FPA Score Methodology

How the score is built

The score is produced in three stages. We publish the structure openly, because a number nobody can interrogate is not worth trusting. What we do not publish is the tuning, and we explain why immediately afterwards.

1. 1Per-90 normalisation

   Every raw counting statistic — goals, shots on target, key passes, completed passes, successful dribbles, tackles, interceptions, duels won — is converted to a per-90-minute rate. This is the first and most consequential correction. A substitute who scores four goals in 500 minutes and a starter who scores nine in 2,400 minutes are not separated by five goals; they are separated by a rate that favours the substitute. Raw season totals reward availability and punish rotation, which is precisely the wrong bias when the object of the exercise is to find players whose minutes have not yet caught up with their ability. We require a minimum sample before a player is rated at all, because a per-90 rate computed from a handful of appearances is noise wearing the costume of signal.

2. 2Percentile conversion within a comparison group

   A per-90 rate means little in isolation. Two tackles per 90 is unremarkable for a holding midfielder and extraordinary for a winger. Each rate is therefore converted into a percentile against a comparison group of players in the same position, so that every metric ends up expressed in the same unit: how unusual is this, for this kind of player? Positional comparison also stops the model from collapsing into a goals-and-assists ranking, which would leave defenders and deep midfielders permanently invisible no matter how well they played.

3. 3Position-weighted aggregation

   The six attribute percentiles — scoring, creativity, passing, dribbling, defending, duels — are then combined into a single figure alongside two contextual inputs: how much the player actually plays, and how young they are. The weighting is not uniform, and it is not the same for every position, because what matters in a centre-forward’s profile is not what matters in a full-back’s. The strongest parts of a player’s profile carry more weight than the weakest, on the reasoning that recruitment is usually a bet on a specific strength rather than on the absence of any weakness.

The result of those three stages is then placed on the 0-100 percentile scale described above, and that is the number shown everywhere on this site.

What we do not publish

We do not publish the individual metric weights, and we do not publish the per-position weighting sets. The reason is straightforward, and we would rather state it plainly than dress it up in language about proprietary methods: the weighting is the part of the model that took real work to arrive at, and it is also the part that would be trivial to copy. The three-stage structure described above is ordinary practice in football analytics and we lose nothing by setting it out. The tuning is not ordinary practice, and we keep it.

The consequence for a reader is real and worth naming: the FPA Score cannot be independently reproduced from this page. We think the honest response to that is not to pretend otherwise, but to accept being judged on results instead — which is what the validation section below is for.

What the score cannot see

The FPA Score is built entirely from match event data. Everything that does not appear in an event feed is invisible to it, and several of those invisible things are among the most decisive factors in whether a transfer succeeds or fails.

Tactical adaptability

A player’s numbers are produced inside one particular system, with one particular set of instructions and one particular set of teammates. The score cannot tell you whether a midfielder whose passing percentile was built at a possession-dominant club will still function in a side that spends most of the match without the ball.

Mentality and temperament

Nothing in an event feed measures how a player responds to being dropped, to a hostile away crowd, or to six months without a start. Recruitment departments treat this as a first-order variable, and rightly so. The score does not address it at all.

Language and cultural adaptation

A move from Japan to a European league is a relocation before it is a transfer. Settling in, language, distance from family, and the ordinary practicalities of a new country have ended careers that the numbers said were about to begin.

Injury history and durability

The score reads minutes played, but it cannot read why the missing minutes are missing. A favourable per-90 rate accumulated across a fragmented season may belong to a rotation option or to a recurring injury risk, and the score cannot tell the two apart.

Uneven data coverage between divisions

Our current data source supplies full defensive detail for J1 but not for J2 and J3, where tackles, interceptions and duels come back empty. Players in those divisions therefore score lower on the defensive attributes than their play warrants, and their overall FPA Scores sit systematically below J1’s. We would rather flag this openly than let a reader assume the gap reflects ability. Until the coverage evens out, compare within a division rather than across them.

The FPA Score is a screening instrument, and it is worth being blunt about the size of that claim. Its purpose is to reduce a database of two thousand players to a shortlist worth someone’s attention, and nothing beyond that. It is not a valuation, not a recommendation, and not a substitute for watching the player. Every name it surfaces is the beginning of the work rather than the conclusion of it.

Data sources

Player, squad and match statistics come from API-Football, covering J1, J2 and J3. The current dataset is the 2024 season. Detailed per-metric enrichment is applied on an ongoing basis, which means the comparison groups sharpen as coverage of the database improves.

Where our editorial scouting reports cite finishing quality, chance creation or defensive volume beyond what the API supplies, those figures are cross-referenced against Football LAB. A number that cannot be traced back to one of these two sources does not go into a published report.

How the score is validated

A scoring model that is never checked against reality is an opinion with decimal places attached. Every scouting verdict we publish carries a date, and we track what subsequently happened to the player: the moves that materialised, the ones that never did, and the calls that turned out to be wrong. That record is the only meaningful test of whether the FPA Score measures anything real, and we keep it public for exactly that reason.