The Execution Layer: Mechanics as the Missing Variable in Transfer Prediction

1. The Summer of Strikers

The 2025–26 Premier League summer transfer window produced the most concentrated test of transfer prediction in the competition's history. English clubs spent over £3 billion in total, with six high-profile forwards alone — Alexander Isak, Benjamin Sesko, Viktor Gyökeres, Hugo Ekitike, João Pedro, and Nick Woltemade — costing a combined £471 million [1]. Liverpool broke the British transfer record twice in a single window. Manchester United, Arsenal, Chelsea, and Newcastle each staked their seasons on the assumption that forward-line investment would resolve attacking deficiencies that had cost them the previous campaign.

By November, the results were stark. Isak, Sesko, and Gyökeres — the three most expensive of the six — had scored a combined six league goals between them [2]. Erling Haaland, the benchmark, had scored fourteen in the same period. By January 2026, Gyökeres had been described as struggling to match the technical level of Arsenal's other attackers [3]. Sesko, despite possessing the physical tools that made him one of the most coveted young forwards in Europe, was managing two goals in seventeen league appearances [4]. Isak, signed for £125 million, had recorded a single shot on target in his opening four league appearances before suffering a season-disrupting injury [5].

The conventional explanation arrived on schedule. Analysts cited "adaptation time." Dr Ian Graham, Liverpool's former director of research, has described a "league exchange rate" — an expected percentage drop in per-90 output when a player moves between leagues. A forward arriving from Portugal's Primeira Liga can expect approximately a 35% average decline in output on a per-90 basis [6]. The adjustment is statistical: take a player's actions in the source environment, apply a discount, project the residual into the target environment.

This is the analytical machinery the market relies on. It is also, we assess, operating on the wrong unit of analysis entirely.

2. Three Layers, One Metric

The transfer prediction infrastructure — from scouting databases to composite models to league exchange rates — measures a single layer of player performance: actions. Goals scored, assists provided, expected goals generated, passes completed, duels won. These are countable events with measurable frequencies. They are the data that populates every scouting platform and every algorithmic shortlist in professional football.

Actions tell you what a player does, and how often. They do not tell you how the player executes those actions, or — critically — whether the execution quality that produced those actions in one environment will produce equivalent actions in another.

Beneath the action layer sit two further layers, each carrying distinct information about a player's capability and each with different implications for transferability.

Technique is the method a player selects to complete an action. A striker presented with a scoring opportunity might drill the ball, curl it to the far post, toe-poke, or side-foot. A defender contesting an aerial situation might jump early for positioning advantage or time a late attack on the ball. Technique tells you how a player attempts to complete an action. It adapts to tactical context — a team that emphasises direct play will demand different technical selections from a team that builds through the thirds. Technique is context-sensitive.

Mechanics is the physical execution quality within a technique. It is how the body generates force, manages balance, times contact, and recovers position. Mechanics tells you why a particular technique succeeds or fails under specific physical conditions. A striker's shooting mechanics — the compactness of their release, the position of their standing foot, the degree of follow-through — determine whether they can reliably generate power, accuracy, and shot speed. Unlike technique, which adapts to tactical instruction, mechanics are grounded in a player's physical structure, training history, and neuromuscular patterning. They are more stable than either actions or technique across environmental changes.

The distinction is consequential: actions are the most context-dependent layer and mechanics the most context-independent, yet the entire transfer prediction infrastructure is built on the context-dependent layer.

A player's goal tally in Portugal is a function of the Portuguese league's defensive standards, the player's team quality, the tactical system, the quality of service, and the player's own capability. When that player moves to England, every contextual factor changes simultaneously. The league exchange rate attempts to adjust for this by applying a blanket discount derived from historical transfer outcomes. But the discount is applied to the action — the aggregate output — without any ability to isolate the component of that output that is genuinely portable: the player's execution quality.

3. Why Mechanics Transfer and Statistics Do Not

The claim that mechanics is the strongest predictor of skill transfer between footballing environments rests on a structural argument about what each layer of performance carries across a change in context.

Actions are environmental products. A striker's goal record is the joint output of their own ability, their teammates' ability to create chances, the opposition's defensive quality, the tactical system's ability to position the striker in dangerous areas, and the volume of minutes played. Change the environment and every input to the goal record changes simultaneously. The action-level data from the source environment is, at best, a noisy proxy for the player's underlying capability. At worst, it is systematically misleading — inflated by a weaker league, deflated by a dysfunctional team, or distorted by a tactical system that created artificial volume.

Technique is partially portable. A player can adapt their technical selections to new tactical demands — learning to play one-touch in a possession system, or to play longer balls in a direct system. But the speed of adaptation and the ceiling of adaptation depend on the mechanical substrate. A player whose reception mechanics involve an open body shape and a directional first touch can adapt to a high-tempo pressing environment. A player whose reception mechanics involve a closed body shape and a controlling first touch will struggle in the same environment regardless of tactical instruction.

Mechanics are the most stable layer. A player's striking mechanics, tackling form, aerial contest positioning, and running gait are products of years of physical development, coaching, and neuromuscular patterning. They do not reset when a player changes clubs. A compact shooting release that generates power from a short backlift will produce powerful shots in Lisbon, London, and Leipzig. A wide shooting release that requires a long backlift to generate equivalent power will be closed down more easily in environments with faster defensive recovery — not because the player has declined, but because the mechanical profile has encountered a context that exposes its limitation.

The layer of player performance least dependent on the environment is the layer most likely to persist when the environment changes. The transfer market's analytical infrastructure is built on the layer most dependent on it.

4. The Compounding Stack

The absence of mechanical assessment does not merely leave a gap in the scouting process. It propagates through every analytical layer that consumes action-level data as its input, and each layer adds the appearance of rigour without correcting the original omission.

Layer 1: Event data collection. Providers record actions — passes, shots, duels, carries. Each recorded event is an action-level observation. The mechanical quality of the action is not captured. Two shots that arrive at the same location with the same xG value may have been produced by fundamentally different mechanical profiles — one compact and repeatable, the other requiring a full backlift that will be closed down in faster defensive environments. The event record does not distinguish between them.

Layer 2: Derived metrics. Expected goals, expected assists, progressive carrying distance, and similar metrics aggregate action-level events into per-90 rates or cumulative values. These metrics inherit the blindness of the event data they consume. A player's xG per 90 in one league is treated as a measure of their shooting capability, but it is in fact a measure of their shooting outcomes in a specific set of conditions. The mechanical capability that produced those outcomes — which may be more or less transferable than the outcomes suggest — is invisible to the metric.

Layer 3: Composite models. Scouting platforms combine derived metrics into composite player ratings, similarity scores, and algorithmic shortlists. These models operate entirely within the action layer. A model that identifies a player as "similar to" a successful Premier League striker on the basis of comparable xG, progressive carries, and aerial duel rates has no mechanism for assessing whether the underlying mechanical profiles are comparable. The similarity is statistical, not physical.

Layer 4: League adjustment. The league exchange rate — the expected drop in per-90 output when transitioning between leagues — is the most revealing symptom of the architectural flaw. The adjustment is a blanket discount applied to the action-level aggregate. It has no mechanism for distinguishing between a player whose mechanical profile predicts successful transfer and a player whose mechanical profile predicts failure. Both receive the same discount, because the variable that separates them is not in the model.

Layer 5: The transfer decision. The club's decision-makers consume the output of layers 1 through 4 and commit investment. At no point in the standard analytical pipeline has anyone assessed whether the player's mechanical execution quality is suited to the demands of the target environment.

Each layer adds analytical sophistication to an input that carries a structural omission. The omission is categorical: the entire pipeline measures the wrong unit of analysis for the question it is trying to answer. The question is "will this player's capability transfer to a new environment?" The pipeline answers "what did this player do in their current environment, adjusted for historical averages?" The distance between these questions is where transfer value is destroyed.

5. Same Statistic, Different Mechanics, Different Prediction

The analytical value of mechanical assessment is most visible where action-level data produces identical readings for players whose physical execution — and therefore whose transferability — diverges entirely.

Consider two Premier League right-backs, both statistically effective tacklers, both deployed in elite systems. Their tackle completion rates, duels won per 90, and defensive action success rates overlap substantially. The action-level data says: equivalent defenders.

The mechanical assessment says something different. The first player contests primarily by using his frame, positioning shoulder-to-shoulder and leveraging upper-body strength. His tackling mechanics are sprint-intensive — he needs to be alongside the attacker before initiating the contest. Post-tackle, the ball is typically retained and the next possession begins immediately. This mechanical profile predicts success in a high-line defensive system where the defender is the last line and the team requires immediate transition from defence to possession. It predicts difficulty in a low-block system where tackles are contested from deeper positions at tighter angles.

The second player challenges on the ground using either foot and from a wider range of angles. His post-tackle outcome is different: from deeper positions, the ball typically goes out of bounds; from higher positions, it ends up loose. He absorbs significant physical punishment — knocks, falls, trampling — requiring a calm, unfazed temperament. This mechanical profile predicts success in a compact, disruptive defensive system — a pressing trap, an insurance role in rest-defence. It predicts difficulty in a possession-first system where tackle outcomes need to initiate the next sequence [7].

This is not a question that action-level data can answer. The tackle completion statistics are comparable. The mechanical profiles are orthogonal. A club signing either player on the basis of defensive statistics alone has a 50% chance of acquiring a mechanical profile that matches their system's demands — and no tool in the standard analytical pipeline to improve those odds.

The same principle operates in attacking play. Viktor Gyökeres arrived at Arsenal from Sporting CP having scored 39 league goals in a single season [8]. His physical profile mapped to what Arsenal lacked: pace, power, willingness to run in behind. Yet by January 2026, observers described him as losing possession cheaply and squandering opportunities by taking too many touches [3]. Arsenal reportedly acknowledged before signing him that they would need to adapt their style to accommodate his profile [9]. The statistical scouting identified the right player for the role but could not predict the mechanical mismatch with the system.

Recent research provides the first empirical evidence that skeletal data captures what event data cannot. Van Haaren et al. (2025), analysing 1,736 one-on-one dribbles from the 2022–23 Champions League using Hawk-Eye SkeleTRACK data (29 anatomical landmarks at 25 Hz), found that defender posture and attacker balance were among the most impactful variables for predicting dribble outcomes — variables entirely invisible to standard event data [10]. This is the first published study using multi-camera skeletal match data for a performance prediction task in football, and it directly validates the thesis that mechanical information carries predictive power that action-level statistics do not.

6. The Data Exists. The Products Do Not.

The conventional assumption — that football lacks the data infrastructure for biomechanical assessment — is wrong. The infrastructure exists. What does not exist is anyone using it for this purpose.

Since 2022, every Premier League match has been captured by Genius Sports' optical tracking system at up to 100 frames per second, recording 29 skeletal points per player in three dimensions. For semi-automated offside technology (SAOT), the system generates up to 10,000 mesh data points per player — hundreds of millions of data points per match [11]. Since the 2025–26 season, every Bundesliga match has been captured by TRACAB's Gen5 system at 50 frames per second, recording 21 skeletal points per player in 3D, producing 140 million data points per match [12]. Hawk-Eye's SkeleTRACK captures 29 skeletal points at 50 fps across FIFA and UEFA competitions [13].

The comparison with basketball is instructive — but not in the direction commonly assumed. The NBA's Hawk-Eye installation, deployed across all 30 arenas from the 2023–24 season, captures 29 skeletal points per player at 60 fps [14]. The Premier League's system captures the same number of skeletal points at a higher frame rate. Football's top leagues are already collecting body-pose data at resolution comparable to — and in some cases exceeding — what the NBA provides.

The divergence is entirely in what is built on top of this data. In October 2025, the NBA launched three new public-facing statistics explicitly derived from pose tracking: Gravity (quantifying defensive attention and space creation), Shot Difficulty (analysing how stance and mechanics affect accuracy), and Defensive Box Score (attributing defensive impact in real time) [15]. Football has no equivalent. The skeletal data that could support biomechanical assessment of shooting mechanics, dribbling posture, tackling form, and running gait is collected every matchday in the Premier League and Bundesliga. It is consumed for officiating and tactical analysis. It is not consumed for the mechanical assessment that this publication argues is the missing variable in transfer prediction.

The Sport Performance Lab (SPL) at Maple Leaf Sports & Entertainment — parent company of the Toronto Raptors, Toronto FC, and the Toronto Maple Leafs — demonstrates what the downstream application looks like. The SPL employs a dedicated biomechanics data scientist whose role involves analysing "player body pose data" to "gain competitive advantages that best inform front office staff of player movement behaviours" [16]. The front office inclusion is the critical detail. This is biomechanics integrated into the player evaluation and acquisition process.

The Raptors' Shooting Lab — which won the 2026 NBA Team Innovation Award — uses the same underlying technology to capture 29 body points at 60 fps and analyse elbow velocity, release angle, stance width, and shot trajectory [17]. In November 2025, the NBA launched a league-wide biomechanics assessment programme: standardised motion capture labs at all 30 team facilities, over 500+ players assessed, using Qualisys, Theia Markerless, Bertec, and BreakAway Data, with P3 as consultant [18]. MLSE's SPL-Open-Data initiative has open-sourced markerless motion capture datasets to build the analytical community [19], and the SPLxUTSPAN Data Challenge on Kaggle invites researchers to predict free throw outcomes from biomechanical data [20].

Devin Pleuler, the Senior Director of R&D for Team Operations at MLSE and leader of the SPL, has observed at a research conference that with the expanded tracking data now available, the competitive moat has shifted. The moat is no longer the data — it is the physicists and biomechanists who can derive new insight from it.

His observation maps precisely to the structural argument of this publication. The cost of collecting biomechanical data in football has already been paid — the cameras are in the stadiums, the skeletal tracking is operational, the data is generated every matchday. What has not been built is the analytical layer that converts skeletal data into mechanical assessment: the framework that asks whether a player's execution quality predicts success in the target environment.

Sony's recent acquisitions sharpen the picture. Between October 2024 and October 2025, Sony acquired KinaTrax (markerless biomechanical motion capture, deployed in 75+ MLB stadiums) and STATSports (wearable GPS tracking, used by 800+ football clubs) to complement Hawk-Eye, which already provides skeletal tracking across 23 of the top 25 global sports leagues [21]. The company now owns every component required to build a biomechanical scouting product for football: optical skeletal capture, validated biomechanical analysis, and wearable physical data. Whether it will is an open question. That the components exist and are commercially assembled under a single corporate entity is a fact.

7. Why the Gap Persists

If football's data infrastructure already captures skeletal data at the resolution required for biomechanical assessment, why has no one built the analytical products?

The answer is paradigmatic, not technical.

Football's analytics community was built by mathematicians, computer scientists, and data scientists who professionalised the field through event data modelling. The intellectual heritage is statistical: expected goals, passing networks, possession value models. Biomechanics — the science of human movement — is a separate disciplinary tradition with different methods, different training, and different institutional homes. The football analytics community and the sports biomechanics community have developed largely in parallel, with minimal integration. The people who build the models and the people who understand human movement operate in different departments, report to different directors, and attend different conferences.

The NBA's advantage is organisational. MLSE's Sport Performance Lab sits across all their teams, integrating analytics, sports science, and biomechanics into a single research function with shared infrastructure and a common rapid-prototyping platform [22]. Football's typical organisational structure separates the data department (which does player evaluation) from the sports science department (which does physical assessment and injury prevention). The biomechanical expertise exists within football clubs — in the medical and sports science staff — but it is not integrated into the scouting and evaluation pipeline.

The DFL's Automated Event Detection system, which entered operational trials in the Bundesliga at the start of the 2025–26 season, represents the first attempt to build analytical products on top of skeletal tracking data in football [12]. Currently it detects on-ball events (passes, shots, set pieces) automatically from skeletal data. Off-ball event detection — including body-part identification and body orientation — is under development. This is a step toward biomechanical analytics, but it remains oriented toward event classification rather than execution quality assessment.

The clubs that bridge the organisational gap — making biomechanical assessment an input to player evaluation, not just a tool for injury prevention or officiating — gain access to a predictive variable that their competitors' models structurally exclude. The advantage is paradigmatic, not incremental. And for the first time, it does not require new data collection infrastructure. The data is already there.

8. What the Standard Metrics Collapse

Each mechanical dimension below decomposes an action that event data records as a single unit. In every case, the decomposition reveals information that the collapsed measurement destroys — information directly relevant to whether a player's capability will transfer to a new environment.

Striking: Why xG Cannot Predict Finishing Transfer

Expected goals measures the probability that a shot taken from a given location, in a given game state, results in a goal. It captures the situation. It does not capture how the body produced the shot. The kinetic chain involved in a football strike — hip generating the greatest moment, knee reaching maximum force at full extension, ankle applying spin and trajectory control [23] — produces measurable differences in execution quality. A compact release (short backlift, rapid contact) predicts the ability to shoot effectively under defensive pressure: the snap release that produces a goal in tight spaces in the Portuguese league will produce comparable goals in the Premier League, because the mechanical profile is suited to compressed time and space. A wide release (long backlift, full wind-up) predicts power in open situations but vulnerability to fast-closing defenders. xG treats both shots identically if they originate from the same location.

Shot preparation — the pre-execution sequence of creating separation from a marker, taking a directional set touch against the defender's momentum, and establishing an open stance for the strike — is itself a mechanical skill distinct from the shot. The player who can manufacture shooting angles through separation mechanics possesses a capability invisible to xG, which measures the shot location but not the mechanical process that created the opportunity. Football analytics has measured receptions and it has measured shooting. The unmeasured middle — how players manufacture the angle between receiving the ball and striking it — is a mechanical dimension with direct implications for how goal-scoring ability transfers between environments [24].

Dribbling: Why Carry Statistics Mislead

A "successful dribble" in event data is a binary outcome: the player advanced the ball past a defender, or they did not. This collapses three distinct phases — each with different mechanical demands — into a single result [25].

The preparation phase determines the range of available attack options: the player's orientation when receiving, their spatial relationship with defenders, the momentum they carry into the contest. The reception phase sets the foundation: first-touch direction, posture, ball proximity. The attack phase determines the outcome: direction changes, acceleration, feints, stride control. A player can be elite in one phase and limited in another — and the limitation is invisible in the collapsed statistic.

Manchester United's Alejandro Garnacho provides a case in point. He ranks in the 99th percentile for carries into the penalty area and the 93rd for progressive carries — elite at advancing the ball over distance. He sits in the 5th percentile for take-on success rate, despite attempting 5.71 take-ons per 90 (85th percentile) [25]. The carry statistics say world-class dribbler. The take-on statistics say the opposite. The mechanical analysis identifies where the discrepancy originates: his take-on mechanics resemble a template built for a different physical profile — more explosive in short bursts, faster from a standing start. The preparation and reception phases are strong; the attack phase encounters a mismatch between mechanical template and physical substrate.

Van Haaren et al.'s analysis of 1,736 dribbles confirms that the mechanical dimension carries predictive information the event record destroys. Attacker lean angle, defender body posture, and the relative spatial alignment between the two players at the moment of engagement — all skeletal-data variables, invisible to event tagging — were among the strongest predictors of dribble success [10].

The broader concept at work is techno-physical capacity — the interaction between a player's physical build and their technical execution range. Every athlete's physical structure determines a range of feasible motion. Intersecting that range with their trained perception-action patterns yields the set of actions they can consistently execute under competitive pressure. Scouting that evaluates technique without reference to the physical context in which it operates is evaluating a different variable than the one that determines performance in the target environment.

Defending: Why Tackle Statistics Produce Opposite Predictions

The defensive case was introduced in Section 5, but its analytical implication extends further. The standard tackle metric records an outcome: the ball was won or it was not. The mechanical assessment asks three questions the outcome alone cannot answer: From which angles and at what proximity can the player effectively contest? What happens post-tackle — is the ball retained, loose, or out of play? And what physical and mental exertion does the action require [7]?

These questions produce system-fit predictions that tackle statistics cannot. A player whose tackling mechanics are sprint-intensive will excel in a high-line system and struggle in a low-block. A player whose tackling mechanics allow ground-level challenges from multiple angles will excel in pressing traps and disruptive schemes. The tackle completion rate may be identical. The mechanical profile — and therefore the system-fit prediction — is entirely different.

Running: Why Distance Covered Is Structurally Incomplete

Tracking data captures distance covered per 90, sprint counts, and top speed. These are action-level measures of running. They do not capture purpose, sustainability, or contest readiness at the end of a run.

Running capacity decomposes into at least four tiers [26]: players who cannot sustain high-intensity sprinting across a match; players who can sprint but without covering purpose; players who run and cover, combining physical output with positional intelligence; and players who run, cover, and maintain aggression into subsequent duels. Modern pressing systems have made the highest tier a premium asset. Liverpool manager Arne Slot, after his side's 3-0 victory over Manchester United in September 2025, stated the requirement directly:

This capacity is mechanical: it depends on stride efficiency, energy management, deceleration patterns, and the neuromuscular ability to transition from running to contesting without degradation. A player's 11.5km per 90 may represent purposeless running or aggressive covering. The number is the same. The transfer prediction is not.

Spatial Control: The Contest Primitive

Across dribbling, tackling, aerial duels, and physical contests, a single mechanical concept recurs: the zone of spatial control around a player's centre of gravity, within which they can effectively manipulate the ball and resist encroachment [27]. The stiff arm — deployed across football, basketball, rugby, and American football — is a mechanical tool for expanding this zone. Players who wield it as a trained pattern possess a duel advantage that statistics capture as an outcome but not as a mechanism. The aerial contest dimension, examined in our prior research (SW-R-2026-001), similarly identifies the pre-contact phase as analytically significant — positioning and timing decisions that are mechanical, not tactical, and that the standard aerial duel metric structurally excludes [28].

The standard metrics record what happened. Mechanical assessment identifies the physical process that determined the outcome — and it is the physical process, not the recorded outcome, that predicts whether the capability transfers to a new environment.