Why You Make Bad Decisions Late in a Zwift Race

A Zwift race is a physical test and an attention test running on the same hardware at once. I ride Zwift several times a week and I am training for Ironman Copenhagen, so my weeks contain 40-minute races where I think constantly and three-hour endurance rides where I think about nothing. Both fail in mental ways, and the failures are opposites. I am an engineering student, not a sports scientist, so nothing below rests on a study I cannot verify. It is mechanism, arithmetic from this site's own climb model, and my own riding, labelled as such.

What a Zwift race actually asks you to watch

List what you genuinely monitor in the last ten minutes of a hard race. The pack, or at least whether a gap has opened. The gradient ahead, because a 4% ramp changes what the pack does. Your power against a plan. Your PowerUp. A chat feed moving in the corner. And the gap to the rider in front, which you are converting into a decision about whether you can close it. All of it at threshold.

Only two pay off continuously. The rest are periodic or worthless in the moment: most of what is on your screen is not earning its share of your attention.

Three kinds of load, and only one is worth having

Cognitive load theory splits mental effort three ways, which tells you what to attack. Intrinsic load is baked into the task: holding 4.2 W/kg while deciding whether to follow a move is inherently hard, and only fitness and practice help. Extraneous load is effort spent on things that do not help, like a cluttered screen or a phone beside the bars, and it is free to delete. Germane load builds a model you reuse, such as knowing where this course splits. Cut extraneous, accept intrinsic, spend the rest on germane. Most riders do the reverse, because extraneous load is the entertaining kind.

Why decision quality falls as effort rises

The mechanism, as best I can argue it

Three things compete for the same narrow channel of attention while you race. Hard effort produces a loud stream of internal signals - breathing, burning legs, the urge to ease off - and suppressing them is itself an act of attention that costs more the harder you go. Working memory is small: "I am 8 seconds down, 2 km left, the climb starts in 400 m, I hold an Aero" is already near the limit, and each new item pushes an old one out. And tired attention narrows, keeping the most salient stream, which under effort is your own suffering, while dropping the peripheral ones where the tactical information lives. That is a prediction rather than a statistic: judgement degrades in step with accumulated effort, fastest in the streams you watched least closely.

What that predicts about race mistakes

A race concentrates its most important decisions in the window where you are least able to make them. Nothing in the first ten minutes of a 40-minute race decides the result. Almost everything in the last five does: when to move up, whether to follow the move that sticks, when to spend the PowerUp, when to sprint. So the errors to expect are not random. They cluster late, they look like "I saw the gap open and did nothing for four seconds", and in hindsight they feel like decisions you would obviously have got right at the start. My bad races are rarely a fitness story. They are a story about a four-second delay.

What a late mistake costs

The size of each cost depends on the course and the pack, so no single percentage would survive contact with a different race. The mechanism is knowable, though, and that is what lets you rank them.

If these tactics are new, read Zwift racing for beginners and the PowerUps guide first, or browse the racing section.

The opposite problem: the solo endurance ride

A race overloads attention. A three-hour zone 2 ride starves it, and that is also a problem. Nothing needs deciding: no pack, no PowerUp, no gradient that changes what you do. One power number and a clock.

Why the easier ride feels mentally harder

Outdoors the same three hours is full of small forced decisions. You steer, read the surface, brake for a junction, pick a line, shift for the next rise. None of it is difficult, and that is the point: it occupies attention cheaply. You also coast, where indoors you almost never do. The mechanism I think explains the rest is this: perceived effort is not a direct readout of the muscles, it is what your attention lands on. Outdoors it is spread across a dozen small tasks, so the signal from your legs gets a small share. Indoors, with nothing competing, it gets nearly all of it. That is reasoning, not a measured result, but it predicts what every indoor rider notices: the hardest part of a long trainer ride is not the legs. I compared the two settings in Zwift versus outdoor cycling and virtual hills, real wind.

Boredom is a limiter, not a character flaw

Abandon a four-hour ride at 2:40 because you cannot stand another minute and the session failed for a reason you can engineer around. Four things help, for different reasons:

• Structure. Ten minutes at the top of zone 2 every half hour costs almost nothing physiologically and gives attention a job.
• Segmentation. Three hours is unbearable as one object; six blocks of thirty minutes is fine. Only the counting changes.
• Group rides. Others restore the cheap monitoring you lose indoors. See group ride dynamics.
• Route choice. Terrain variation forces small adjustments, which is half of why route selection shapes what you finish.

The honest caveat about entertainment: a film reduces boredom by taking attention, not by creating it. That is a fine trade on an easy ride and a bad one on an interval session, where your power will drift during the interesting parts and get corrected during the dull ones. Check the trace afterwards and see.

Pacing is a cognitive skill

Holding a power target is not purely physical. It is a vigilance task: read a number, compare it to a target, correct, several times a minute, for as long as the effort lasts. Vigilance degrades in a predictable order.

The first thing to go is not average power, it is variability. Early on your corrections are small and frequent. Later they become large and infrequent, because you check less often and let the number drift further before you notice, so the trace roughens while the average still looks fine. That is a free early warning: if your 3-second power swings more than it did twenty minutes ago while your 30-second average is on target, attention is failing before your legs are. Simplify at once - one number, nothing else, and a deliberately shorter checking interval.

ERG mode removes the vigilance task by making the trainer hold the target for you. That is a real reduction in load, and why hard intervals often feel more achievable in ERG. It removes the practice too, though. If every interval is in ERG, you have never rehearsed what a race or a climb attempt demands: producing a steady target out of your own attention, unaided. I use ERG for short, fiddly, high-intensity work and free ride for anything long enough for pacing to matter.

Protocols that actually reduce load

1. Pre-decide the race

A decision made at threshold is made badly; the same decision made in the start pen is made well and costs nothing to execute later. Write the plan as trigger and response, so that in the moment you are matching a pattern, not reasoning.

Five rules is about the limit of what I can recall mid-race; a fifteen-rule plan has moved the load rather than removed it. Knowing your category also removes a class of uncertainty about who to follow, which is why the racing category calculator and the FTP calculator come before the plan, not after it.

2. Simplify the screen

Decide before the start which two numbers you may look at and treat the rest as decoration. For a race that is power and the pack; for a climb attempt, power and elapsed time. Close the tabs, put the phone face down, turn off the second screen. It is the cheapest item here: no fitness, no willpower, just a decision made in advance.

3. Chunk the effort, and save judgement for the end

Attention holds a small object far better than a large one, so never carry "40 minutes at threshold" in your head. Carry "four minutes to the next boundary" and rebuild the object each time you reach one, using whatever the course provides: laps, banners, hairpins, kilometre markers. Then budget judgement like the depleting resource it is: spend it early working out where the race will split, then coast on your rules so you reach the last ten minutes with something left to think with.

What this means for a climb time attempt

A solo climb attempt is the purest version of the pacing problem: no pack, no PowerUp decision, no tactics, just one number held for the best part of an hour. That is why lapses of attention show up so cleanly in the result. Alpe du Zwift is 12.2 km with 1,036 m of gain, roughly 8.5% average, over 21 hairpins. The fit behind this site's Alpe du Zwift calculator is time_seconds = 148.60*(W/kg)² - 1954.08*(W/kg) + 8329.87, and it gives this:

Roughly 3.2 W/kg is what it takes to break the hour, and near that point the model's slope is about 1,000 seconds per W/kg. For a 75 kg rider a 5 W sag is 0.067 W/kg, worth about 67 seconds across the whole climb. Let it appear only in the final third, once vigilance starts to fail, and my rough estimate is still around 20 seconds lost: the difference between 59:58 and 60:18, from nothing but checking your power less often.

So pace the Alpe as an attention problem. Fix the target beforehand so there is no arithmetic to do on the bike, use the 21 hairpins as chunk boundaries, and check power on a deliberate rhythm, because whenever you happen to look becomes less often as you tire. Ven-Top, Zwift's Mont Ventoux in the France world, is roughly 20.9 km and about 1,534 m of gain, so the same vigilance task runs longer. Set that target with the Ven-Top calculator or the general climb time calculator, compare the two on Alpe versus Ven-Top, and take the detailed pacing from conquer Alpe du Zwift.

Telling mental fatigue apart from physical fatigue

They feel similar from the inside and call for opposite responses. Physical fatigue means less power for the same perceived effort, and simplifying the task does not help. Mental fatigue means the same power feels harder, and cutting what you track does help.

• 3-second power swings more than earlier while the average is unchanged: attention.
• You read a number, fail to register it, and read it again: attention.
• You notice a gap only after it has doubled: attention.
• You push as hard as before and the number is lower, or cadence collapses on the flat: legs.

Get it backwards and you rest when you should have restructured. There is a planning consequence too: a hard race after a mentally heavy day of work or study is not the same session as the identical race on a quiet Sunday, so I put my hardest racing where my head is freshest. The training section and the 10-day training week cover how to lay that out across a block.

For the deeper version of how perception shapes what you can do on a trainer, I wrote about it in ghost in the machine.