RPIX 2.7.0
How the ranking works
Version 2.7.0, adopted 12 October 2026. Provisional until the backtest. Version 2.5 (adopted 5 October 2026) and earlier stay in the method history. A score from 0 to 100 for Salvadoran players who are actually playing.
Choose your explanation
Who are you reading this as?
The short version
RPIX compares the overall club record of players in similar positions. Readiness asks who was actually playing immediately before a national-team camp. A higher number is stronger evidence under the same rules—not a promise about the next match.
How to use it in selection
Use RPIX to identify the stronger season profile, Readiness to check immediate match activity, and the position tables to compare like with like. Treat a flagged gap as a review prompt. Before recommending a player, confirm role, availability, eligibility, opponent and tactical fit.
Do not use the selection signal by itself to choose a squad. It orders public evidence; it does not observe training or private medical information.
Model definition and validation
RPIX = G × (0.05 M + 0.35 T + 0.10 R + 0.45 O + 0.05 Y)
RPIX is a deterministic composite index with position-conditioned empirical-Bayes talent percentiles (calibrated true spread per statistic), explicit missingness rules and a versioned eligibility gate. Readiness v1 is a separate deterministic recency index. Neither index has established calibration or out-of-sample predictive validity yet.
Plain-language guide
What RPIX is trying to answer
RPIX is a 0-to-100 comparison of footballers from the same country. A higher number means the available club data describes a stronger overall profile under the published rules; it does not mean the player is guaranteed to perform better in the next match.
Half of the score is what a player does, and half is how much he plays. What he produces on the field for his position counts 45 percent. How regularly and at what level he plays counts another 45 percent: 35 percent for the season and 10 percent for his most recent games. Market context and the credit for a young player already earning minutes count 5 percent each.
Players are compared with others in the same position. A goal means something different for a striker and for a goalkeeper.
Recent form looks at the last 12 league games of the current season: minutes played out of the minutes available, at the level of the league. A game a player missed while on national-team duty does not count against him: we excuse club games from 5 days before his first national-team match (travel and camp) to 1 day after his last one. Missed games outside that window count as zero. If a player has fewer than 3 games to look at, recent form is not used for him.
Readiness is a separate pre-camp signal and never changes RPIX. Recent form is part of the RPIX score.
League factors are the author's judgment, with European leagues rated above leagues in the Americas. They are not yet measured, and each one is listed on the Leagues page.
If a league publishes only part of the statistics for a position, the missing part counts as average, neither good nor bad.
Defenders lose most from this: tackles, interceptions, clearances and aerial duels are not published for most leagues, so a defender is rated on fewer statistics and the overall list leans toward attackers. Use the position rank next to each score to compare players within their position.
What changed in version 2.7
Talent percentile instead of rank percentile. Until 2.6.1 each statistic was ranked inside the position group, so any difference, even one goal of luck in a few games, was stretched across 0 to 100. Since 2.7 we first measure how much players of each position really differ in each statistic, after removing the luck expected from their minutes, and score each player by where his best estimate sits in that range. A statistic with real spread, such as goals for forwards and midfielders, keeps it. A statistic with almost no measurable spread, such as goals and assists for centre-backs or clean sheets for goalkeepers (which mostly reflect the team), now separates players very little, and playing more minutes without scoring no longer lowers a defender. The spread of each statistic is measured once on El Salvador and Guatemala together and published in the methodology record; it is refitted each season. Formula, weights, the 270-minute gate and league factors are unchanged.
What changed in version 2.6
Minimum reference group. A statistic is scored only when at least 8 players of the same country and position group have it. Below 8 it is shown as no reference group and counts as not available, like a statistic the league does not publish. Before, a group of one player was compared with himself and received the neutral 50th percentile; that did not move the score, but it raised coverage, confidence and the data tier without any new information.
Position actually played. A player is scored in the position group he lined up in this season (Transfermarkt per-game position), not only the one on his profile. The group switches only when he has at least 270 league minutes with a known position, at least 60 percent of them in the new group, and the position known for at least half of his league minutes. Otherwise the profile position is used. A goalkeeper is never moved to the outfield or the other way round. A player that Transfermarkt lists only as Attack is scored as a forward. Position ranks compare players within the group they are scored in.
Players with two teams (added in 2.6.1). Some players are on a first-team roster but play for the reserve team. The league he is scored in is now the league-level competition where he played the most minutes this season. Minutes in another league-level competition of the same season also count, at the league factor of that competition, but only if he made at least one appearance there and never for the same league twice. The 270-minute gate counts the minutes across them. A reserve team is a lower level, so playing there earns no bonus; it only stops his minutes from being ignored.
Career context and the Watch asterisk (evidence beside the score, never part of it). Members see each player's league-level history by season on his page. A young player gets a Watch asterisk when all three hold: his current minutes are limited (under 450 league minutes, or not ranked); he is 25 or younger; and he has a documented record, meaning an appearance in the last three seasons in a league rated 0.70 or higher, senior national-team caps, or a youth national-team squad (U20). The asterisk says nothing about why the minutes are limited and does not change RPIX.
How to read the position weights
The weights in the tables are nominal. A player is scored only on the statistics that exist for him, and the weights are re-scaled over those, with the missing part counting as average. Today every centre-back in the ranking has goals, assists and discipline only: tackles, passing and defensive actions are not in the data we use for any centre-back. About 62 percent of a centre-back performance score is therefore the neutral value. The RPIX of a defender says little about how well he defends; it reflects his minutes, his level and his goals, assists and cards. Compare defenders with other defenders and read the confidence label.
Peers are players of the same country and position group in the ranking, not the whole league. Statistics are compared as rates per 90 minutes, pulled toward the group average according to his minutes and to how much players of that position really differ in that statistic (version 2.7). League level is applied after the comparison: performance, playing level, recent form and the youth credit are scaled by the league factor; market value is not.
Other data providers. Only Transfermarkt data enters RPIX. Match reports from other providers, such as Wyscout, can be shown as labelled evidence on a profile and never change the score: providers count actions differently and no group of comparable players has data from the same other provider.
Known limits. With few players in a group, one result can move a score a lot. The measured spread of each statistic rests on a few hundred players and is refitted each season. League factors are provisional judgements.
The score is evidence for comparison, not a scouting verdict. Injuries, tactical instructions, eligibility and private information may remain outside the data.
- Performance 45% what he produces in matches
- Playing level 35% how much he plays, and where
- Recent form 10% his last 12 league games
- Market 5% market value
- Prospect 5% young and already playing
Who gets ranked
Minimum eligibility threshold: 270 current-season league minutes (3 × 90).
The ranking gate is binary: G = 1 at 270 minutes or more; G = 0 below 270. When G = 0, RPIX is withheld from the ranking even if its components can be calculated.
The threshold limits the influence of very small samples; it does not establish that 270 minutes are statistically conclusive. It is provisional and must be evaluated in the published backtest.
"Not ranked — insufficient minutes" is a data-eligibility status, not a judgment of player quality. The profile continues to show the observed statistics, minutes and component values.
What counts and what does not
Counts
- Minutes played, and how much of his club's league matches that is
- The level of his league
- Goals and assists
- Cards (fewer is better)
- Shots on target, tackles, passes and saves, only where the league publishes them
- Market value, as context (5%)
- Age, only when the player is actually playing
Does not count
- Height
- National-team call-ups
- The coach's opinion
- Fan votes
- The AI second opinion
- The player's name, or how famous his club is
Confidence and data tiers
Every score comes with a confidence number from 0 to 100. It says how much data stands behind the score.
High confidence: many minutes, many matches, full statistics. Low confidence: few minutes or missing statistics. Read a low-confidence score with care.
FULL = the league publishes detailed statistics (shots, tackles, passes, saves). BASIC = only minutes, goals, assists and cards.
Confidence never changes the RPIX. It only tells you how far to trust it.
Three separate measures, not one mystery number
A player can be excellent over a season, short of match rhythm this week, and still perform well or badly for his country. Those are different questions, so futbolfutbol does not force them into RPIX.
1. RPIX — club quality active
The longer-term club-season score. RPIX = 45% position-adjusted performance + 35% playing time and competition level over the season + 10% recent form + 5% market context + 5% youth prospect. A national-team call-up or result never adds points.
2. Readiness — immediate preparation active
Calculated at the camp cutoff from the last five known club matches: 55% share of available minutes + 25% start share + 20% freshness since the last appearance. Known injury, suspension or national-team unavailability can cap it. It never changes RPIX.
3. National-team performance proposed, not scored yet
A future match-by-match measure of what happened after selection: position-specific contribution, decisive actions, efficiency, opponent context and discipline. It will be shown separately and will not rewrite the pre-camp recommendation.
Selection signal active in the squad audit
Selection signal v1 = 0.60 × RPIX + 0.40 × Readiness
It compares players only inside the same position group and keeps the same number of squad places used by the coach. It identifies a potential selection gap, not proof of misconduct or tactical certainty.
Readiness v1 uses only evidence currently available on this site. Recent match ratings and private medical information are not guessed. Historical camp snapshots store the cutoff, evidence and formula version so later matches cannot silently rewrite the original assessment.
Scientific status
What is measured, and what still must be demonstrated
RPIX is a versioned quantitative index: a fixed rule combines observable inputs into a reproducible score. It is not yet a scientifically validated predictor. Validation requires testing predictions on data that were not used to choose the formula.
| Scientific property | Meaning | Current status |
|---|---|---|
| Reproducibility | The same frozen inputs and formula version produce the same result. | Implemented |
| Construct validity | Whether the chosen variables adequately represent football quality, readiness and positional performance. | Partially supported by design; not established |
| Reliability | Whether scores remain reasonably stable when the underlying level has not changed, rather than moving mainly because of sampling noise. | Not yet formally measured |
| Predictive validity | Whether today's score predicts later minutes, level-weighted participation or position-adjusted performance better than simpler baselines. | Backtest pending |
| Calibration | What a difference such as 5 or 10 RPIX points means in observed future outcomes. | Not yet established |
| External validity | Whether conclusions generalise across countries, leagues, positions and seasons. | Not yet established |
Accordingly, the site uses the words score, index, comparison, signal and hypothesis. It does not describe RPIX as proof, certainty or a definitive measure of talent.
National-team call-ups
Call-ups never change RPIX. We show them next to the ranking so you can compare.
We detect each camp from club line-ups: the date from which most called-up players are left out of their clubs' squads. For example, the September 2026 camp ran from 15 to 30 September. Club matches during a camp do not count against a player.
A player is flagged only if he was left out of his club's squad in his last 2 known matches before the camp.
AI second opinion
- What it is. The AI second opinion comes from a 9-billion-parameter model that runs on our own machines, not in a cloud service. We use it like a second reviewer: it reads a short table of numbers and says, in plain words, whether a called-up player looks justified. It is a readable opinion, not a measurement.
- What it sees. For each player called up (played or on the bench) in a senior national-team window, only numbers that are already on this site, with the name replaced by "Player A": position group, age, club league and league factor, data tier, league minutes, club matches and appearances this season, his last 5 club league matches before the camp, status, RPIX with O, T, M and Y, confidence, the call-up check result and national-team caps. It also sees the same numbers for the 3 highest-ranked players at the same position who were not called up. It is told to judge only from club form and those numbers, and to answer "borderline" when the data is thin.
- The question is neutral and identical for every player. "Based only on this data, should this player be in this national-team squad?" Same instructions, same settings (temperature 0, fixed seed 20261001) and same prompt version for everyone; the version is shown with every answer. The wording never mentions favouritism, suspicion or what anyone believes. The answer is Yes, Borderline or No, with two short reasons that cite the numbers.
- It is never part of RPIX. The AI does not change RPIX, the call-up check, the selection gap or any score, and it is never used to compute them.
- The proof is the numbers. The evidence is the selection gap and the call-up check, which anyone can recompute. The AI only adds a readable opinion next to them; if it is wrong, the numbers still stand.
- Anyone can check it. Every answer is stored with the exact numbers it was shown (open "What was asked" beside each answer), the model, the machine that ran it, the date and the prompt version. If the AI cannot answer, or its reply is not valid, nothing is shown: we never make up an answer. Only the site administrator can ask the AI; visitors never trigger it.
How voting works
- Who can vote. Anyone who signs up and confirms their email. The confirmation link is sent by email and works once. Accounts are free.
- Verification. You can apply as a Reporter or as Coach / club staff. An admin checks that role. Until then your votes count as a fan's. A verified role shows a badge ("Verified press", "Verified coaching staff") and has its own results group.
- Public names. Your display name, role and vote date are shown publicly, next to your vote. Your email is never shown. Privacy.
- Three questions. Is this RPIX fair (Underrated, Fair, Overrated, with a reason of 10 to 200 characters); should he be in the latest squad (Yes or No); and how much confidence you have in the coach's selection (1 to 5). One vote per question; you can change or remove it, and every change is logged.
- Reasons are reviewed. An admin approves each reason before it is public. The vote itself counts right away.
- 5-vote threshold. Fans (everyone with a confirmed email), Verified press and Verified coaching staff are shown separately, each only from 5 votes. Fan score = 50 + 50 x average vote.
- Votes never change RPIX. A separate "Community RPIX" will be published only after 1,000 votes, and it can differ from RPIX by at most +5 or -5 points. The official RPIX stays as it is.
Data sources and updates
- Every figure on a player page carries its source link and an "as of" date. Market values, profile data and per-match statistics come from Transfermarkt; national-team call-ups come from the same per-game data.
- Data is refreshed by hand and the ranking is recomputed from the stored figures (age is recomputed daily). Anything we cannot confirm is left empty and shown as "not collected".
Fairness
We judge players on merits. The same rules apply to everyone in the same situation.
We never tune a rule for a player. We never guess missing data. A missing statistic is missing information, not a bad game.
The formula was frozen before any test of it. It changes only with a new, published version.
Does it work? Our hypothesis
RPIX is a hypothesis, not a fact. We test it.
H1 Does RPIX today predict how a player's market value changes over the next 12 months, better than market value alone?
H2 Does it predict how much he will play, weighted by league level, better than his past minutes?
H3 Does a high performance score (O) repeat in the next period, or is it noise?
Status: the backtest has not been run yet. The plan is on the Backtest page.
Technical specification
Formulas, parameters and decision rules
The complete reproducible specification is available to registered members.