From Muscle to Movement: The Engineering Chain Behind Your Zwift Speed

Between the contraction of a muscle fibre and the avatar moving on your screen, your effort crosses six or seven interfaces. Every one of them takes a cut, and one of them decides what number you are shown.

This is the part of indoor cycling that mechanical engineering is actually useful for, and it is the part riders most often get wrong when their numbers disagree with someone else's. If you have ever wondered why your trainer and your power meter differ by a few percent, why low cadence feels so different at the same wattage, or what exactly Zwift is doing with the figure it receives, this is the chain that answers all three.

Two companion articles cover the neighbouring ground: the physics of what makes you fast once the power exists, and how this site's prediction models were fitted. This one is about the machinery in between.

The chain, end to end

Laid out in order, with what each stage does to the signal:

The last row is the important one. Zwift does not know what you actually produced. It knows what it was told, and it treats that as fact. Everything downstream of the measurement, including every estimate on this site, inherits whatever error was introduced at that point.

Torque and cadence: the same watts, very different loads

Power is not something your legs produce directly. What they produce is force on a pedal, which becomes torque about the bottom bracket, and power is that torque multiplied by how fast the cranks are turning:

Run 250 W through that at two different cadences and the difference is larger than most riders expect. Assuming 172.5 mm cranks:

Identical power, 50% more force per stroke at 60 rpm than at 90. That is why grinding feels muscular and spinning feels cardiovascular: they are genuinely different loads on the tissue, even though the number on the screen has not moved.

These are averages across a full revolution, and the real picture is spikier still. You cannot apply useful force at the top and bottom of the stroke, so the peak torque during the downstroke is well above the average, and the numbers above understate what your knee actually experiences.

Two practical consequences follow. Low-cadence work is strength work, and should be programmed and recovered from as such rather than treated as ordinary tempo. And if your knees complain on climbs, cadence is the first thing to look at, well before saddle height.

Where the power is measured, and why two devices disagree

This is the single most useful thing in the article, because it resolves an argument that happens constantly and has a completely mundane answer.

Power meters measure at different points along the chain, and the chain loses energy as it goes. So they are not measuring the same quantity:

• Pedal and crank-based meters measure before the chain. They see the full output of your legs.
• Hub-based meters measure after the chain, so drivetrain losses have already been subtracted.
• Smart trainers measure after the chain too, and usually infer power from resistance and speed rather than measuring torque directly.

Chain efficiency is commonly quoted between 95 and 98% depending on cleanliness, lubrication and how cross-chained you are. So a crank meter reading 250 W and a trainer reading 242 W are not in conflict. They are both right, about different things, and the 8 W gap is the chain.

Add each device's own accuracy specification, frequently plus or minus 1 to 2%, and two devices can legitimately differ by 4 to 5% while both remain within tolerance. If yours differ by that much, nothing is broken.

Drift, and why calibration is not optional on some trainers

A measurement can be accurate on day one and wrong three months later. Two mechanisms cause this and both are mechanical rather than electronic.

Temperature. A trainer that has been running for twenty minutes is warmer than one that has just been switched on, and warmer components deform slightly differently under the same load. This is why a spindown calibration is supposed to happen after a warm-up rather than before one. Calibrating cold and then riding warm calibrates for a machine that no longer exists.

Belt tension and tyre pressure. On wheel-on trainers, the force pressing the roller against the tyre determines how much of your effort reaches the flywheel. Tyre pressure drops over a week, the contact changes, and the reported power drifts with it. This is the main reason wheel-on units are less repeatable than direct-drive ones, and it is entirely fixable by checking pressure before every ride.

Several modern direct-drive trainers need no user calibration at all and say so. If yours does, follow the manufacturer's procedure and do it after ten minutes of riding, not before.

Practical guidance for the setup around all this is in the setup guide.

What Zwift does with the number

Once the figure arrives, Zwift runs its own physics. It takes your power, your profile weight and height, the in-game weight and drag of your equipment, the gradient and surface you are on, and any draft you are receiving, and returns a speed.

Two things about that are worth knowing. The first is that your profile weight is an input to the physics rather than a display value, so a wrong number there changes your speed, not just your statistics. The second is that Zwift has never published the exact values it uses for drag area or rolling resistance, which is precisely why the calculators on this site fit curves to recorded finishing times rather than modelling the game from first principles. You cannot model a system whose constants are unpublished, but you can fit a curve to what it produces.

The consequence for the equipment question is covered in the equipment article: in Zwift, a bike is a small set of numbers rather than a machine.

Why the same legs make fewer watts indoors

The chain above explains part of a common complaint, and it is worth separating the parts that are measurement from the parts that are real.

• Measurement, not fitness: if you compare a crank meter outdoors with a trainer indoors, several percent of the difference is simply the drivetrain sitting between them.
• Real, and thermal: without a 30 km/h headwind your core temperature climbs, and heart rate rises for the same power. This one is genuine and it is the largest effect for most riders.
• Real, and mechanical: a bike locked to a frame cannot move underneath you, so a sprint recruits less of your upper body and core. This matters for short efforts far more than for an hour.
• Real, and continuity: outdoors you coast. Indoors you do not, so the same average power represents more total work.

The fix for the largest of those is a bigger fan, which resolves more indoor power complaints than any other single change. The rest is covered in the comparison of indoor and outdoor riding.

What this means for every number on this site

Everything the calculators here produce sits downstream of a measurement made by a device with a tolerance. The curve behind the Alpe estimate was fitted to times posted by riders whose own trainers had their own errors, and it will be applied to a power figure supplied by yours.

That is not a reason to distrust the output. It is a reason to read it at the precision it deserves. A 2% trainer tolerance on a 3.33 W/kg rider is roughly the same size as the model's own error band, which is why the data article argues that knowing your W/kg to better than about a tenth is not realistically available to you.

Use the estimate as a target to pace against and a way to compare two scenarios, which is what it is genuinely good for. Do not read the seconds.

The short version

• Power is torque times angular velocity, so 250 W at 60 rpm is 50% more pedal force than at 90 rpm.
• Crank meters measure before the chain and trainers after it, so a 2 to 5% disagreement is the drivetrain rather than a fault.
• Pick one measuring device and keep it. Consistency beats accuracy for tracking progress.
• Calibrate after a warm-up, not before, and check tyre pressure on wheel-on trainers.
• Zwift treats the number it receives as truth, and every estimate on this site inherits whatever error came with it.