What we measure, and how we check it

• The dataMatch-by-match performance data from data providers (shots, passes, duels, defensive actions, minutes) and market values from Transfermarkt. Every page shows the date of its data.
• The LevelA number from 0 to 100: how many of every 100 players in the database he outperforms, measured match by match against his position in his own league. It doesn’t compare leagues: an 80 in a second division isn’t an 80 in a first.
• Does the Level hold?Today’s Level resembles next season’s: rank correlation 0.41 over 66,826 season pairs, against 0.37 for the raw match rating. It measures performance, not price, and barely works for goalkeepers (0.12).
• The expected price in 12 monthsA model trained on earlier seasons estimates what a player will be worth on Transfermarkt in a year, from his performance, age, minutes, league, position and current price. It is the only future figure on the player page.
• How it is testedBackwards, without peeking: every July from 2019 to 2025, a model trained only on earlier seasons gave a 12-month price, and we compared it with what happened: 36,543 cases from 17,449 players. Always against two simple rules: the price stays put, and every value goes down.
• Contracts and agenciesFrom public sources, and they can lag: contract dates and agencies are not confirmed with the club or the agency. “No agency” means none is listed. 2,536 agencies, with 13,653 players linked to one.
• What we don’t doWe don’t publish the formulas. We don’t claim it works in every league, age or price range: the Accuracy page shows where it is weaker.

All the accuracy figures, including where it is weaker →