Your pacing plan stops working at hour three
A calculator anchored on threshold hands you the same number for ninety minutes and for six hours. On a long day the limit is not your threshold. It is how your power holds.
2026-09-20 · 8 min read · Brevet · AI-assisted
Drafted with AI assistance. A human review of this version is still pending — we say so rather than imply one. The science is sourced and real; the article isn't medical advice.
The short version: A lot of pacing calculators start from your threshold and give you a slice of it — the same slice whether the day is ninety minutes or six hours. Threshold is a fresh-legs number. What limits a long day is how your power holds as the work piles up, and that lands far lower than people expect. We know, because the first pacing engine we built made exactly this mistake.
The second climb
Kilometre sixty. The second big climb of the day starts, and the sticker on your stem says 250 watts.
Your legs say 215.
You didn't go out too hard — you rode the plan. You ate. It isn't hot. The number is simply not there, and there are eighty kilometres left.
This is the most common way a long day falls apart, and it is rarely a fitness problem. It's an arithmetic problem, committed weeks earlier, at a desk, by whatever produced that sticker.
We built the wrong plan first
We have an engine that predicts how fast you'll ride a lap. It works the way pacing tools commonly work: it takes your Critical Power — close cousin of FTP, the power you can hold for the better part of an hour — and spreads effort around it. For a time trial or a thirty-minute climb, that's the right anchor.
Then we asked it about a gran fondo. 140 kilometres, 2,790 metres of climbing, a rider with a CP of 260 watts.
It answered: 258 watts. Ninety-nine percent of CP. For five hours.
The "optimised" version was worse. It spent 228 minutes above CP and ran the rider's reserve down to zero on the way. No error, no warning. Just arithmetic doing what it was told, and a finishing time of 4 h 56 that nobody on earth with that CP could ride.
Nothing in the maths was broken. The anchor was. CP describes about an hour of fresh legs. Ask it about hour five and it keeps giving you the answer for hour one.
The limit on a long day has a name
Sports science calls it durability: the things a lab measures on fresh legs — threshold, efficiency, the power at a given heart rate — drift during prolonged exercise, and they drift differently in different riders. Maunder and colleagues made the case in 2021 that a profile without it describes an athlete who only exists for the first hour.
So the pacing card we ship doesn't start from threshold. It starts from how your power holds. For a rider in the middle of the range that means about 83 % of CP in the first hour, 76 % after it — and less again once the day runs past three hours, and past five.
For that gran fondo: an intensity factor of 0.70 and 6 h 11 instead of 4 h 56. On paper that looks like giving up an hour and a quarter. In practice it is the difference between a plan and a wish. The 4 h 56 was never on offer.
Now look at the gold line again. It's a staircase.
Your body does not fall off a ledge at 3:00:00. Our model does, because it has three duration brackets and we haven't earned a smooth curve yet — that needs data on how riders fade between hours three and five, and we would rather collect it than invent it. The staircase has a consequence we only found by testing our own tool. Take a flat 96.5 km route and that same 260-watt rider. At the under-three-hours level they finish in 2 h 56. At the over-three-hours level, 3 h 01. Both answers agree with themselves. A calculator that iterates lands on whichever one it happened to start nearer to.
Ours now walks in from both ends. Where the two disagree it takes the easier plan, and prints on the card that it did. A plan that turns out slightly too easy is a good day out. The other kind is the second climb.
Power belongs on the climbs. Some of it.
One number all day is not the fastest way to spend your energy, and this part is old news. Swain modelled it in 1997: vary power by 10 % with the gradient and a hilly 10 km time trial drops from 24:20 to 22:47. Atkinson, Peacock and Passfield repeated the exercise in 2007 with a better-validated model: 126 seconds saved by a 289-watt rider on a course of alternating one-kilometre ramps — and the largest savings for the riders with the lowest power who could vary it the most.
The reason is plain physics. Uphill you are slow, almost all your power goes into lifting you, and every extra watt buys real time. Downhill you are fast, the air takes your watts to the third power, and pedalling harder at 60 km/h buys you almost nothing.
So why only fifteen percent? Why not thirty?
Because variable power isn't free. Normalised Power — the common yardstick for what a ride cost you — averages power to the fourth. A second ridden 15 % above your level is priced at 1.75 times a normal second. At 30 % above, it's 2.9 times.
The card moves power toward the climbs by up to 15 %, backs right off on the descents, and then scales the whole plan until its Normalised Power equals the level you can hold. That last step is the one that matters. Without it, "more on the climbs" quietly becomes "harder overall", and you are back at the second climb.
One honest caveat: Normalised Power is a model, not a law of nature. The fourth power is its author's fit to how variable efforts cost riders — useful, widely shared, and not the last word. We use it because everyone can check it. We don't pretend it's physiology carved in stone.
What your reserve actually buys
Above CP you draw on a finite reserve, usually written W′; Skiba and colleagues gave us the model for how it drains and refills. We assumed a bigger reserve would mean a faster finish, and wrote a test that said so.
The test was wrong.
We paced a two-hour route with two steep three-kilometre walls, ridden all-out, for a rider with a small reserve (8 kJ) and a large one (40 kJ). The large reserve spent 1,714 seconds above CP, the small one 793. The difference at the line: ten seconds in two hours.
Once a plan is held to one Normalised Power, a wall that has to be capped at CP hands its watts to the flat. Your reserve decides where the watts go. It barely decides when you arrive. The card never spends more than half of it, because a plan that uses up the reserve it was given has stopped being a plan.
What no card can know
The wind. The heat. Whether you'll sit in a group for forty kilometres or ride alone into a valley headwind. How you slept. Whether the second feed station has run out of water.
So the card gives a time range, not a time. Targets are bands, not points — a single number on a top tube gets chased. Descents are capped at 55 km/h, because corners set your speed downhill, not terminal velocity. It assumes calm air and twenty degrees, and says so.
And there's a limit on our side. For a member, how their power holds is read from their own long rides — which takes real efforts of two hours and more before it means anything. The public card has no history to read, so it asks you, offers four answers, and defaults to the middle one rather than the flattering one.
Try it on your route
The pacing card takes the GPX of your event, your CP and your weight, and gives you watt bands per climb, elapsed-time checks and a range for the finish, laid out to print for your top tube. No account. Your browser reads the file and sends only the shape of the road — distances and elevations. Where the road is never leaves your device.
Pace the day you can finish. The number for that is lower than your threshold, and it is supposed to be.
Where the numbers come from
- Maunder E, Seiler S, Mildenhall MJ, Kilding AE, Plews DJ. The Importance of "Durability" in the Physiological Profiling of Endurance Athletes. Sports Medicine 2021;51(8):1619–1628. PubMed
- Swain DP. A model for optimizing cycling performance by varying power on hills and in wind. Medicine & Science in Sports & Exercise 1997;29(8):1104–1108. PubMed
- Atkinson G, Peacock O, Passfield L. Variable versus constant power strategies during cycling time-trials. Journal of Sports Sciences 2007;25(9):1001–1009. PubMed
- Skiba PF, Chidnok W, Vanhatalo A, Jones AM. Modeling the expenditure and reconstitution of work capacity above critical power. Medicine & Science in Sports & Exercise 2012;44(8):1526–1532. PubMed
- Normalised Power: Allen H, Coggan A. Training and Racing with a Power Meter. VeloPress.
The figures about our own engine are from runs on 20 September 2026 on a synthetic 140 km route, both plans computed for the same rider in the same air; they describe our software, not a study.
A note on what this is
Brevet is a training and coaching tool, not a medical device. A pacing card is a calculation on the numbers you enter and the shape of a route — a performance estimate, never a health assessment and never a promise of a finishing time. The targets are bands to orient yourself by. How you ride is your decision.
Our methods are written up in plain language on the Science page, with the limits stated honestly — including where a model of ours has steps it should not have.