Durability: Why You Fade Late in a Race and How to Train It

Two runners show up with the same VO2 max, the same threshold pace and the same measured economy, and then at 30 km one of them is holding pace while the other is coming apart, and nothing in the lab test predicted the gap. Durability is the quality that did predict it. Almost nobody trains it on purpose.

What durability means in endurance training

Durability is the resistance of your physiology to prolonged work. The research group that named it defined it as the time of onset and magnitude of deterioration in physiological profiling characteristics during prolonged exercise (Maunder and colleagues, Sports Medicine, 2021), which in plain terms asks how deep into a long effort your engine is still the engine you tested.

Andrew Jones went further. He argued the property, which he calls physiological resilience, belongs inside performance models as a fourth dimension next to the three classic determinants (The Journal of Physiology, 2024). Critical power after two hours of work falls by around 10% on average, but individual athletes have landed anywhere from about 1% to 31%, so two athletes can start a race with matching lab profiles and finish at opposite ends of that spread. No fresh test sees it coming.

One clarification first. Durability is not the same thing as tissue capacity, which is how much structural load your tendons, bones and muscles absorb before they break down, while durability is metabolic and neuromuscular: your power, pace, economy and thresholds holding their shape over hours. You can be durable and fragile at once. Or the reverse.

What falls apart, and when

The three classic determinants of endurance performance are VO2 max, exercise economy and the highest sustainable fraction of VO2 max, and together they account for roughly 72% of the variation in marathon performance (Hunter and colleagues, Experimental Physiology, 2025). All three are measured on fresh legs. All three move once you have been working for a while.

The cleanest running data comes from a study of 14 well-trained male marathon runners (VO2 max 63.1, marathon time 2:46) who were profiled fresh and then again after 90 and 120 minutes of prolonged running (Zanini and colleagues, Scandinavian Journal of Medicine and Science in Sports, 2025):

• Running economy deteriorated 4.2% at 90 minutes and 5.8% at 120 minutes. Same pace, more oxygen. The decline ran close to linear.
• VO2 peak fell 3.1% at 90 minutes and 7.1% at 120 minutes. Note the shape: the drop more than doubled in the second half hour. This one accelerates.
• Speed at lactate threshold dropped 3.0% and 6.6%. Threshold speed sits closest to marathon pace, so a 6.6% haircut on it is the difference between a negative split and a death march.

Cycling fills in the same picture. That average 10% loss of critical power after 120 minutes of heavy-intensity riding covers individual responses spanning 1% to 31%, and after roughly four hours of work, 20-minute and 6-minute time trial power dropped 2.9% and 10% on average. The individual spread was wide again: minus 8.5% to plus 1.1% on the 20-minute effort, and minus 31% to plus 1% on the 6-minute one (Hunter and colleagues, 2025).

That spread is the whole point. The group average is a footnote. Where you personally sit inside a range that wide is the number worth knowing.

Your lab numbers do not predict it

Here is the finding that should change how you read your own testing: in 20 well-trained male amateur racers tested fresh and again after 1,000 kJ of accumulated work, researchers found no significant correlation between durability and either VO2 max or relative FTP (Barsumyan and colleagues, BMC Sports Science, Medicine and Rehabilitation, 2025). The authors flag the small sample themselves, so read it as an early signal rather than a settled law. It points where the rest of the field points: an athlete's peak numbers say little about how much of them survives hour three.

The same research group compared seven successful amateur road racers, meaning a domestic podium or a UCI world championship qualification in the previous season, against seven who raced regularly without those results. The riders completed a 20-minute effort fresh, then rode at 70-80% of that power until they had accumulated 1,000 kJ of work, then repeated the effort. The successful group lost 6.5% of their 20-minute power. The less successful group lost 12.5%, close to double (Frontiers in Sports and Active Living, 2025). Five-minute power showed no group difference. The separation lived in the long, sustained effort after fatigue had set in.

Be careful with those two percentages. The groups were not matched: the successful riders were lighter, trained more hours and had a higher VO2 max, and the authors corrected for that with statistics rather than by design. Seven per group is thin. Trust the direction, not the decimal.

If you have ever tested well and raced badly, this is the mechanism, because your FTP or critical power number describes you at your freshest while races are contested by a version of you that shows up two hours later.

Read your own durability from a Garmin file

You do not need a lab. The field measure is decoupling, the drift between internal workload (heart rate) and external workload (pace or power) across a long effort, so if your pace holds while your heart rate climbs, your cost per unit of speed is rising and your durability is being spent. The same drift is what makes heart rate an unreliable pacing signal on race day, a separate problem with the same root.

The largest analysis of this ran on 82,303 recreational marathon runners (Smyth and colleagues, Sports Medicine, 2022), with decoupling calculated as the ratio of internal to external workload in the 35-40 km segment compared with the 5-10 km segment. The findings:

• Average decoupling was 1.16, and it first appeared at 25.2 km. The average runner had started drifting before 30 km, with a wide spread around that mean (SD 9.9 km).
• Low decouplers (under 1.1) held on until 33.4 km. High decouplers (1.2 and above) started drifting at 19.1 km. Fourteen kilometers of difference in when the wheels came off.
• Low decouplers finished in 217 minutes, high decouplers in 238 minutes. The durable group also raced at 88% of critical speed against 82% for the fragile group. More durable meant they could risk a harder relative pace and still hold it.

To run this on yourself, take a long run or ride of at least 90 minutes at steady effort, on flat ground, in stable weather, then compare heart rate divided by pace (or power) in the first third against the last third. The study band above puts a good result under 1.1, which is under 10% drift; coaches commonly work to a tighter 5% target on a steady, flat effort. Sit above 10% and durability is your limiter, whatever your threshold test says. Hills, heat and a fast start all inflate the number. Compare like with like, and track the trend across a block rather than obsessing over one file.

How to train it

The reassuring news: durability responds to training, and it responds to more than one kind. A 10-week randomized study put 35 sedentary and recreationally active adults into either low-intensity work at 6.8 hours per week or high-intensity intervals at 1.6 hours per week (Matomäki and colleagues, Frontiers in Physiology, 2023), and both groups improved durability, with no difference between them. In the interval group the size of the drift fell from 8.8% to 5.4% and its onset moved from 108 minutes to 137 minutes. The low-intensity group moved the same way, from 7.7% to 6.3% with onset going 106 to 131 minutes, without clearing statistical significance on its own.

These were untrained people, so the size of the effect will not carry straight across to a trained athlete. What carries is the shape of it: two wildly different weekly doses bought the same protection.

That does not mean volume is optional. Reviews of the field point toward work performed in the heavy and severe intensity domains as the stronger driver of durability, rather than raw accumulated kilojoules (Hunter and colleagues, 2025), so time spent above easy is what forces the adaptation. The long session is the platform you deliver it on.

What that looks like in a week:

• One properly long session, progressed slowly. Duration is the stimulus that exposes the decay in the first place, so add to it in small steps, not jumps. The old 10% per week rule of thumb works as a ceiling, not as a law, and the rest of the week has to stay honest. A weekly volume planner keeps the long session from quietly eating your recovery.
• Quality placed at the end of the long session, not the start. Coaches call these fatigue-state efforts: 2-3 blocks of 6-10 minutes at threshold after two hours of steady work. This is established coaching practice rather than a protocol with its own trial behind it, so treat it as a tool and watch how you recover from it. Earn it first: it belongs to an athlete already handling two-hour sessions comfortably, not to one whose long run is still growing. Once every 10-14 days is plenty.
• Keep the easy work easy. The base that lets you repeat long sessions comes from zone 2 work that leaves you able to train tomorrow, and durability is built across months of repeated exposure, which requires you to still be training in month three.
• Lift. In a 10-week randomized trial, 28 well-trained male runners added supplementary strength and plyometric work, and the strength group improved running economy durability across a 90-minute heavy-intensity run and substantially increased time to exhaustion at the end of it (Zanini and colleagues, Medicine and Science in Sports and Exercise, 2025). The runner's strength template covers what to lift and when.
• Track load honestly. Long sessions with late quality cost more than the duration suggests. A TSS calculator gives you a number to compare weeks against.

Fueling is half of durability

Much of the mechanism behind late-race decay is metabolic. The leading explanation: as type I fibres deplete glycogen, type II fibres take over more of the work, and they cost more oxygen per unit of force, so your economy falls because you have quietly changed which muscle is doing the job. How much fuel you have left, and how much carbohydrate you can take on mid-effort, sets how early that switch happens.

This is why durability testing insists on controlled fueling. In the amateur cycling study above, researchers standardized carbohydrate at 60 g per hour in a 1:1 glucose to fructose blend so that any drop in power was a physiological result rather than a fueling artifact. If you eat differently on two long sessions, you cannot compare their decoupling numbers and learn anything.

Race-day intakes across endurance sport have climbed well above 60 g per hour in recent years. How much your gut tolerates is itself trainable. Start with gut training and carb tolerance, then build the plan for a specific event with a race nutrition planner and the long session fueling guide.

Practice the intake in training at the rate you intend to race at. An unfuelled long run trains one thing. A fuelled one trains a different thing. Both have a place, but only one of them rehearses the race.

The mistakes that leave durability untrained

• Testing fresh and racing tired. A 20-minute FTP test or a 5k time trial on rested legs tells you about the top of your range, so add one fatigued test per block: the same effort after 90 minutes or 1,000 kJ of steady work. That second number is the one your race performance obeys. It is a hard session in its own right, so put it on a day you can recover from and keep it out of the last three weeks before a race.
• Chasing the peak number instead of the held number. Durability tracked neither VO2 max nor relative FTP in the cycling data above, and adding 10 watts to a fresh test buys you nothing if you shed 12% of it by hour three.
• Long sessions that never get long enough. The running study found economy and threshold speed decayed most between 90 and 120 minutes, so if your longest run stops at 75 minutes all year, you have never trained in the window where the problem lives.
• Stacking fatigue-state work every week. Late-session quality is a heavy stimulus. Every long ride ending in threshold blocks is how a good idea turns into a stale block. Once every 10-14 days, then let the adaptation land.
• Comparing decoupling across incomparable days. Heat, hills, a fast opening 10 minutes and a skipped gel all move the ratio, so standardise the route, the pacing and the fueling, or the trend you think you are reading is noise.
• Treating durability as a race-week fix. The physiology here is slow. Months of repeated long exposure moves it. Three weeks out from a marathon, your durability is already what it is.

How Movement Rebels fits

Durability is invisible in the single number most athletes look at. It is visible in the file underneath it. The Movement Rebels coach reads your Garmin or Apple Health sync and sees the last 14 activities with per-km splits, pace, cadence and average heart rate, so a long run where your pace held while your heart rate climbed reads as what it is instead of a good session. Post-activity analysis lands after each sync, not a week later.

It also sees your logged lifts and PRs, your morning readiness check-ins and your training intensity distribution. The Sunday plan is written against what those last 14 activities showed, so a week where your late-run economy faded gets planned differently from a week where it held. And if the single-leg and calf work has quietly dropped out of your strength log, you can tell the coach in chat to add that session back into the week, or have it rebuild the week around what is missing. Planned sessions push to your Garmin watch when auto push is on.

Add one fatigued test to your next block. Whatever that number says, it describes the athlete who will be on the start line.