Published Oct 5, 2026
by Markus Rummel. Last month I presented our durability research at the Science in Cycling symposium in Bromont, Quebec. The question behind the talk: can you measure how well an athlete holds up over a long ride from the data they already record in training, without a lab test? In other words, can you measure your cycling durability from everyday rides? The short answer is yes, and this post explains how, what we found, and why it matters for your training.
You can download the slides here (PDF). A full paper based on the talk will appear in the Journal of Science and Cycling; we will link it here once it is out.
Durability is how much, and how early, your physiology deteriorates during prolonged exercise [1]. Two riders with the same threshold power can look very different after three hours: one still rides at the same effort, the other has slowed down or has to work much harder for the same power. In long races that difference often decides the result.
Researchers measure durability in two ways [2]:
So far, almost all of this research comes from lab protocols (ride for hours at a fixed power, test before and after) or from professional race data. Lab protocols are one-off snapshots and nobody repeats them every two weeks. Amateur athletes, the vast majority of us, are hardly studied at all.
We used training data from 105 amateur cyclists and triathletes who consented to research use of their data: 4,850 rides recorded with a power meter and a chest strap clean enough for beat-to-beat heart rate analysis (fewer than 5% artifacts). Accumulated work is measured in kilojoules per kilogram of body weight (kJ/kg); for a 77 kg rider, 15 kJ/kg is about 1,150 kJ, roughly two to three hours of solid endurance riding.
Within each ride we looked for 5-minute stretches at a meaningful intensity, at least 85% of the athlete's critical power (CP). We then compared stretches ridden at the same power (within 10%) early and late in the ride. If the power is the same but heart rate is higher later, that difference is durability, measured with no special test.
At the same power, the internal cost rises steadily. Comparing each ride's first hard 5 minutes with later stretches at the same power, after 15 kJ/kg heart rate was 10 bpm higher, DFA alpha 1 was 0.17 lower and breathing was 4 breaths per minute faster (median across athletes). Power itself barely moved, so this is not an intensity effect.
We also compared all pairs of same-power efforts within a ride, a design borrowed from the professional race studies [3, 7] but applied to heart rate and breathing instead of power. Heart rate drift was about 7 bpm at both 15 and 25 kJ/kg of work between the two efforts, and in line with what a recent lab study found after 2.5 hours of riding [6].
It is robust. The drift showed up whether we defined hard efforts as above 85%, 95% or 105% of CP, and when we matched power to within 2% instead of 10%. Rerunning everything on four months of newer data gave the same picture.
It is repeatable per athlete. Splitting each athlete's rides into two halves, the heart rate drift measure agreed well between the halves (correlation 0.72). Differences between athletes were larger than the measurement noise, so the measure tells athletes apart.
The obvious alternative is to check how much your best power drops late in rides. For pros that works, because racing forces maximal efforts after hours of work. For amateurs it mostly does not. Late in a training ride most of us do not go all out, so a lower late-ride power usually means we chose to ride easier, not that we could not ride harder. In our data this power-based measure was much less repeatable (correlation 0.37). Physiological durability only needs you to ride at a steady effort, which you do anyway. For amateur athletes it is not just the convenient way to measure durability; it is the one that works.
Because the measurement needs no test, it can be updated every two weeks from your normal training. Below is one athlete over 20 months: the durability estimate changes threefold within a season. Part of that is training, part is heat and accumulated fatigue, which is why the most useful comparison is you against your own baseline, not against other athletes.
If you train with a power meter and a chest strap, you are already recording everything needed to measure your durability. We are building this into AI Endurance so your training plan can track and target durability alongside your thresholds. Questions or thoughts? Join the discussion on our forum.
This work builds on our earlier study with Stefano Andriolo and Thomas Gronwald, relating cycling power and DFA alpha 1 from everyday workout data [8].
Get your AI Endurance best training plan into TrainingPeaks. From there, you can execute your TrainingPeaks workouts in Zwift and many other apps. Connect your AI Endurance account once and any changes will automatically be synced with TrainingPeaks.
by Stefano Andriolo. Building on previous work, we refine a method to accurately determine the relationship between DFA alpha 1 and power. This method can be used to track fitness and thresholds of an athlete. We find in some cases ramp detection tends to overestimate thresholds, a finding mirrored in recent physiological papers. On the other hand, thresholds based on clustering of DFA alpha 1 values tend to agree well with this new method. We propose a hybrid lab and everyday workout experiment to further study the relationship.
Execute your AI Endurance personalized training plan directly from your Garmin watch or bike computer. No more writing down workouts or remembering interval sets. Get step-by-step instructions as Garmin custom workouts and a Garmin Connect training plan with only a few clicks.
We explain how AI Endurance uses ChatGPT to help guide your triathlon, running and cycling training.
