When you stop pedalling, only two forces slow you on the flat: rolling resistance, which is near constant, and aero drag, which scales with speed squared. Newton gives m·a = Crr·m·g + ½·ρ·CdA·v² for each segment, with a the measured deceleration and v² averaged over the drop. Two segments at well separated speeds make two equations; the tool solves them for the two unknowns. The fast segment is dominated by aero, the slow one by rolling, which is what lets the maths tell them apart. Mass is bumped by 1.5 percent for the spinning wheels, which act like extra weight when decelerating. The deceleration is treated as constant within each segment, so keep the speed drops modest; that approximation costs only a couple of percent.

Find a flat, smooth, straight road with no wind and no traffic. Get up to speed, stop pedalling, hold your normal riding position dead still, and time the drop between two speeds on your head unit; no braking, no freewheeling out of position. Ride the same stretch in both directions and enter both times, the tool averages them, which cancels most of any slope or breeze you could not feel. Repeat each segment two or three times and use your most consistent pair. Keep the fast segment genuinely fast and the slow one genuinely slow; if they are close together the two equations look the same and the split between CdA and Crr goes soft. The result is a field estimate: good for tracking change and feeding the race tools, not a substitute for a controlled velodrome test.

Total drag is F = ½·ρ·CdA·v², drag power P = F·v. CdA here is built from your choices, not measured, so treat it as indicative. Pressure (form) drag comes from the size of the wake; for a rider it is the large majority of the total, moved most by position, frontal area, body size and helmet. Skin friction is the viscous drag along the surface, the smaller slice, and the one fabric changes. Interference is the extra drag where parts meet.

The air works the whole machine, not just the rider: it climbs over the back, deflects across the wheels and frame, and the spinning wheels shed their own wind into the wake. Fabric follows the wind-tunnel findings (Kyle & Burke; Brownlie; Oggiano): a textured or tripped skinsuit trips the boundary layer turbulent earlier on the rounded limbs, delaying separation and cutting form drag by more than it adds friction. That trade is speed-dependent, so the fabric's effect on CdA here changes with the speed box; see the crossover explainer below. The flow view shows that attachment: with a tripped suit at speed, a teardrop lid or a tucked position the streamlines hug the surface and follow it further round the back before separating into a later, narrower wake; loose kit lets go early and the wake rolls up large. Watch the dots too. They speed up where the flow squeezes over the back, then stall in the wake, the velocity deficit your legs are paying for.

A textured skinsuit works by tripping the boundary layer turbulent so it clings to your arms and shoulders for longer, separating later into a smaller wake. The catch is that the trip only bites once the air is moving fast enough. Below that, you have paid the texture's friction and bought no attachment, so the suit is slower than plain race kit. The chart shows the watts each fabric costs or saves against standard kit across speed, for the rider built above. The teal line crossing zero is the crossover: below it the textured suit costs you, above it it pays, and the faster you go the more it pays. This is why a suit chosen for a 50 km/h crit can be the wrong suit for a 25 km/h climb, and it is real tunnel behaviour (Oggiano; Brownlie), not a quirk of this tool.

Tick Live CFD solve and the air stops being drawn and starts being computed. A lattice-Boltzmann solver runs the flow around your rider in real time; lattice-Boltzmann recovers the same Navier-Stokes equations a wind tunnel obeys, in the low-speed limit, so this is real fluid dynamics rather than an illustration. The shape the air flows around is lifted straight from your rider's silhouette, so separation, the wake and the draft pocket all pour off the true outline, helmet, position and bodies included. The rider is then painted back on top. Colour is air speed: teal near freestream, red where the air is slow, which is the velocity deficit you pay for in watts.

Two honest limits. It solves a 2D slice, so it tells the truth in profile but cannot see the flow round the legs, the spinning wheels or yaw, and the drag it implies is sectional and indicative, not a measured CdA. And fabric is the exception: a tripped skinsuit works by tripping the boundary layer, which lives below the grid, so it still moves the CdA number but the solver will not pretend to draw it. Everything geometric, position, helmet, build, height, bike and the second rider, does change the solved flow for real.

The air behind a rider is slower than the air in front; it has lost momentum to the body. That lost momentum is the drag, near enough by definition, and it is how a tunnel measures the wake. So the red region in the live solve is not decoration, it is the bill. Fold the body smaller, hold the flow attached for longer, shrink that red pocket, and the watts come down. Watch the dots in the schematic for the same story: they speed up where the flow squeezes over the back, then crawl in the wake.

Add the second rider and the follower sits in the lead's velocity deficit, the slow, broken air the first body leaves behind. Lower oncoming wind speed means lower drag, which is why the wheel is worth holding. In the live solve the follower is not handed a discount; it is dropped into the slow pocket the solver already created, and the saving falls out of that. The figure here is a rough one anchored to a real deficit; the gap, the overlap and any crosswind all move it in the real world.

A small thing the live solve gets right on its own. A disc wheel is drawn solid, so the solver treats it as a solid barrier and the air has to go round it. A spoked wheel is drawn as a rim and spokes, mostly open, so the air passes through, which is roughly what a spoked wheel does. Neither behaviour was coded in; it falls out of using the real artwork as the shape the air meets.