How the ranking works

• G = 1 if current-season league minutes are 270 or more, otherwise 0 (not ranked).
• M (15%) = 100 x clamp( log10( value_eur / 25,000 ) / 3 , 0 , 1 ). Market context. The league factor is not applied to M.
• T (30%) = 100 x clamp( s / e(age) , 0 , 1 ) x L, where s = league minutes / (club league matches played x 90) and L is the league factor.
• O (45%) = O_raw x sqrt(L). O_raw is the weighted mean of his positional percentiles (see Positions).
• Y (10%) = 100 x max(0, T - 100 x s x L) / 50, and 0 from age 23. 50 is the largest possible gain, for a player aged 18 or younger who plays half the minutes in a league with L = 1.00 (T = 100, base 50). So Y runs from 0 to 100.

Results are rounded to one decimal. Ranking ties break on O, then T, then confidence, then league minutes; market value is never a tie-breaker. Unknown inputs are never guessed.

Each statistic is a rate per 90 minutes. To stop a few good games from creating an unrealistic rating, every rate is pulled toward the average of his position with a 450-minute prior: adjusted rate = (n + prior x 450 / 90) / ((minutes + 450) / 90). His own rate then counts for minutes / (minutes + 450): 40% at 300 minutes, 50% at 450, 67% at 900, 80% at 1,800 and 89% at 3,600. The adjusted rate becomes a percentile among players of the same position group in our data: 90 means roughly the top 10% of the group, 50 the middle, 10 near the bottom.

• Unknown is not zero. When a league does not publish a statistic (for example tackles in El Salvador), its weight is removed and the other weights are re-scaled to add up to 100%. This lowers confidence, never the score by itself.
• Passing contribution = 50% completed passes per 90 (percentile) + 50% pass accuracy (percentile). Accuracy uses a prior of 150 attempts toward the group average.
• Discipline = yellow cards + 3 x red cards + 2 x own goals, per 90, adjusted like any rate; fewer is a higher percentile. A second yellow counts once, as the resulting red, and not also as a yellow.
• Team statistics (clean sheets, goals conceded, wins, table position) are never scored for outfield players. For goalkeepers the clean-sheet rate counts because preventing goals is his job; saves weigh more whenever they exist.
• Goalkeeper fallback: expected goals against is not in our data, so the goalkeeper weights are re-scaled over saves, clean-sheet rate, penalty saves, passing and discipline. A penalty-save figure counts only for a goalkeeper who faced at least one penalty (prior of 3 attempts).
• Winger and centre-forward share one FW group in this version.
• Reference groups (ranked-eligible players with the core inputs, today):
  • FW: 5 players
  • MF: 7 players
  • FB: 3 players
  • CB: 4 players
  • GK: 4 players
  Goals, assists and cards exist in every league, so FULL and BASIC players share one pool for them. Shots on target, tackles and passes are compared only among players whose league publishes them. With so few players, a lone player in a group sits at percentile 50 for such a metric. We use mid-rank percentiles (players below + half the ties, divided by the group size) and show the group size next to every percentile.

1. 1
   Every rule applies to every player in the same situation. No weight, threshold, league factor or positional rule is changed to help or hurt a named player.
2. 2
   Unknown data is never guessed.
3. 3
   Team outcomes are not automatically treated as individual performance.
4. 4
   Players are compared primarily with others in similar positions.
5. 5
   Small samples are adjusted so a few strong games do not create an unrealistic rating.
6. 6
   Missing advanced statistics are missing information, not poor performance.
7. 7
   Age changes how playing time is read, but is not rewarded again in several components. Only T and Y use it.
8. 8
   National-team selection is not a scoring component, because the coach controls it. Height is not scored either; it is only shown on the profile.
9. 9
   Market value is context, not the definition of football quality, and has the second-lowest weight (15%).
10. 10
    The formula is frozen before any formal backtest is evaluated.

• Interceptions, clearances, blocks and aerial duels: centre-back and full-back defending is under-measured. Those weights are dropped, not scored as zero.
• Expected goals and assists, shot location, goalkeeper post-shot expected goals, progressive passes and carries, physical tracking, tactical role.
• Team strength is not added to anyone's score. A future version may use team context to normalise statistics, only when enough data exists.
• Future (v2.3 candidates): expected goals and assists, progressive passes and carries, interceptions, clearances, aerial duels, goalkeeper post-shot expected goals, opponent strength, team-possession normalisation, tactical roles, separate winger and striker models, multi-season ageing curves, larger comparison groups, extra full-back metrics. None is added until a reliable source exists.

• H1 RPIX at time t predicts the next 12 months of change in log10 Transfermarkt market value better than market value alone and better than minutes share alone.
• H2 RPIX at t predicts level-weighted minutes (sum of minutes x league factor) over the next 12 months better than prior level-weighted minutes, market value and raw minutes share.
• H3 Players with higher O at t keep a higher position-adjusted O in the next competitive period (the performance signal repeats rather than being noise).
• H4 Selection gap, descriptive only: for each senior camp, how often lower-RPIX or inactive players are called while higher-RPIX active players at the same position are not. No pass or fail. See the Selection gap page.

The plan, comparisons and statistics are on the Backtest page. It has not been run yet.

Version 2.2 stays provisional until the 2025-10-01 to 2026-10-01 backtest is completed. Weights change only with a new published version; they are never tuned quietly.