ROUND_12 · ZANDVOORT
Dutch GP 2026did we call it right?
Before the race we ran this GP 400 times through our engine, starting from the real qualifying grid, and published the odds. The race has now been run: here is the contrast, misses included.
Leer en españolVERDICT · RACE RUN
What we saidand what happened
The model called the top six and got the order wrong: its three favourites were the podium, in exactly reverse order
The model did NOT have Lando Norris as the favourite: it gave him a 18.3% chance of winning.
2.20
Avg. position error
across 16 finishers
0.87
Order correlation
1 = exact order
8/8
Front of the race
of the predicted top
8/10
Points scorers
called right
| Actual | Driver | Grid | Predicted | Diff |
|---|---|---|---|---|
| 1 | Lando NorrisMcLarenSecond win in a row for McLaren, 11.5s clear, on a three-stop in a wet-to-dry race. | 1 | 4.10 | -3.10 |
| 2 | Kimi AntonelliMercedesMoved ahead of team-mate Russell late on after a team order. | 3 | 3.60 | -1.60 |
| 3 | George RussellMercedesThe model's favourite (25% of wins) finished third. | 2 | 3.70 | -0.70 |
| 4 | Lewis HamiltonFerrari | 5 | 4.70 | -0.70 |
| 5 | Charles LeclercFerrari | 6 | 4.80 | +0.20 |
| 6 | Oscar PiastriMcLaren32s off the winner, on a four-stop. | 4 | 4.40 | +1.60 |
| 7 | Liam LawsonRed BullIncluding a 10s penalty for failing to slow under yellow flags. | 8 | 8.50 | -1.50 |
| 8 | Nico HulkenbergAudiThe model had him 15th on average: the biggest surprise of the day. | 13 | 15.70 | -7.70 |
| 9 | Fernando AlonsoAston MartinFrom 18th on the grid into the points, on just two stops. | 18 | 13.20 | -4.20 |
| 10 | Pierre GaslyAlpine | 11 | 12.50 | -2.50 |
| 11 | Yuki TsunodaRacing Bulls | 12 | 12.70 | -1.70 |
| 12 | Arvid LindbladRacing Bulls | 10 | 12.90 | -0.90 |
| 13 | Gabriel BortoletoAudi | 9 | 16.10 | -3.10 |
| 14 | Franco ColapintoAlpineIncluding a 10s penalty for failing to slow under yellow flags. | 14 | 13.90 | +0.10 |
| 15 | Sergio PerezCadillac | 22 | 17.90 | -2.90 |
| 16 | Carlos SainzWilliamsIncluding a 10s penalty for causing a collision. | 17 | 13.30 | +2.70 |
| DNF | Alexander AlbonWilliams | 16 | 15.10 | — |
| DNF | Valtteri BottasCadillac | 21 | 19.00 | — |
| DNF | Esteban OconHaasStopped beside the circuit on lap 56. | 15 | 14.90 | — |
| DNF | Lance StrollAston Martin | 19 | 18.60 | — |
| DNF | Oliver BearmanHaasPower unit failure, lap 2. | 20 | 16.20 | — |
| DNF | Max VerstappenRed BullCrashed heavily at the final corner on lap 1 on the damp surface, at his home race, bringing out the red flag. The model gave him a 20% podium chance from 7th on the grid. | 7 | 5.70 | — |
Where the model got it wrong
- Nico Hulkenberg predicted around P15.70, finished 8th (7.70 places off).
- Fernando Alonso predicted around P13.20, finished 9th (4.20 places off).
- Gabriel Bortoleto predicted around P16.10, finished 13th (3.10 places off).
- Lando Norris predicted around P4.10, finished 1th (3.10 places off).
- Did not see Nico Hulkenberg, Fernando Alonso scoring.
The pattern is clear: the model orders the front of the race well and underrates the midfield, because each car's performance is set per season and does not capture a given weekend's form. That is the biggest known limitation, and the next one to attack.
Official result: Wikipedia. The probabilities were published BEFORE the race and have not been touched since.
The preview, exactly as published
Favourite
George Russell 25%
Wins 25% of the simulations and finishes on the podium in 57.8%. Starts P2.
Driver by driver
| Grid | Driver | Win | Podium | Avg. finish |
|---|---|---|---|---|
| P1 | Lando Norris | 18.3% | 53% | P4.1 |
| P2 | George Russell | 25% | 57.8% | P3.7 |
| P3 | Kimi Antonelli | 20% | 56.3% | P3.6 |
| P4 | Oscar Piastri | 13% | 42% | P4.4 |
| P5 | Lewis Hamilton | 12.5% | 36.3% | P4.7 |
| P6 | Charles Leclerc | 8.8% | 34.8% | P4.8 |
| P7 | Max Verstappen | 2.5% | 20% | P5.7 |
| P8 | Liam Lawson | 0% | 0% | P8.5 |
| P9 | Gabriel Bortoleto | 0% | 0% | P16.1 |
| P10 | Arvid Lindblad | 0% | 0% | P12.9 |
| P11 | Pierre Gasly | 0% | 0% | P12.5 |
| P12 | Yuki Tsunoda | 0% | 0% | P12.7 |
| P13 | Nico Hulkenberg | 0% | 0% | P15.7 |
| P14 | Franco Colapinto | 0% | 0% | P13.9 |
| P15 | Esteban Ocon | 0% | 0% | P14.9 |
| P16 | Alexander Albon | 0% | 0% | P15.1 |
| P17 | Carlos Sainz | 0% | 0% | P13.3 |
| P18 | Fernando Alonso | 0% | 0% | P13.2 |
| P19 | Lance Stroll | 0% | 0% | P18.6 |
| P20 | Oliver Bearman | 0% | 0% | P16.2 |
| P21 | Valtteri Bottas | 0% | 0% | P19 |
| P22 | Sergio Perez | 0% | 0% | P17.9 |
Opinion · our reading
Verstappen starts seventh at home, and the number that explains it is not his
The first thing that jumps out of this simulation is that Max Verstappen wins 2.5% of the races. At Zandvoort. Where he has three wins and the grandstands are painted orange. It is hard to write without feeling the model has lost its mind.
But the number behind that 2.5% is not Verstappen’s: it is the car’s. In our database the 2026 Red Bull is the fourth car on the grid, behind Mercedes, McLaren and Ferrari, and not by a shade. Starting seventh is not a Saturday accident; it is roughly where that car belongs.
The proof sits two rows further down, and it is the most interesting thing about the weekend. Liam Lawson drives THE SAME car and starts right behind him, eighth. The simulation gives him a 0% win chance, a 0% podium chance and an average finish of P8.5: meaning he ends up half a place BEHIND where he started. Verstappen, in identical machinery from an adjacent slot, averages P5.7: he gains 1.3 places and reaches the podium in one race out of five. That gap (20% podium against 0%) is the only thing in this race that can be cleanly credited to the driver, because everything else is held equal.
And this is where the model probably falls short. Our engine does not know this is Zandvoort, or that Verstappen has won there three times; it has no field for a home crowd, for reading a race that falls apart, or for managing a tyre better than anyone when the window is narrow. It says 2.5% because it is being asked purely about raw pace in that car. If Sunday brings rain, a safety car or a strange strategy, that 2.5% is worth nothing, and Zandvoort is a circuit where all three happen.
My personal call, and this is opinion rather than data: the 20% podium looks low to me and the 2.5% win looks about right. Verstappen has a car to fight for a podium and not to win, and that is what has made this season different from the previous ones. The sad part is that the interesting question at his home race is no longer whether he wins, but how many places he can steal from a car that has no more to give.
How the model works
The engine accounts for how much starting at the front is worth at each circuit, and nowhere is it worth as much as at Monaco. Since the 2026 Azerbaijan GP that weight is calibrated track by track against the real races from 2014 to 2025; earlier previews ran on a simpler formula that derived it from how easy it is to overtake on each layout. And even so, the favourite is not the man on pole: Lando Norris stays at 18.3% and is edged out by George Russell on 25%, starting 2th. Not even at the circuit where the grid matters most does one starting slot cover the gap between the cars.
A favourite who only wins one race in four is not a broken model: that is what a real Grand Prix looks like. A retirement, a safety car at the wrong moment or a slow stop flips the result, and the engine simulates all of them. That is why 6 drivers have a genuine shot rather than a single name circled in red.
The most interesting column is the average finish: Fernando Alonso starts P18 and averages P13.2, the biggest gain on the grid: the driver the model thinks is most underrated by where he qualified.
The numbers come from 400 full race simulations (72 laps) with car and driver ratings updated to this point of the season. This is not betting advice and not an official prediction: it is what our model said. The odds were published before the race and have not been touched since: the contrast with the real result is above.