PolePredictbeta

Learn · 3 min read

Why an AI can rank the grid but still miss the winner

Ordering twenty-two cars and picking one winner are very different problems.

Our grid correlation can look strong while our winner picks fail. That isn't a contradiction. It comes from what each measure asks.

Ranking the field is averaging

Getting most cars into roughly the right places is forgiving. Swap two drivers in midfield and the score barely moves. Fast cars usually finish ahead of slow cars, so a model that understands relative pace does well.

The winner is one outcome

Picking the winner is all or nothing. When two or three cars are close in pace, small events decide it: a start, a safety car, a strategy call. The model can have the right driver second and still score a miss.

What we do about it

We publish win probabilities, not just a name, so you can see when a call was a coin flip. And we publish every miss. Compare any race on the scorecards page.

Next: Scorecards · Beat the AI

← All explainers