your eleven are combined into a team rating, adjusted for the shape you chose and for how well the players complement each other, and the two sides are played out with a Dixon-Coles model — the standard way of turning team strength into a scoreline.

It is the share of decisive matches you would win: draws are set aside. That is not a presentational choice. Roughly a quarter of football matches end level, and those draws come off both sides, so on the unconditional scale two identical teams each win about 36% and no XI that has ever existed reaches 55%. Conditioning on a result puts a perfectly matched side at exactly 50% — which is what makes the number mean anything.

A player's card shows the minutes he actually played that season, and the model takes them seriously. Someone with eight outstanding matches is cheap, and he is cheap for a reason: his profile is discounted in proportion to how little of the season it rests on. A full campaign counts fully, around a thousand minutes counts for roughly 60%, and below 270 minutes a player is not draftable at all.

That makes the bargain wonderkid a real trap rather than a free win — one worth walking into with your eyes open.

You can sign an eleventh outfielder when the slot he belongs in is already full. He will be placed wherever costs the side least, and every one of his ratings is reduced in proportion to how far the move is: a defender shifting into midfield gives up a little, a midfielder in goal gives up almost everything.

The reshuffle happens automatically and can move somebody you signed earlier, so a fifth centre-back does not simply displace himself.

A squad list often has more names than you can sign. Two reasons, both about honesty rather than tidiness. A player who moved clubs mid-season has statistics that belong to two clubs at once — the feed reports season totals, not per-club ones — so attributing them to either would be inventing a number. And a player who never got on the pitch has no statistics to judge at all.

Those genuine mid-season movers are shown in the squad, greyed out, with both of their clubs named — Jorginho appears under Arsenal 2022/23 and Chelsea, and can be signed for neither. Players who never got on the pitch are left out entirely.

Detailed statistics do not exist for 2003/04. Arsenal's attacking profile is reconstructed statistically from what they actually did; their defensive profile is not — the reconstruction could not be validated there, so it is set to that of a typical side instead. The whole is then checked against sides that really did take 90 points.

Detailed statistics did not exist in 2003/04. Arsenal's attacking profile is statistically reconstructed from their actual results, and their defensive profile is set to that of a typical side, because the reconstruction could not be validated there. The whole is then checked to be as strong as the sides that really did take 90 points.

Defending, mostly. Tackles, interceptions and clearances measure how much defending a side did, and a good defence is precisely one that has to do less — measured across 400 real seasons, those categories barely predict goals conceded at all. The defensive rating therefore leans on the player's own numbers and on the record of the club he played for, and it is the least trustworthy half of the model.

It also cannot see tactics, managers, injuries, or whether eleven strangers would actually get along.

Statistics from the Premier League. Market values from Transfermarkt, indexed to 2026 euros so a 2009 price and a 2024 price mean the same thing. Prices before 2006/07 do not exist, which is why those seasons are playable only in sandbox and carry no cost.

The scoreline model is fitted on 7 380 real Premier League matches.