Zandvoort said goodbye with Verstappen in the wall, and our model saw none of it

Zandvoort said goodbye to Formula 1 after thirty-five visits, and it did it with the worst possible script for half the grandstands: Max Verstappen, at home, in the last race this track will ever see, put it in the wall on lap 1. Not in an overtake, not in a fight. At the final corner, on a damp surface, on his own. I am writing this on Monday and it still feels like an ending nobody would have dared to write.

What we said on Friday

We do something here that is genuinely nerve-wracking: we take the real starting grid, run it through our simulator hundreds of times, and publish the odds before the race is run. No edits afterwards. We did it in Hungary and we did it at Zandvoort, which means now comes the uncomfortable part: putting the result next to it.

Our favourite was George Russell, at 25%. Behind him, Antonelli at 20% and Norris at 18.3%. Norris won. So we did not call the winner, and I am not going to dress that up.

But look at the podium: Norris, Antonelli, Russell. Those are exactly our three favourites, in exactly the reverse order. And if you look at the top six instead, they are also the six the model put at the front, shuffled. The average error for the race was 2.2 places across sixteen classified finishers. The full comparison is here, driver by driver, misses included.

My reading, and it is an opinion: that is roughly the ceiling of what a simulator can do. Getting who will be at the front right is a question of car, driver and grid, and that can be modelled. Getting the order right between three cars covered by three tenths is something else. It comes down to a pit stop, a red flag and who does not make a mistake. Anyone selling you a model that nails that is selling you smoke.

The box that does not exist

We gave Verstappen a 20% chance of a podium from seventh on the grid. He was gone on lap 1. And this is the part of the weekend that interests me most, because it is not a bug in the model: it is a limit of it, which is a different thing.

Our engine knows how fast a car is, how good a driver is and how often it rains at each circuit. What it does not have, and we say so openly on the how-it-works page, is a box for “it is your last home race, there are ninety thousand people in orange, the track is half wet and you have just decided that today you are not lifting”. That is not a variable. I am not even sure it should be one.

There is an ugly temptation in simulation work: believing that if the model did not see it, it did not matter. It works exactly the other way round. What decided this race was precisely what no number in our database contains.

Antonelli is not a prospect any more

On Friday we left three questions on the table. The first was whether Antonelli’s first half had been a hot streak. The answer: 242 points and a 59-point lead over Russell, his own team-mate, with Hamilton level with Russell on points and behind him on countback.

And one detail from Sunday says more than the points do: Antonelli passed Russell late on a team order. Mercedes has already chosen, and it has done it in August. I think it is the right call, and also the kind that ages badly if the championship tightens, because Russell has just been told live on air what his role is.

For context on what that is worth, we have 1,200 measured seasons on what a team’s second seat actually changes. The short answer is that sometimes it is worth a constructors’ title and sometimes it is worth nothing. Mercedes are 87 points clear of Ferrari; they can afford to choose.

McLaren have arrived, but they have arrived late

Second question: whether Hungary had been a one-off. It was not, because Norris won again, two in a row, and yet he is still 83 points back with eleven races to go. Arithmetically he is alive. Competitively he needs something to happen to Antonelli, and “something happening” is not a plan.

The third question was Lawson, promoted into the Red Bull after Hadjar’s injury. He finished seventh with a ten-second penalty on top. On a weekend when the other car in his colours crashed out on lap 1, scoring and breaking nothing is exactly what he needed to do. It is not a headline, but it is an argument.

And that is it: it does not come back

What annoys me most is that Zandvoort is leaving just as it had made clear what it was for. A track where overtaking is hard forces the race to be decided somewhere else: in the stop, in the tyre, in who calls it right when the rain arrives. And Sunday was exactly that: rain at the start, a red flag, four stops on some cars, and a winner who was nobody’s favourite.

Less spectacle per lap, more story per race. I think that is a good trade, and the calendar has just decided the opposite.

Why we publish the misses

We could publish the predictions and go quiet when they turn out wrong. That is the normal thing to do, in fact. We do not, for a very simple reason: a prediction you only show when it was right is not a prediction, it is advertising.

So it is all there: the model’s running scoreboard, race by race, with the average error, the order correlation and the list of places where it goes most wrong, always the same ones, the midfield, where strategy and judgement decide rather than the car. With two races that is not a track record, it is an anecdote. In eleven more it might be something else.

See you at Monza, with the odds published before the race. As always.