A prediction is a whole lap, not a single number. The pole time is just the integral of everything below it. Every channel on this page comes out of the frozen model, reproduced bit-for-bit from the pre-registered call. None of it is drawn by hand.
The lap under a load-sensitive tyre and the friction ellipse. Peak , slowest corner .
Each dot is one point on the lap: how hard the car brakes or accelerates (horizontal) against how hard it corners (vertical). The tyre can only spend its grip once, so the cloud is bounded by the friction ellipse. Braking reaches , cornering . Nothing sits outside the budget, because the solver never lets it.
braking points, where the model sheds speed into a corner.
of the 4 MJ budget, spent over , worth against the same lap with no ERS.
Deploying the fixed 4 MJ where it pays turns a lap into . The shipped config is the simple greedy deploy, which is canonical and easy to defend. A Bellman DP finds a tighter optimum and is kept as a secondary.
2026 cars trade downforce for drag on the straights. In the model, straight mode drops the downforce coefficient 5.0 → 3.0 and drag 1.0 → 0.6. Those are illustrative values at a nominal CL, since only the grip×load product is identified (see method). of the lap runs in low-drag mode.
Every channel above is reproduced from the pre-registered frozen model (sha256 …). Dry pole only, no driver names, just the limit of the 2026 car. How it works →