Why preserving raw pressure data matters in wearable sports technology

About this paper. This is a technical paper written by BOUND. It has not been peer-reviewed. It sets out an argument supported by published measurement and validation research, and it describes a design position BOUND has taken. It does not report any validated BOUND result: V1 prototype boards are currently undergoing bench validation.

Abstract

Wearable sports technology reports numbers: load, contact time, force, asymmetry. Each of those numbers is the end of a chain that starts with a physical sensor and passes through electronics, firmware, filtering and algorithms. Validation frameworks from digital medicine and sport science already treat every link in that chain as part of the measurement ^{1 2}. For pressure-sensing insoles, the chain is unusually exposed. The sensors themselves change their output with loading history, sustained load, temperature and mounting ^{3 4 5}. Commercial systems differ widely in accuracy ⁶. Insoles can be reliable while being systematically biased ⁷, and their validity has not yet been systematically reviewed ⁸. Elsewhere in sport, a software update has changed the metrics produced from unchanged data ⁹. We argue that, in this situation, pressure wearables should preserve sample-level data, together with the version and calibration state that produced each result. Without that record, errors cannot be corrected, results cannot be independently checked, and improvements cannot be applied to past data.

1. Background

Most performance staff never see what a wearable actually measured. They see what it reported. Between the two sits a series of decisions: how the sensor was sampled, how the signal was filtered, which events were detected, which thresholds were applied and which model converted the result into a familiar unit.

The V3 framework for biometric monitoring technologies describes this as a data supply chain. It separates verification, which asks whether the sensor’s sample-level output is correct, from analytical validation, which asks whether the algorithm turns those samples into an accurate metric ¹. The same paper notes that “raw data” means different things at different points in that chain, and proposes “sample-level data” as the consistent unit to evaluate ¹. We use that meaning here: the sensor values as they leave the device, before the processing that turns them into reported metrics.

In sport, Düking and colleagues make the practical version of the same point. Their recommendations for evaluating wearable data ask manufacturers to provide access to raw data, and ask evaluators to report firmware and software versions, sampling and filtering ². Reviewing GPS and inertial devices, Malone and colleagues note that many metrics reach practitioners without clear explanation of how they are obtained, and that processing choices affect them ¹⁰.

2. The argument

Our argument is simple. If a reported metric depends on sensor behaviour that changes over time and on processing that changes between versions, then the only durable record of what happened in a session is the sample-level data, stored with enough context to reproduce any result derived from it.

Three consequences follow:

1. Correction. When a calibration or algorithm is found to be wrong, past sessions can be reprocessed rather than discarded or left wrong.
2. Independent checking. Researchers and club analysts can compare a device against reference equipment using what the device actually recorded, not what an earlier version of its software made of it.
3. Transparency. Any number shown to a coach can be traced to the readings and the version of processing behind it.

None of this is new as an idea. Physical-activity research went through a similar shift, from proprietary activity counts toward analysis of raw acceleration signals and shared methods ¹¹. We think pressure-sensing wearables in sport are at an earlier point on the same path.

3. Evidence

3.1 The sensor is not a fixed ruler

Many low-cost pressure insoles use resistive sensing elements, including force-sensitive resistors. Their output is not a fixed function of the force applied.

Bench studies show three effects. First, history matters. Hall and colleagues needed a calibration model that included the sensor’s previous outputs to correct hysteresis and drift, and found that prolonged shear loading changed sensor gain by roughly 100% ³. Second, sustained load matters: resistance creeps under constant force, and the drive voltage used by the circuit affects how much ⁵. Third, environment and mounting matter: temperature, sensor curvature and the compliance of the surface the sensor sits on all shifted calibration curves, leading the authors to recommend calibrating each sensor individually in conditions close to its use ⁴.

These studies come from hand-force and prosthetic applications, not football insoles. We treat them as evidence about the sensing technology, transferable by analogy, not as measurements of any insole product.

3.2 Systems differ, and validity is not settled

Comparisons of commercial in-shoe systems show how much results depend on the system. Under the same bench loading, Price and colleagues found root-mean-square errors of 2.6 kPa for one system and 27.0 kPa for two others, with regional contact-area differences ranging from −97% to +249% ⁶. An earlier technical assessment of five devices found that a resistive system reached good accuracy only after a complex ad hoc calibration ¹². Even a well-performing capacitive system showed average errors under 10%, with error rising at higher pressures ¹³.

Wireless insoles compared directly with force plates add a further point. In healthy adults performing functional activities, insoles under-read vertical ground reaction force while showing fair-to-excellent test-retest reliability ⁷. A device can repeat the same error consistently. Reliability is not accuracy.

Finally, the evidence base is incomplete. A 2025 protocol for a systematic review states that the validity and reliability of wireless pressure insoles have not yet been systematically evaluated ⁸. Most validation work so far uses walking or simple tasks in the lab, not football movements on grass.

3.3 Processing and versions change the number

We found no published study showing a firmware update changing pressure-insole outputs. The clearest evidence of version effects comes from GPS in football. Buchheit and colleagues reported that between-unit variation ranged from 1% for maximal speed to 56% for the number of decelerations above 4 m/s², and that a manufacturer software update produced large decreases in counted accelerations derived from the same recordings ⁹. The physical movement did not change. The software did.

The general lesson, that filtering and processing decisions change reported metrics, is recognised across wearable research ^{10 2}. We see no reason to expect pressure-derived metrics to be immune.

4. Implications

4.1 What preserving raw data should mean

In our view, preserving raw data in a pressure wearable means at least four things:

• Sample-level data kept unchanged. The sensor values as recorded, with timestamps and sequence information so that missing data can be detected rather than silently filled.
• Derived results stored separately, never overwriting the samples they came from.
• Version metadata on every result. Which firmware, which processing version and which calibration produced it.
• Calibration state recorded. Including when no validated calibration exists, in which case raw values should be presented as raw values, not as forces.

4.2 BOUND’s position

BOUND’s development pipeline was designed this way. Its insole streams raw sensor readings with timestamps and sequence numbers. Processing keeps those readings separate from derived results and records which processing version produced each result. No calibration model from raw readings to force has yet been validated, so BOUND shows no force values. This pipeline is in development and has not yet been used to record real players. These are design decisions. They are not evidence that any BOUND measure is valid; that evidence can only come from comparison against reference equipment, which is the next stage of our validation work.

4.3 For clubs and researchers

For anyone evaluating a pressure wearable, the practical questions follow directly. Can you get the sample-level data? Is each result tagged with the version of processing that produced it? Is the calibration method described, and how was it validated, on what population and in which movements? These questions align with existing recommendations for evaluating wearable data ².

5. Limitations

This paper has several limitations. It is not a systematic review; sources were selected to test a specific argument. Several of the cited studies were assessed from their abstracts. The sensor-physics evidence comes from bench work outside football insoles. The evidence for version effects comes from GPS, not pressure sensing. Preserving raw data has costs that we have not quantified, including storage, data protection and the discipline of versioning every processing step. Finally, keeping raw data does not by itself make a measurement valid. It makes validation, correction and scrutiny possible.

6. Conclusion

Pressure-sensing wearables sit on sensors whose output drifts and depends on context, on processing that changes between versions, and on a validation evidence base that is still being built. In that situation, the reported metric is the least durable part of the record. The sample-level data, kept with its version and calibration history, is what allows a number to be corrected, checked and trusted later. We think that should be the default for the category, and it is the standard we are holding ourselves to.