Pulling real standings and results to build the model.
Pick a contender, bend their assumptions, and watch the live Monte Carlo shift. Defaults come straight from this season's real results.
Not a single average. The remaining season is simulated tens of thousands of times, sampling a noisy result for each contender in every race, then counting how often each one ends up champion. Every default is derived from real data, no hand-set numbers.
Remaining Grands Prix and sprints come straight from the live schedule, so the points still on the table are exact, not assumed.
Each driver's average finish is an outlier-robust mean of their real classifications, then sampled as a noisy position every race, so a fast car still has the odd quiet Sunday.
DNF rates come from real retirements, but a clean run of seven races isn't a 0% car. Each rate is shrunk toward a sensible prior so small samples don't lie.
The one manual knob. A handful of races can't justify extrapolating a development slope, so you set it per driver rather than the model guessing.
The top contenders are all modelled at once, and some races are stolen by an outsider, a rate also estimated from who's actually been winning.
After every simulated season the points are tallied and the champion recorded. Each driver's share of those wins is their probability.