Scientific transparency vs black boxes
Every time we compare gpexe with a new client’s previously used system, the request is always the same:
“I want to keep seeing the same number of accelerations.”
It is an absolutely reasonable request. At first, we also think the comparison will be straightforward: simply set the same thresholds on gpexe as those used by the other system and compare the results.
Instead, almost every time, something unexpected happens: the numbers do not match.
Evidently, there can be other reasons behind these differences. However, the real issue arises when we try to understand why. In some cases, even when applying the software’s declared thresholds to the exported data, the resulting number of accelerations does not match the one provided by the system itself.
How is this possible? It is precisely this question that prompted us to write this article. Because the real topic is not accelerations. The real topic is knowing what we are actually measuring.
Welcome to the black box
When results do not align, the first step is always the same: trying to reconstruct the calculation method.
- By what threshold is an acceleration identified?
- Is there also a duration threshold—meaning a minimum time during which the acceleration must remain above the threshold?
- What filtering is applied to the speed before the calculation?
These seem like trivial questions. In reality, very often, there is no answer.
Some software programs do not disclose all their calculation rules. Others do not even allow you to export the instantaneous data needed to verify the results independently. In other cases, despite having access to both the data and the declared parameters, the count is simply not reproducible.
It is important to clarify that we are not talking about a specific manufacturer. We have encountered situations of this kind when comparing various systems.
This, in our view, is the true “black box”: not knowing exactly how a number is generated, upon which daily decisions are subsequently made.
An objective measure must be reproducible
In scientific research, there is a fundamental principle: a measure is reliable only if it is verifiable and reproducible. If two people have access to the same data and apply the same calculation rules, they must obtain the exact same result.
Of course, every company is free to design proprietary algorithms and develop innovative metrics. However, this should not prevent the user from understanding what data is being processed, under what criteria, and what the metrics they observe every day actually mean.
Transparency does not limit innovation. It makes it credible.
gpexe web app: example of instantaneous data visualisation showing raw speed, filtered speed, acceleration, acceleration events, and bursts
The test of truth
When analysing the number of accelerations produced by your system, try answering these questions:
- Do you know precisely the threshold your system uses to identify an acceleration? (e.g., 2.5 m/s²)
- Do you know if a duration threshold is also applied? (e.g., 0.5 s)
- Can you export the instantaneous speed and acceleration data to independently verify the calculation?
- Do you have access to the raw data and know the filtering method used?
- Can you pinpoint the exact moment each acceleration was detected?
- Have you ever compared these events with the match video?
- Conversely, have you ever watched a video to count what you consider acceleration events manually, and then compared that number with the one provided by your tracking system?
If you answered “no” to even one of these questions, you are likely using a metric whose meaning you do not fully comprehend.
Our invitation is simple: demand transparency. Because the issue is not whether you measure 50 or 100 accelerations. The issue is knowing with absolute certainty what those numbers represent.
Transparency is the starting point
The next step is to ask ourselves whether the metrics we have been using for years are truly the most representative of athletic performance.
If you answered “yes” to all the questions in the previous test, you are probably ready for this shift in perspective: stop counting accelerations and start measuring bursts.
But that is another story. If you want to start exploring it, we invite you to read our dedicated articles:
- MPEs vs bursts – part 1: understanding high-intensity demands in team sports
- MPEs vs bursts – part 2: how metabolic and mechanical intensity really differ
- How easy is it to stumble over acceleration and deceleration?
- Acceleration: overcome its limits towards a new mechanical analysis.
Author: Valerio Roberti