The climb starts and your instinct is to protect the average. Ease back a touch, keep the number steady, save it for the flat bit after. It feels like the disciplined thing to do — the pacing-by-the-book move.

It's also the slow way up.

On any time trial with real hills or real wind, holding one flat wattage from start to finish isn't the safe strategy. It's the strategy that leaves free time on the table — time a rider can bank simply by moving watts from where they're cheap to where they're expensive, without doing one joule more work over the whole course.

Most pacing advice for amateur time-triallists still says the opposite: find your number, hold your number, don't chase the terrain. That advice is right for a flat, still course. It's the wrong lesson applied to the wrong course, and the gap between the two shows up as real, measurable seconds on race day.

Section 01What Swain's model predicts

In 1997, exercise physiologist David Swain built a mathematical model of cycling speed that accounted for the two forces that actually decide how fast a bike moves: aerodynamic drag and the pull of gravity on a gradient (Swain 1997). He then asked a simple question of the model — for a fixed amount of total work, is it faster to hold power constant, or to vary it with the terrain and still average the same number?

Run the same model on wind instead of hills — a 40 km time trial split into 5 km blocks of headwind and tailwind at 16 km/h — and the picture changes. Varying power with the wind still wins, but the margin shrinks to 29 seconds over a full hour: 60:21.2 versus 60:50.2, about 0.8% (Swain 1997). Terrain and wind are not interchangeable. Hills reward variable pacing far more than wind does, at least at these realistic parameters.

The physics behind this isn't exotic. Aerodynamic drag rises with roughly the cube of speed, but the pull of gravity on a gradient doesn't care how fast you're moving — it's a fixed cost per second regardless of speed. At low speed, almost none of your power is being spent fighting the air, so extra watts translate almost directly into extra climbing speed. At high speed, the opposite is true: drag eats a growing share of every additional watt, so pushing harder on a fast descent or a tailwind stretch buys comparatively little. Move watts from the fast, air-dominated stretches of a course to the slow, gravity-dominated ones, and you bank time without spending any extra energy across the full ride — which is exactly the trade Swain's model is calculating.

Section 02Real riders, real roads: does it hold up?

A model is a model until someone tests it on an actual bike. In 2011, twenty experienced cyclists rode a 4 km undulating course four times each — twice holding a constant power output averaging 253 W, twice varying power with the terrain at an average of 260 W. Normalised to a common 255 W to make the two conditions comparable, the constant-power trials took 411 ± 31.1 seconds and the variable-power trials took 399 ± 29.5 seconds: 12 ± 8 seconds faster, or about 2.9%, and the difference was statistically significant (p < 0.001) (Cangley et al. 2011).

A separate test reached the same conclusion from the wind side of the equation. Seven cyclists rode a simulated 16.1 km time trial with a headwind for the first half of the course and a tailwind for the second. Riding roughly 5% harder than their own average into the headwind and easing off with the tailwind beat both a flat constant-power effort and the riders' own self-selected pacing — and the deliberately variable strategy was the fastest of the three approaches tested (Atkinson & Brunskill 2000).

The real-world gain is smaller than the model's best case — undulating roads aren't a clean 10% grade alternating on a metronome, and no rider holds a perfect pacing plan for four kilometres. Wind gusts, gear changes, and pacing drift all eat into the theoretical number the moment a real rider is holding the bars instead of a spreadsheet. But the direction and the statistical significance both hold up outside the model, which is the part that matters: this isn't a modelling artefact that evaporates on the road. Twenty riders, a real undulating course, and a result that survived contact with reality at p < 0.001.

Section 03The protocol for a hilly or windy TT

1. Set your surge before the climb, not on it.

   Plan a modest power bump — roughly 5–10% above your target average — for the uphill and headwind sections, decided in advance, not chased mid-effort.
2. Ease off deliberately on the descent and the tailwind.

   The strategy only pays out if the recovery half actually happens. Riding the downhill hard "to bank extra time" cancels the gain from the climb.
3. Don't turn the surge into a redline.

   Push deep enough into a climb and you start drawing into the finite work capacity above critical power (Skiba et al. 2012) — spend it early and there's less left when the course asks for it again.
4. Know the course before you ride it.

   Variable pacing needs real gradient or wind information to execute. Guessing at the terrain ahead defeats the purpose.

Section 04Where this gets misread

Section 05Applying it with HELIOS

Climbing Mode reads the gradient of the climb ahead and sets a wattage target for it before you're on the slope — live grade, distance to the summit, and elevation gain folded into one number. That's the same call Swain's model makes with an equation instead of a screen: push a controlled notch harder while gravity, not aerodynamic drag, is the dominant cost, because the course is faster at the same average effort. The target is an estimate built from the gradient ahead, not a power-meter reading — it's there to set the plan before the climb, not to referee it.

That's also the piece most riders are missing when they try to do this by feel alone. The protocol above only works if you know the surge is coming before your legs do — a target set at the base of the climb, not a number you improvise halfway up it.

Section 06Counterpoint: where this stops working

Every number above assumes the surge stays inside a modest band — roughly 5–10% above target average, matching both Swain's model parameters and Atkinson's field test. Go deeper than that on a climb and you're no longer banking gravity's discount — you're drawing into the finite work capacity that sits above critical power, a store that has to be paid back before it can be spent again (Skiba et al. 2012). Spend too much of it on the way up, and there's less left for the finish, exactly when you can least afford it.

The size of the win is also entirely course-dependent. Swain's own model shows the gap directly: 6.4% predicted on the hilly course tested, against just 0.8% on the windy-but-flat one, at matched parameters (Swain 1997). On a genuinely flat, still time trial, there's no gradient or wind to exploit, and variable pacing has nothing to work with.

None of this works blind, either. Every study here assumes the rider knows what's coming — the gradient, the wind, the distance remaining — and adjusts in advance. Guess wrong about the terrain ahead and a deliberate surge just becomes an accidental overcook.

Section 07Bottom line

On a hilly or windy time trial, holding one flat number feels disciplined, but it isn't the fast option. Push a controlled 5–10% harder into the climbs and the headwinds, ease off on the descents and the tailwinds, and you cross the line faster at the same average power — a call a 1997 physics model made first, and a 2011 field test with 20 riders on real roads confirmed almost three decades later.

The discipline isn't in finding the surge. It's in planning it before the climb starts, and actually taking the discount on the other side.