01What this is, and what it isn't
This platform turns raw match events into judgments about how teams play and what players actually do. It is a scouting instrument, not a stats dump. Everything in it is built to answer one of two questions: what kind of player is this, and what kind of team is this.
Every number you will see comes from one source: event data scraped from WhoScored, covering 254 played matches and 393,129 individual events across the 2026 MLS season to date. Nothing is bought, nothing is estimated from box scores.
What it is good at
- Style description. How a side builds, where they progress, how they behave when losing. These are the most reliable outputs because they aggregate thousands of possessions.
- Role identification. What a player does with the ball relative to others in his position. Not how good he is — what he is.
- Finding profiles you specify. The search tool answers questions like "left-footed right-sided carriers under 23" in a way a stats site cannot.
What it cannot do
Honest limitations
- No off-ball information. Event data records the player on the ball. A striker's runs that drag a centre-back out of position are invisible. Defensive positioning is invisible. This is the single biggest limitation and it never goes away.
- No defender positions. We can see where a pass started and ended, not who was between those points. This is why passes are labelled by trajectory, not as "line-breaking" — we cannot verify a line was broken.
- Unused substitutes are invisible. Squad data only contains players who touched the ball. "Fringe" means plays rarely and late, not "end of the bench".
- Age is a snapshot. WhoScored reports age at match time, not date of birth, so a player can read a year light. We store the highest age observed.
- One season. Roughly 17 games per side. Team style is firm at this sample; individual finishing and anything rate-based on low volume is not.
A scout who knows these limits will get far more out of this than one who doesn't. Most bad conclusions from event data come from forgetting what the data cannot see.
02How the data is built
Everything is a layer on top of the layer below. Understanding this order explains why some numbers are trustworthy and others need care.
- Events. Every pass, carry, tackle, shot, with coordinates, outcome, and qualifiers. 393,129 of them. This is the raw material.
- Possession sequences. Events grouped into unbroken spells of control by one team. 65,186 sequences. A sequence ends when the ball is lost, the game stops, or a set piece intervenes.
- Team style. Sequences aggregated per club, then standardised against the league so every trait reads as "how unusual is this".
- Player involvement. Each player's touches mapped back onto the sequence they belong to, and where in that sequence they occurred.
- Chain roles. What share of a player's involvements look like initiating, progressing, carrying, finishing, and so on.
- Percentiles. Every metric ranked within position pool, so a centre-back is compared against centre-backs.
League benchmarks — memorise these
You cannot judge whether a number is high without knowing the middle. For MLS 2026:
| Benchmark | League average |
|---|---|
| Passes per possession | 3.61 |
| Seconds per possession | 10.2 |
| xT per possession | 0.0164 |
| Possessions ending in a shot | 9.2% |
A side averaging five passes per possession is not "slightly above average", it is a long way out. The spread is narrower than intuition suggests.
03The metric families
Seven families. Each answers a different question. The skill is knowing which one to reach for.
Expected threat (xT)
Divide the pitch into a grid. Each cell has a value: the probability a possession starting there ends in a goal. Move the ball from a low-value cell to a high-value one and you have added threat. That difference is xT.
It is the platform's core currency because it values moving the ball, not just the final action. A midfielder who never shoots can generate serious xT.
Why xT beats counting passes
A completed pass in your own half from one centre-back to another adds essentially nothing. A completed pass from the halfway line into the channel behind the full-back adds a lot. Pass completion percentage treats them identically. xT does not.
Read it as: per 90, split into xt_pass_90 and
xt_carry_90. Carrying and passing are different skills and separating them
is often where the insight is.
The trap: xT rewards risk-free progression. A player who only ever plays the safe forward pass in front of a low block accumulates xT without ever unlocking anything. Always read xT alongside chance creation.
Sequences and team style
Every possession is tagged with how it started, how it travelled, and how it ended. Those tags aggregate into a style profile: does a side build from deep, go over the top, work the ball wide, switch play, hold it up.
The important refinement is route productivity. Knowing a side goes through the middle 35% of the time is trivia. Knowing that route produces a shot far more often than the league average is a scouting report.
Real examples
| Side | Signature route | Share | Verdict |
|---|---|---|---|
| Inter Miami | Through the middle | 34.9% | Effective |
| San Diego | Patient build | 24.6% | Unproductive |
Miami funnelling centrally is Messi and Busquets in a phone box, and it works. San Diego build patiently and it does not pay — the definition of sterile possession, and a squad with a specific hole in it.
Chain roles
Eleven behaviours, each expressed as a share of a player's total involvements: Initiator, Bridge, Progressor, Carrier, Vertical, Support angle, Individual, Creator, Box threat, Finisher, Tempo.
These describe function, not quality. A poor player and an excellent one can share a role profile. That separation is deliberate: figure out what a player is, then ask separately how well he does it.
Archetypes from the current data
| Player | Archetype |
|---|---|
| Lionel Messi | Finisher / Vertical Passer |
| Carles Gil | Progressor / Third-Man Bridge |
| Rodrigo De Paul | Third-Man Bridge / Carrier |
| Héctor Herrera | Third-Man Bridge / Tempo Setter |
Note Messi reads as a finisher rather than a creator. At 39 he is positioned to end moves, not build them. The data captures the change in role without being told.
Pass trajectory
Forward passes classified by how they travel and where they land. Over (chipped or long), around (worked laterally), or through (played on the floor between bodies). Landing inside, in behind, or outside.
This is the closest thing to describing a passing style rather than counting passes.
Carles Gil's distribution
Over 27.1% · Around 67.4% · Through 5.5%
Inside 22.6% · In behind 45.6% · Outside 31.9%
Overwhelmingly a worker of angles rather than a chipper. Nearly half his forward passes land in behind, which is why he creates from deep rather than in the box.
Read the label carefully
This is trajectory, not "line-breaking". We know the ball's path, not the defenders' positions. A through-pass classification means the ball travelled on the floor into a central area; it does not prove a line was split.
Chain-position value
Almost nobody measures this and it may be the platform's most distinctive output. For every possession that ended in a shot, which players were involved three or more actions before the end?
The assist and key pass are already well counted. The pass that starts the move is not.
Early involvement in shot-ending chains, per 90
| Player | Club | Value |
|---|---|---|
| Héctor Herrera | Houston | 7.02 |
| Jeppe Tverskov | San Diego | 5.57 |
| Tristan Muyumba | Atlanta | 5.16 |
| Manu Duah | San Diego | 5.08 |
Deep midfielders and a ball-playing centre-back — precisely the players whose contribution disappears from conventional metrics. If you want the man whose team's dangerous moves run through him rather than end with him, this is the column.
Squad role and leverage
Two axes. Selection share is minutes played as a proportion of minutes available while the player was at the club, so a July signing is not punished for games he could not have played. Leverage is the share of his minutes played with the score within one goal.
The minutes-inflated trap
A player with strong per-90 numbers accumulated at 3-0 is not the same as one producing at 1-1. Leverage separates them, and it is invisible in every conventional stat.
Current flags include Dejan Joveljić (98.5% of available minutes but 71.8% leverage) and Braian Ojeda (98.3% / 68.2%). Both play almost everything; a disproportionate share of it comes with the game already decided.
This does not make them bad players. It means their rate stats deserve a second look before you pay for them.
Game state
Every possession is tagged with the score at the moment it began. That single addition unlocks the difference between a team's identity and its reaction.
Directness by game state
| Side | Winning | Losing | Swing |
|---|---|---|---|
| Portland | 0.318 | 0.377 | +0.059 |
| Chicago | 0.383 | 0.424 | +0.041 |
Portland change character sharply when chasing. A side whose numbers barely move between states has a settled identity; a side that swings hard is reactive. Both are useful to know before you play them.
04Reading the maps
On the Players page, click any value in a ranking table to plot the actions behind it. The number tells you how much; the map tells you where, and where is usually the more useful half.
Pass maps
Look for three things, in this order:
- Origin cluster. Where does he receive? A midfielder whose passes all start in his own half is a different player from one starting between the lines, even with identical volume.
- Direction and fan. A narrow fan pointing forward means a specialist. A wide fan means a distributor. Neither is better; they suit different systems.
- The empty regions. What he never attempts is as informative as what he does. A deep midfielder with no passes into the left channel may have a genuine limitation.
Carry maps
Carries show ball-progression under his own power. Long lines from deep are a ball-carrying midfielder; short repeated lines near the touchline are a winger isolating a full-back.
Watch for: carries that end in the same place repeatedly. That is usually a player who runs into a wall — carrying without penetrating.
Shot maps
Point position is shot location, size should be read as chance quality. The question is never "how many" but "from where".
- Tight cluster inside the six-yard box: a genuine penalty-box striker.
- Scattered attempts from 25 yards: either a specialist or a player with poor shot selection. The conversion column tells you which.
- Everything from one side of the box: a cutting-in wide forward, not a centre-forward.
xT maps
The newest view, and the one requiring the most care. Each arrow is a pass or carry weighted by threat added.
Keep the negatives visible
A backwards pass subtracts threat. Showing only positive actions flatters everybody equally and teaches you nothing. A player whose map is dense with small positives and few negatives is a low-risk progressor; one with large positives and large negatives is a gambler. Both exist and you should be able to tell them apart at a glance.
05Building a shortlist
The Search page is the recruitment tool. It loads every profiled player once and filters instantly, so exploration is cheap — change a slider and look again.
The method
- Define the role, not the position. "Left centre-back" is a position. "A centre-back who can carry out of pressure and hit a diagonal" is a role. Only the second can be searched.
- Set the pool and the floor first. Position pool, then minimum 90s. Six is the sensible default; below that, rate stats are noise.
- Add one behavioural filter. Trait sliders filter on percentile within pool. Start with one at 70 and see how many survive.
- Sort by the thing that matters. Filtering decides who qualifies; sorting decides who leads. "Efficient" is not one number — choose whether you mean threat generated, volume, or penetration.
- Read the survivors as profiles, not ranks. Open two or three cards and compare. The list is a starting point, never an answer.
Worked example: the original question
"The most efficient right-sided carriers who are left-footed."
Pool: wide. Side: right. Foot: left. Floor: 6 nineties. Sort: xT from carries.
| Player | Club | Age | xT/carry 90 |
|---|---|---|---|
| Gabriel Pec | LA Galaxy | 25 | 0.381 |
| Cavan Sullivan | Philadelphia | 16 | 0.376 |
| Aiden Hezarkhani | Salt Lake | 19 | 0.338 |
| Anders Dreyer | San Diego | 28 | 0.329 |
Note what this surfaced: an established international, a sixteen-year-old, and two in between. That age spread is the point — the query described a profile and the tool returned everyone who fits it regardless of reputation.
06Worked cases
Case 1 — We need a ball-playing centre-back under 23
Approach. Pool CB, age cap 23, floor 6 nineties, progression trait at 70th percentile, sorted by progressive passes per 90.
Result. Justin Che (New York, 22) at 11.36 progressive passes per 90, Olwethu Makhanya (Philadelphia, 22) at 8.08, Lucas Herrington (Colorado, 18) at 7.47.
Now think. Volume is not quality. Open each card and check progressive pass completion. A defender attempting many and completing few may be forced into it by a poor structure rather than choosing it. Then check chain-position value: does his progression actually feed moves that end in shots?
Case 2 — Our striker's numbers look great. Should we trust them?
Approach. Check the game-state adjusted output before anything else.
Result. Sam Surridge reads 0.544 raw xG per 90, but 0.452 once weighted for whether the game was live. Rafael Navarro drops from 0.401 to 0.320.
Now think. Neither is a bad player, but a meaningful share of their output arrived in games already decided. Cross-check leverage on the squad-role view. If selection share is high and leverage is low, you are looking at a player whose rate stats are inflated by garbage time.
Case 3 — How do we beat this team?
Approach. Team page press profile plus the route breakdown.
Result. Philadelphia contain direct play 2.38 standard deviations better than the league, but are neutral against short build-up (-0.11). Salt Lake are similar: excellent against direct (1.82), unremarkable against short (0.02).
Now think. Going long against either is playing to their strength. The plan is patient build and playing through them, and the personnel question follows from that: do we have midfielders who can receive under pressure?
Case 4 — Find me a cheaper version of a player we can't afford
Approach. Open the target's card, read the "a bit like" comparisons, then use the full similarity function which compares across 32 dimensions rather than chain roles alone.
Now think. Similarity is stylistic, not qualitative. The closest match to an elite player is often a player who does the same things less well — which is exactly what you want if the price gap is large and the system fit matters more than peak quality. Always check squad role: a similar profile who is a rotation player at a weaker club is a different proposition from a key player.
07Traps and judgment
Everything below has caught someone out. Most of them have caught me.
The pool trap
Percentiles are computed within position pool. A player at the 85th percentile for progression among wingers is not comparable to the 85th percentile among centre-backs — the underlying distributions are completely different. Always know which pool you are reading.
The small-sample trap
Six nineties is the floor for a reason. Below it, finishing metrics in particular are close to meaningless: a player can be at the 99th percentile for conversion on four goals from six shots. Role percentages stabilise faster than rate stats.
The style-is-not-quality trap
Chain roles describe function. A player can be the league's most extreme progressor and a poor footballer. The archetype tells you where he fits; the metrics tell you how well.
The team-effect trap
A midfielder at a possession-dominant side will accumulate progressive passes because his team has the ball more. Some of what looks like individual quality is team context. Compare within similar teams where you can, and treat volume metrics with more suspicion than efficiency ones.
The absence-of-evidence trap
If a player shows nothing for aerial duels, that may mean he is poor in the air — or that his team never puts him in aerial situations. Event data records what happened, not what was possible.
The single most useful habit
Before believing any number, ask: what would make this number high for a reason other than the one I'm assuming? Nearly every mistake in scouting analytics is an unexamined alternative explanation.
08Test yourself
Three kinds. The quiz checks comprehension, the exercises check reasoning, the tasks check that you can actually drive the tool.
Part A — Comprehension quiz
Ten questions. Click an answer to see whether it's right and why.
Part B — Reasoning exercises
Written answers. Think first, then open the model answer. There is usually more than one defensible response — the reasoning matters more than matching mine.
B1. A winger sits at the 95th percentile for take-ons but the 20th for xT from carries. What is happening, and would you sign him?
He beats his man frequently but the ball does not end up in more dangerous places afterwards. Three plausible explanations: he dribbles laterally or backwards; he beats his man in low-value areas; or he takes on so often that successful attempts are diluted by a poor success rate — check take-on percentage to separate these.
Whether to sign depends on system. In a side that needs someone to hold width and occupy a full-back, the take-ons have value even without progression. In a side that needs penetration, this profile is decorative. The metric does not answer the question; it tells you which question to ask the coach.
B2. Two centre-backs have identical progressive pass volume. One has far higher chain-position value. What does that tell you?
Volume says they attempt the same amount of progression. Chain-position value says one of them progresses into moves that actually end in shots, while the other's progression dissipates. That is the difference between a defender who breaks lines usefully and one who moves the ball forward into a dead end.
Caveat: team quality contaminates this. A centre-back at a side with excellent forwards will look better on chain-position value regardless of his own contribution. Compare their teams' shot-ending rates before concluding.
B3. Your side is 68% possession and bottom-third for xT per possession. Diagnose it, and say what you would buy.
Sterile control. The side can keep the ball but cannot move it into dangerous areas. Check the route breakdown to find where possessions die: if the signature route is patient build with an unproductive verdict, the problem is the final third, not the build-up.
Then check the squad for a chance-creation gap. If no player is above the 60th percentile in-pool for creation, the diagnosis is straightforward. What you buy is a line-breaking passer or a runner in behind, and which one depends on whether opponents are sitting deep against you — a side facing low blocks needs the passer.
B4. A player is flagged minutes-inflated. Give two reasons this might be unfair.
First, he may play for a dominant side that wins by three regularly, so low leverage reflects his team rather than his usage — the flag is measured against league-wide leverage, so a whole squad can be caught.
Second, a substitute who habitually comes on to close out games will have low leverage by design; that is a defined and valuable role, not a sign his output is hollow. The flag is a prompt to look closer, never a verdict.
B5. Why is "line-breaking passes" labelled "pass trajectory" here, and why does that distinction matter for a scouting report?
Because we have the ball's path but not the defenders' positions. We can say a pass travelled on the floor into a central area; we cannot say it split a defensive line.
It matters because a scouting report that claims verified line-breaking is making a claim the data does not support. If challenged in a meeting, you would have to concede the point, which undermines everything else in the document. Precision about what a metric proves is what makes the rest credible.
Part C — Tasks in the tool
Do these in the site itself. Each has a checkable answer.
C1. Find the three most creative attacking midfielders under 25 with at least eight 90s. Which one has the lowest pass completion, and why might that be acceptable?
How: Search → pool AM → age 25 → minutes 8 → creation trait 70 → sort by xA per 90.
The point: creators often complete fewer passes because through balls and passes in behind fail more often than safe ones. Low completion in a high-creation profile is frequently a feature. Check whether his incomplete passes are ambitious ones by opening the pass map.
C2. Pick any team. Identify their signature route and whether it pays off. Then name one opponent trait that would neutralise them.
How: Insights → filter by that team → read the team profile card.
The point: a side reliant on one productive route is vulnerable to an opponent strong against exactly that route. Cross-reference with the press profile of the teams they play worst against.
C3. Use the rankings drill-through: pick the league leader for xT from passing, plot his actions, and describe his passing shape in one sentence.
How: Players → Rank tab → metric xT from passing → click the top value.
The point: this is the habit worth building. Never accept a ranking without looking at the actions behind it. A single map will tell you in seconds whether the number reflects what you assumed.
C4. Find a player whose archetype does not match his listed position, and explain whether it is a misuse or a deliberate tactical choice.
How: Insights → Recruitment → look for misfit profiles.
The point: a striker reading as a Link Player is usually a false nine by design. A centre-back reading as a Carrier may be covering for a lack of midfield progression. Distinguishing intent from accident requires watching the games — the data only flags the anomaly.
09Quick reference
| Term | Meaning | Watch for |
|---|---|---|
| xT | Threat added by moving the ball between pitch zones | Rewards safe progression; pair with creation |
| Sequence | Unbroken spell of possession by one team | Ends on loss, stoppage, or set piece |
| Chain role | Share of involvements matching a behaviour | Function, not quality |
| Pool | Position group used for percentile comparison | CB, FB, CM, AM, W, ST |
| Trajectory | Over / around / through, by destination | Not verified line-breaking |
| Chain-position value | Early involvement in shot-ending moves | Contaminated by team quality |
| Selection share | Minutes as a share of minutes available | Availability-adjusted, not raw |
| Leverage | Share of minutes with the score within one | Low = output may be inflated |
| Directness | Net upfield progress per unit of ball travel | Compare across game states |
| Archetype | Two highest-ranked chain roles | Descriptive label, not a grade |
10Credits and influences
Almost nothing here is a wholly original idea. The work below shaped what this platform measures and how it presents it. Where the implementation departs from the original, that is noted, because the differences usually exist for a reason worth knowing.
ContributedThe framework this platform is built on. Segmenting a match into unbroken possession sequences, tagging each by how it started, travelled and ended, and then decomposing a player's involvements into a set of repeatable roles.
How ours differsBuilt on WhoScored event data, and validated against his published Benfica reference figures before being trusted (3.55 passes per sequence against his 3.6, 9.9 seconds against 9.6). The player layer is our own extension: chain roles are expressed as percentiles within position pool, and carries had to be derived from consecutive same-player touches because the event feed has no carry event.
ContributedExpected threat, the idea that a pitch can be valued as a grid of goal probabilities so that moving the ball between zones has a measurable worth. It is the core currency of most of what this site reports.
How ours differsA 12x8 grid applied to both passes and derived carries, reported separately as threat from passing and threat from carrying, since they are different skills. Extended to event-level maps where every arrow is one action, and the arrows sum exactly to the published per-90 figure rather than approximating it.
ContributedTwo things. Classifying passes by how they beat a block, over, around or through, crossed against where the ball lands. And the scouting card format: banded trait pills, an archetype label, and a stylistic comparison.
How ours differsOne difference matters a great deal. Theirs is labelled line-breaking; ours is labelled trajectory, because WhoScored provides no defender positions and we cannot verify a defensive line was actually broken. Claiming otherwise would be asserting something the data does not support.
ContributedThe idea of splitting a team's style by scoreline, so that identity can be separated from reaction. Their winning, drawing and losing chart is the direct inspiration for ours.
How ours differsComputed from our own sequence layer on top of a game-state spine built from goal timelines, with own goals credited to the correct side. We added the losing-minus-winning swing as a single reactivity number, which turns the chart into something sortable rather than only readable.
ContributedThe progression quadrant: how often a player attempts to progress against how often it comes off, with the median lines turning it into four readable player types rather than a cloud of dots.
How ours differsWe separated tendency from volume. Progressive pass tendency is the share of a player's passing that is progressive, which is a different statement from how many he plays, and a low-volume player in a low-possession side can still have a high tendency.
ContributedThe argument that raw output should be adjusted for the context it was produced in, demonstrated by inferring match difficulty from closing betting odds.
How ours differsWe took the argument and rejected the method. Odds introduce an external data dependency with licensing and reliability questions attached. Instead we adjust for whether the game was still live, weighting output by score margin, which is computable from data we already hold and defensible without a third party. The idea is theirs; the implementation is deliberately different.
Data and tooling
Every figure on this site derives from WhoScored's match event feed. Its limits are this platform's limits, which is why section 01 spends as long on what the data cannot see as on what it can.
The Python package underneath the ingest pipeline. The multi-league scraper, the caching behaviour and the league configuration all sit on top of it.
This list is incomplete
These are the influences with a direct, traceable line into something on this site. Plenty of other people's writing and charts shaped how the platform thinks without leaving a single identifiable feature behind. If you recognise your work here and it is credited wrongly, or not credited at all, that is an oversight rather than a decision.