How it works

Where the numbers come from, what each forecast means and how far you can trust it.

How far you can trust it

Not a fortune teller

We give probabilities, not predictions. When we say a rider has a 30% chance of winning, we mean that out of 10 situations like this they win about 3. The favourite often does not win.

The starting grid is the single most important piece of information in MotoGP. Before the sprint, our race forecast orders the riders almost exactly like the grid; what it adds is the uncertainty range and the chance of retirement.

The sprint is the only weekend data that makes us change our mind: it is real race pace, in traffic, the day before. With it, the forecast does beat the grid.

How the forecast has done

Every forecast is compared with the grid, the benchmark to beat. All computed on past races, using only the information available before each race.

Forecast	Races	Order (model / grid)	Position error (model / calibrated grid)	Error with the raw grid	Favourite wins	80% range hit rate
Race (after the grid)	219 (2015–2026)	0.76 / 0.76	2.8 / 2.8	3.8	37%	84%
Race (after the sprint)	76 (2023–2026)	0.79 / 0.75	2.5 / 2.7	3.8	45%	85%
Sprint	76 (2023–2026)	0.82 / 0.82	2.5 / 2.5	3.0	45%	87%

The model does not order riders better than the grid: the predicted order is practically the same. What it does is calibrate, and a calibrated grid with no model is off by almost the same. The only forecast that genuinely improves on the grid, in both order and error, is the one after the sprint.

Error: average positions off, compared with the calibrated grid (the grid turned into expected positions with a simple regression, no model). If the model does not beat it, its edge over the raw grid is only calibration.

By season (race)

Season	Races	Order (model / grid)	Position error (model / calibrated grid)	Favourite wins
2026	15	0.79 / 0.79	2.5 / 2.6	27%
2025	22	0.71 / 0.71	2.9 / 2.9	50%
2024	20	0.74 / 0.74	2.8 / 2.8	40%
2023	20	0.79 / 0.79	2.6 / 2.7	40%
2022	20	0.75 / 0.75	3.0 / 3.1	15%
2021	18	0.68 / 0.68	3.1 / 3.1	33%
2020	14	0.71 / 0.71	3.0 / 3.0	36%
2019	19	0.84 / 0.84	2.5 / 2.4	32%
2018	18	0.79 / 0.79	2.9 / 2.9	28%
2017	17	0.76 / 0.76	2.8 / 2.8	35%
2016	18	0.67 / 0.67	2.9 / 2.9	56%
2015	18	0.88 / 0.88	2.4 / 2.5	50%

What we do and how

The weekend forecasts

Friday, after Practice: who will go from Q1 to Q2.

Saturday, after Q1: pole and front-row chances.

Saturday, after Q2: sprint and race forecast, from the official grid (with penalties).

Saturday afternoon, after the sprint: race forecast updated with sprint pace.

Monday: what happened versus what we forecast.

How to read the data

Descriptive does not mean predictive. Rider and circuit profiles describe what has happened; only the weekend forecasts look ahead.

Always with the sample size. When a figure comes from few cases, we say so, and show the usual range instead of a single number.

The title simulation is "if things carry on like this", not a calibrated forecast, and we show it as a range between weighing the whole year or only recent form.

The grid is always the official one, with penalties applied.

Where the data comes from

Results, sessions and grids: Dorna and the official FIM PDFs (laps, sectors, tyres, incidents), since 2010.

Weather: Open-Meteo, at each circuit's location.

Data updates itself during the weekend, a few minutes after each session.