A player profile should separate in-possession, out-of-possession and transition work. A single final score is useful only after the club decides how much each part matters for the role it is buying.

The model should compare a player with the jobs he will actually perform, not with everyone who shares his listed position.

Listed position is only the first filter

Position tells you where to start looking. It shouldn’t be where the model stops.

Firmino and Lewandowski both go down as centre-forwards, but they solve completely different problems. One drops, combines, creates space for runners. The other pins defenders, attacks the box, finishes at volume. Same position, different job entirely.

Full-backs have the same problem. The player who holds width, receives high, and creates chances down the line is doing different work from the one protecting the rest defense and securing the back post. Comparing them against one template doesn’t tell you much.

That’s what role-based grouping is for. Stats Perform’s Role Discovery separates players through spatial footprint, Playing Styles engagement, expected pass completion, Possession Value, and movement-chain involvement — so Firmino ends up benchmarked against creative attacking threats, Lewandowski against advanced forwards. StatsBomb’s full-back recruitment walkthrough does the same thing within one position, building separate templates for a Trent-style creator and a Pavard-style defender.

The question isn’t who’s the better full-back in the abstract. It’s who fits the work the team needs done.

Split the evidence by phase before comparing players

A player’s job changes constantly throughout a match — receiving, pressing, retreating, recovering, countering. One average across all of it can hide where he’s actually helping or falling short.

Phase also changes the peer group.

A midfielder who looks ordinary as a ball-winner might be elite in the five seconds after a turnover. A winger who produces in settled possession might offer nothing on a deep counter. A centre-back who passes cleanly against a low block might struggle defending space after a loss.

A 2025 Data Mining and Knowledge Discovery paper shows how this can work in practice. The authors labeled match phases frame-by-frame from video across 15 first halves, then trained models on 25Hz tracking data. Their FeatGRU model hit 80% frame-wise accuracy against merged expert annotations. Team-state labels were more subjective than basic game-state ones, and transitions were derived rather than explicitly labeled — worth keeping in mind.

For recruitment, that subjectivity isn’t a reason to avoid phase labels. It’s a reason to treat them as uncertain starting points rather than hard facts. Still better than collapsing everything into one score and hoping it captures the job.

Defensive roles need their own phase labels

“Good out of possession” covers too much ground to be useful on its own.

A midfielder who can press aggressively in a high block might still get exposed once his team drops and has to protect space on the retreat. A winger who’s intense in the counter-press might be unreliable defending the far-side full-back in a settled shape. A full-back who wins duels wide might not be comfortable stepping inside to shut down the half-space.

All defensive work, none of it the same job.

SkillCorner’s out-of-possession model breaks this down — separating pressure, pressing, recovery press, and counter-press (defined as pressure within three seconds of a turnover), and tagging actions by high, mid, and low block. High-block pressing and low-block retreating get scored separately instead of disappearing into one off-ball number.

That changes the shortlist. SkillCorner’s pressing-playmakers workflow filters attacking midfielders through high-block pressing engagements, forced-backward outcomes, line-break prevention, and pressing intensity, then layers in an in-possession playmaker profile. You’re not looking for “creative midfielder who also defends.” You’re looking for someone who creates in possession and handles the specific defensive role that slot requires.

Hybrid players need multi-role cards

Some players change role within the same possession sequence, not just match to match.

The inverted full-back is the obvious case. In possession he steps into central midfield to receive and combine. When the ball is lost he might hold central briefly to screen or counter-press. Once the team sets without the ball, he drops back outside the centre-backs. Three different jobs in the space of one phase.

Coaches’ Voice’s inverted full-back explainer describes it clearly: central in possession, involved in the defensive transition, back in the defensive line once things settle.

A single full-back score struggles here. Overweight wide defending and you underrate the central receiving the coach actually cares about. Overweight possession value and you miss whether he can recover outside the centre-backs when an attack breaks down.

SCOUTED’s midfield essay quotes Jon Mackenzie on phase-specific skillsets and Jake Entwistle on phase-based recruitment for multi-phasing midfielders. Those observations don’t prove anything about transfer outcomes, but they point at the real modeling problem: some players need different peer groups for different parts of the same profile.

One card per job the club expects him to do.

The buyer’s game model should set the weights

The final score should be local to the buying club, not universal.

A full-back in a possession-dominant side makes different runs, receives in different areas, and defends in different situations than a full-back who spends most of the match without the ball. The physical and tactical load shifts before the player does anything differently himself.

Liu, Yang, Chen and García-de-Alcaraz’s 2021 CSL study ran this across all 240 Chinese Super League matches from 2018 — 6,163 player match observations tracked at 25 measures per second. K-means clustering split observations into high-possession and low-possession groups (56.53% and 43.98% average possession).

The physical demands moved with the possession profile. High-possession teams covered more high-intensity distance in possession and less out of it. Full-backs in high-possession observations logged higher sprint and high-intensity totals; attackers in those observations logged lower ones. One league-season study doesn’t prove transfer outcomes, but it shows why the same position can carry a different load depending on how the team plays.

A club building through central spaces weights in-possession security differently than one defending low and hitting on the counter. A man-marking press creates different demands than a zonal block. The same player can rank highly against one phase-role peer group and look ordinary against another. The buying club chooses which comparison matters.

Context and video decide whether the role transfers

A phase-role profile narrows the search. It can’t close the decision.

A player with strong high-block pressing numbers in the Bundesliga might land at a Premier League club that presses differently, changes his distances, or asks him to cover different spaces. A full-back who receives inside one structure might not get the same angles or support elsewhere. A midfielder who looks press-resistant with certain teammates might not find the same options with different ones.

StatsBomb 360 adds the surrounding picture — teammate and opponent locations around each event, line-breaking pass classification, receiver space, defender positioning. Most useful when it points toward the right clips.

Show phase-role scores alongside the club’s weights, then send the uncertain parts to video: central receiving under pressure, high-block pressing, recovery runs after loss, the first forward pass after a regain, wide defending in the back line. Those clips decide whether the profile holds up in a new team, under a new coach, against new match demands.