Methodology

Values are model estimates on an uncalibrated scale. They are not predictions of a transfer fee and have not been validated against observed fees.

A transfer valuation only means something measured against observed transfer fees, and we hold none. The base value below is a stated anchor rather than a fitted parameter, and every factor is a multiplier on it. The relative ordering of players is defensible, because it comes from the factors. The absolute euro figure is not, until the following exist:

• —Observed transfer fees for players in the covered leagues
• —Contract expiry dates, which dominate fees in the final 18 months
• —A held-out backtest of estimate against realised fee

The base value is what a 50th-percentile, 27-year-old, fully-available midfielder in the reference league is taken to be worth. Every valuation on the site is that number multiplied by the factors below.

Each player’s weekly output is weighted by position, discounted by minutes as a credibility multiplier, and compared against an exponentially weighted average of his own previous weeks — so the index moves on surprise rather than on output. That index is converted to a percentile within his position group, and the percentile drives the valuation.

The map from percentile to multiplier is exponential, not linear. The gap between the 50th and 60th percentile is small money; the gap between the 95th and the 99th is most of a transfer record. A linear map would make every squad player look like a third of a star, which is the most common way a naive valuation model embarrasses itself.

Interpolated linearly between these points; ages outside the table clamp rather than extrapolate. The premium in the early twenties is potential — a buyer is paying for seasons not yet played — which is why it sits above the peak-age multiplier rather than below it. A player whose date of birth we do not hold gets a neutral multiplier and the driver is marked unavailable on his page.

A competition with no strength rating gets a neutral multiplier and is marked unavailable rather than guessed at. Availability discounts the valuation directly — a player who cannot get on the pitch is worth less than one who can, whatever his rate statistics say — and is floored rather than allowed to reach zero, so one factor cannot annihilate a valuation.

Confidence is built from four things that genuinely vary between players: minutes played, matchweeks observed, how many peers the percentile was taken against, and how far into the tail of the distribution the estimate sits. The fit is densest in the middle, so the model is least certain exactly where the headline names are.

Confidence is then capped at 72%. Without contract length or observed fees, no valuation here deserves to read higher, however many minutes a player has played. A model that can print 95% on data this thin is lying about something more important than the percentage.

The range widens as confidence falls, and never collapses to a point. It is the honest shape of the answer, not a decoration on it.

270 minutes

Below this a per-90 rate is shown but never ranked, and the player is excluded from other players’ percentile populations too. A player with twenty minutes and one goal rates at 4.5 goals per 90; that is arithmetic, not a finding.

900 minutes

The floor for appearing on the market boards. An absent player’s index drifts down while his performance level does not, so injury sorts to the top of a value ranking. Availability is a factor in the valuation for the same reason.

Why these move

Both bars are a share of the minutes actually available so far, capped at a full season’s 270 and 900 and floored at one complete match. Fixed thresholds would rank nobody in August and a share alone would rank everybody, so they ramp: the floor reaches 270 by matchweek 12, the credibility mark reaches 900 by matchweek 17, and neither moves again. If you compared two visits a month apart and the numbers differed, that is why.

• Goals
• Assists
• Shots
• Shots on target
• Key passes
• Expected goals
• Expected assists
• Tackles
• Interceptions
• Clearances
• Saves
• Goals conceded
• Yellow cards
• Red cards
• Minutes
• Clean sheets
• Age
• Position
• Competition

Stating this plainly matters more than any single number on the site. The feed carries no passes attempted or completed, no possession, no duels or aerials, no carries or progressive actions, no pressures, no touch locations, no formations and no substitution timing. It carries no injuries, no contracts, no wages and no transfer fees.

Two consequences follow, and both are visible in the product rather than only described here. Player DNA has no ball-progression or pressing axis, because there is nothing to compute one from — an axis built on a proxy would be the most confidently wrong thing on the page, since a radar invites its arms to be compared directly. And player pages carry no comparable-transfers or contract panel, because both need data we do not hold.

A fee is the outcome of a negotiation between two clubs and a player, and is dominated by things absent from any performance model: contract length, release clauses, wage demands, agent fees, the selling club’s financial position, competition between buyers, and the timing of the window. A player with twelve months left is worth a fraction of the same player with four years left, and nothing in his statistics would tell you that.

What is published here is one question answered consistently across every player in the league: given what he has done, his age, his position and his division, where does he sit relative to everyone else? That is a useful question. It is not the same question as “what will he cost”.