Durability: Why Your Fresh FTP Lies About Hour Three

Durability is how well your physiology holds up as work accumulates. Every number you know about yourself -- FTP, threshold, VO2max -- was measured on fresh legs, and none of them stay put. Critical power falls by around 10% after two hours of hard riding, but the individual range runs from under 1% to about 32%, which means two athletes with identical lab numbers can be completely different athletes at hour three. Among amateur cyclists, the successful ones lose 6.5% of their 20-minute power after 1000 kJ while the less successful lose 12.5%. The uncomfortable part is how little we know about training it: of the methods everyone recommends, only heavy strength training has been tested directly and shown to work.

The Number That Moves

What Actually Decays

The classical model of endurance performance has three pillars: VO2max, the fraction of it you can sustain, and how economically you move. All three are measured in a laboratory on a rested athlete, and all three were quietly assumed to be stable properties.

They are not. Andrew Jones, reviewing the evidence in 2024, proposed resilience as a fourth dimension precisely because the other three degrade -- and degrade at wildly different rates between people. His headline figure: critical power falls by an average of 10% after two hours of heavy-intensity cycling, but individual changes range from under 1% to roughly 32%.

Sit with that range for a moment. Two riders finish a lab test with the same critical power. Two hours later, one has lost almost nothing and the other has lost a third of it. No fresh-state test would have told them apart.

Two Things Fall, and They Fall Differently

Clark and colleagues put nine cyclists through two hours of heavy riding and then re-measured their power-duration relationship. End-test power -- effectively critical power -- dropped 8%, from 306 to 282 W. But the finite work capacity above it dropped 20%, from 18.3 to 14.7 kJ.

That second number matters more than it looks. Your anaerobic reserve is what you spend on climbs, surges and finishes. Losing a fifth of it means the attack you could cover at the start of a race is the attack that ends your day four hours in.

Why It Happens

Several mechanisms run in parallel. Muscle glycogen depletes, and as the body shifts toward fat oxidation the oxygen cost of the same power rises -- fat yields less ATP per litre of oxygen than carbohydrate. Slow-twitch fibres fatigue, so the nervous system recruits higher-threshold fibres that are less efficient and more glycolytic. Neuromuscular fatigue accumulates both centrally and peripherally. Plasma volume falls with sweating and blood is diverted to the skin, so stroke volume drops and heart rate climbs to compensate.

The practical consequence is that the intensity domains shift underneath you. The power that was comfortably moderate at the start creeps into the heavy domain by hour three without you changing anything. You did not slow down; the boundary moved.

Is It Really Independent?

This is where honesty is required. Spragg and colleagues found that retention of critical power correlated strongly with relative VO2max (r = 0.835), gross efficiency (r = 0.869) and inversely with carbohydrate oxidation (r = -0.702). On that basis durability looks less like a separate quality and more like a consequence of having a bigger aerobic engine.

But that study had ten riders, all professionals, mean age 19.2. With a sample that small and that homogeneous, a correlation of 0.835 carries an enormous confidence interval, and the finding may not generalise at all to a 40-year-old riding nine hours a week. Studies in amateur populations have generally failed to find a relationship between durability and fresh VO2max or FTP.

The reasonable position: among elite athletes who all have big engines, durability may largely track the engine. Among the rest of us it appears to be its own axis.

Running Is the Same Story

Hunter and Muniz-Pumares tested 18 London Marathon runners -- average age 41, average finish 3:17, in other words people like the readers of this article. After just 90 minutes of running, VO2peak fell from 56.7 to 53.4 mL/kg/min and speed at lactate threshold fell from 12.8 to 12.1 km/h.

Convert that: threshold pace went from 4:41/km to 4:58/km. Seventeen seconds per kilometre, after ninety minutes. And the size of that drop correlated with actual marathon finishing time (r = 0.680) -- better than fresh-state laboratory numbers managed.

Example: Testing Yourself, and What the Answer Means

The amateur study used a protocol you can run on a trainer. Take a rider with a fresh 20-minute power of 250 W.

The test:

At roughly 180 W, accumulating 1000 kJ takes about 93 minutes, so the whole test runs a bit over two hours.

What the second number tells you:

Both riders own a 250 W FTP. On a flat 40 km time trial they would finish close together. On the bike leg of a 70.3 -- roughly 1600 kJ of work -- they are not the same athlete, and the second rider will start the run considerably worse off.

What it changes about pacing. If you know you belong to the 12.5% group, setting race power from your fresh FTP is a mistake with a predictable ending. Start below the number your fresh test suggests and let the negative split come to you.

What it changes about fuelling. This is where the research gets specific. Clark's 2019 study found carbohydrate ingestion completely negated the drop in end-test power at two hours -- but had no effect at all on the work capacity above it, which fell regardless. So eating properly protects your sustainable power. It does not protect your ability to surge. Fuel aggressively from the first minutes, and separately accept that your matches get damper as the day goes on, whatever you eat.

Practical Rules

1. Stop treating your FTP as a property of yourself. It is a measurement taken under one specific condition -- rested -- and your race will not be run under that condition. For anything over about two hours, the relevant number is the one you can produce late, not the one you produced fresh.

2. Test it before you try to train it. The 1000 kJ protocol above costs you one long ride and gives you a real number. Repeat it in the same conditions -- same trainer or same route, similar temperature, same fuelling -- or the comparison is worthless.

3. Lift heavy. It is the only method with direct evidence. In the trial that tested this properly, 19 female duathletes did four lower-body exercises at 3 x 4-10RM twice a week for 11 weeks. The strength group improved 5-minute power measured after three hours of cycling by 7.0%, and after 1.5 hours of running by 4.7%. The endurance-only group did not improve at all. Note that the test was deliberately conducted under fatigue -- this is evidence about durability specifically, not about fresh power.

4. Treat the popular methods as plausible, not proven. Intervals at the end of a long ride, high aerobic volume, block periodization, low-cadence work -- all of these are recommended constantly and all of them have a coherent physiological rationale. None has been tested in a controlled trial measuring durability before and after. That does not make them wrong. It means you should not abandon something that works for you on the strength of a mechanism story, and you should not expect a guaranteed return either.

5. Do not starve yourself in pursuit of it. Training with low carbohydrate availability is often justified as durability work. The evidence for benefit is thin, and the risks are real: relative energy deficiency, disrupted hormonal function, and compromised bone. The bone data most often cited comes from academy footballers in a single session rather than endurance athletes, so treat it as a warning flag rather than a measurement of your risk -- but a flag is still a flag. Women and athletes over 40 carry the highest exposure here.

6. Watch decoupling on your long rides as a free proxy. When heart rate drifts upward while power stays flat, that drift tracks the underlying decline in your first ventilatory threshold; a 51-cyclist study built a working prediction model from exactly these markers. It is not a substitute for a real test, and heat, dehydration and hills all inflate it -- but it costs nothing and it moves in the right direction.

7. Runners: measure the last third, not the average. Hold a fixed heart rate for the first 5 km of a long run and again for the last 5 km, and compare the pace. The gap is your durability. The marathon study found that the decline in threshold speed after 90 minutes predicted finishing time better than any fresh measurement did.

Evidence Base

Durability is a young field. The term was formalised in 2021, most of the primary work has appeared since, and there are as yet very few systematic reviews. That shapes what can honestly be claimed.

What is well established. That physiological parameters decline with accumulated work is beyond dispute and has been shown repeatedly. Clark et al. (2018) demonstrated the shape of it: 8% off end-test power and 20% off the work capacity above it after two hours, in nine cyclists. Clark et al. (2019), in 16 cyclists, refined the timeline -- nothing measurable at 40 or 80 minutes, a 9% loss by two hours -- and separated the effects of fuelling.

That it discriminates performance is also well supported. Muriel et al. (2022) showed that WorldTour and ProTeam riders separate under fatigue and not before, while repeatability of near-maximal efforts did not differ between them at all. Hunter and Muniz-Pumares (2025) found the change in threshold speed after 90 minutes of running correlated with marathon time at r = 0.680. Barsumyan et al. (2025) found the 6.5% versus 12.5% split among 14 amateurs -- and notably found no group difference in the 5-minute effort or in heart rate, so the discriminating signal was specifically the sustained effort.

What is genuinely thin: how to train it. A search for controlled trials that measured durability before and after an intervention returns essentially one positive result. Vikmoen et al. (2017) randomised 19 well-trained female duathletes and tested them after prolonged submaximal work, which makes it a durability outcome rather than a fitness outcome. Everything else in the standard advice rests on mechanism and coaching consensus. This is worth stating plainly because the gap between how confidently durability training is discussed and how much has actually been tested is unusually wide.

Three limitations deserve attention.

Sample sizes are small. The studies above involve 9, 10, 14, 15, 16, 18 and 19 participants. Correlations from samples of that size are unstable, and the strong relationships Spragg et al. reported between durability and VO2max come from ten riders with a mean age of 19.2 -- an exceptionally narrow group.

Sex differences are unresolved and the data conflict. Laboratory work has found women losing less power than men over prolonged exercise, while field analyses of professional cyclists have found the opposite pattern. Neither picture can currently be relied upon, and the underlying reason for the discrepancy is unknown.

Athletes over 40 are barely studied. There is essentially no controlled evidence on durability in masters athletes, despite this being the group most likely to read an article like this one. Recommendations aimed at that population come from coaching practice, not trials.

One measurement note. Quantifying prior work in kilojoules alone is known to be insufficient, because the intensity of that work matters as much as its volume -- the same 1500 kJ accumulated through hard intervals and through steady endurance riding do not produce the same subsequent decline. Any comparison of your own tests needs the preceding work to match in intensity distribution, not just in total.