What GPS still misses about football movement

Watch a top-level match with the tracking data beside it and something odd happens. The numbers that summarise a player’s evening (total distance, high-speed metres, peak speed) are almost never the moments anyone remembers. The goal-saving recovery run is in there. The sharp stop, the plant of the foot, the turn that loses a marker, the deceleration that kills a counter-attack: those are much harder to see.

GPS and its relatives, now usually combined with inertial sensors in the same vest, have transformed how clubs understand physical load. This piece is not an argument against them. It is an attempt to be precise about what they measure well, where their accuracy and meaning run out, and why the most mechanical parts of football remain difficult to measure for everyone, including the companies building alternatives. That includes BOUND.

What tracking does well

Start with the good news, because there is plenty of it. A review of 22 studies concluded that GPS units of all sampling rates measure total distance with adequate reliability, that 10 Hz units are the most valid and reliable, and that moving to 15 Hz added no further benefit ¹. In a direct comparison during straight-line running, 10 Hz units measured instantaneous velocity two to three times more accurately than 5 Hz units ².

A larger systematic review of wearable microtechnology in team sports, covering 72 studies, reached a more cautious but compatible conclusion: validity and reliability depend on the brand and model, the sampling rate, the type of movement and its intensity ³. In plain terms, modern tracking is good at “how far” and “how fast”, particularly when a club uses the same system consistently.

That is not a small thing. Distance and speed underpin a great deal of useful training planning.

Where accuracy runs out: the short, sharp actions

The trouble starts with change. Velocity is easy to measure when it is steady and harder when it is changing quickly. In a controlled test, the accuracy of 10 Hz GPS declined as acceleration increased, and was described as compromised above 4 m/s² ⁴. Older, lower-rate units struggled more with high-intensity running and runs involving a change of direction ¹.

The bigger problem is not raw accuracy but how short actions are counted. An “acceleration” in a tracking report is usually a count of efforts above a threshold, produced after the raw signal has been filtered and processed. Thornton and colleagues found that while distance and speed were consistent between units, threshold-based acceleration and deceleration counts differed substantially between manufacturers and between processing methods ⁵. In football data, Buchheit and colleagues reported that some units recorded two to six times more acceleration and deceleration occurrences than others, and that a manufacturer’s software update changed the counts produced from the same recordings ⁶. Reviews of the field note that sampling rate, device fit, satellite signal and filtering all affect outputs, with no consensus on how data should be processed or reported ⁷.

The practical implication is easy to miss. An acceleration count from one system, in one software version, with one set of thresholds, is not directly comparable with a count from another. Comparing a player’s numbers this month with last season’s can be misleading if anything in that chain has changed.

Derived metrics carry the same risk at a larger scale. Metabolic power, which estimates energy cost from speed and acceleration, has been influential and contested. In one soccer study it showed only a small correlation (r = 0.24) with metabolic power calculated from measured oxygen uptake, and substantially underestimated the energy cost of drills ⁸. That study used older 4 Hz units and a small sample, so we draw no verdict on the metric here beyond this: a derived number inherits every uncertainty of the signals beneath it.

Football is a braking sport

Here is the finding that, for us, frames the whole problem. A systematic review and meta-analysis of elite team-sport match play found that in soccer, high-intensity decelerations were more frequent than high-intensity accelerations, and that soccer showed the largest such difference of the team sports studied ⁹. The authors flag important caveats: the underlying data are tracking-derived, thresholds vary, and heterogeneity between studies was high. But the direction is consistent. Footballers stop hard more often than they start hard.

Stopping is mechanically demanding: the body has to absorb its momentum through the legs. A tracking system sees the change in speed, not the forces behind it. A study of 45 elite matches using accelerometer-based load suggested that position tracking alone may underestimate player load, because many intense actions happen without large changes in a player’s position on the pitch ¹⁰. We would treat that as a pointer rather than a proof, since the accelerometer metric itself is vendor-derived.

For performance staff, this cuts two ways. If hard decelerations are frequent and mechanically demanding, they are worth paying attention to. But the tracking-derived counts that describe them are exactly the numbers most sensitive to thresholds and processing, as the previous section showed. The action that may matter most is the one we currently measure least consistently.

Distance is not force

It helps to separate two questions. One is how hard the heart, lungs and muscles are working, the physiological load. The other is how much mechanical stress tissues such as bone, tendon and muscle are taking, the biomechanical load. Vanrenterghem and colleagues proposed that monitoring should treat these as separate pathways, with different measures and different adaptations ¹¹. It is a conceptual framework rather than an experimental result, but we find it a clarifying one.

Tracking systems sit mostly on the physiological side. Can the accelerometer inside the same vest fill in the mechanical side? So far, the evidence says only partly. In a lab study of running and cutting, accelerometers worn on the trunk, pelvis and shin related only weakly to whole-body loading measured with force plates, and generally overestimated it ¹². A later study tried to estimate ground reaction forces from the accelerations of body segments and found large errors, especially for decelerations and high-speed running, with errors growing as fewer segments were measured ¹³. A single unit between the shoulder blades is a long way from the ground.

Do insoles fill the gap? Not yet

This is where we have to be careful, because BOUND is building an insole. The obvious idea is to measure force where it is applied: under the foot. Some of the evidence is encouraging. A wireless insole agreed well with force plates during hopping, walking and running at 10 km/h ¹⁴. Another study found an insole accurate for contact time, impulse and peak force during running, but not for how quickly force rose, which the authors linked to the insole’s 100 Hz sampling ¹⁵.

Football is not walking or jogging. In a study of maximal sprinting, a pressure-sensor insole was generally not valid for vertical force or timing, agreeing only on step frequency ¹⁶. That study involved one athlete and one product, so it should not be over-read, but it points in a clear direction. And in the research we reviewed, we found no validation of pressure insoles in cutting or hard deceleration, which are exactly the actions that make football different. Pressure sensors respond to force pressing into them, so insoles of this kind estimate vertical force and do not directly measure the horizontal braking forces involved in stopping.

So the honest summary is that insoles are promising for some measures in some movements, and unproven for the movements football cares about most. BOUND’s own V1 prototype boards are currently undergoing bench validation. We have no validated result to report, and we will not claim to have solved this.

What this means for staff

None of this argues for less tracking. It argues for using each measure for what it is good at.

• Trust distance and speed more than counts. Distance, speed and peak speed from a consistent system are well supported. Acceleration and deceleration counts are more sensitive to brand, thresholds and processing, so treat changes in them cautiously.
• Know your thresholds. Ask which speed and acceleration thresholds define each band and count, and keep them fixed across seasons unless you deliberately reset your baselines.
• Keep the chain constant, and write it down. Same system, same thresholds, same processing, and a record of software versions. When something changes, mark it in the data.
• Read derived metrics as models. They can be useful, but they are only as good as the signals and assumptions beneath them.
• Keep the mechanical question open. How much force a player’s legs absorbed in a session, especially in braking and turning, remains genuinely hard to measure in the field. Anyone who says otherwise should be able to show validation in those movements, against force plates, in football conditions.

Governing bodies have started to formalise part of this. FIFA’s testing programme for electronic performance and tracking systems assesses positional tracking against reference systems, and FIFA is explicit that approving a device’s hardware is not an endorsement of the quality of the information it produces ¹⁷. There is, as yet, no equivalent standard for the forces behind football movement. That gap is the most interesting measurement problem in the game, and we think it is worth being precise about before anyone claims to have closed it.