Before Rally del Paraguay we ran the season 10,000 times and made Elfyn Evans the favorite. Then four of the 10 Rally1 crews retired, a 100-to-1 longshot stood on the podium – and yet Evans’ title chance hardly moved. Here’s what our model got right, what it got wrong, and why the quiet number is the one that matters.
A ‘0.4-retirement expectation’ meets a four-retirement rally
Let’s start with the miss, because it’s the biggest number in this piece and it’s usually the most fun for people when reading about algorithms. Summing our pre-event retirement probabilities across the Rally1 field gave an expected attrition of just 0.41 cars. Paraguay took four. Four!
Takamoto Katsuta and Thierry Neuville on Friday’s opening stage, Sébastien Ogier three stages later, and Jon Armstrong before the weekend was out. On our own numbers that’s roughly a 1-in-1200 weekend – which is a roundabout way of saying the numbers were probably wrong, not the weekend.
However, it was still relatively unlikely. Our retirement rates are tuned to how often front-runners actually fail to be classified, and second-year gravel rounds with tire-shredding surfaces evidently deserve a category of their own.
There’s a subtler lesson underneath: the model retires cars more or less independently, and Paraguay’s retirements weren’t independent – same compounds, same stone-strewn roads, everyone’s rubber failing the same way. Price that correlation and a four-car weekend stops being a 1-in-1200 event and becomes what it actually is: rare, but the kind of rare that happens. Both fixes are in hand; Paraguay’s results are already in the training data for the next fit.
What the crash map did see coming
Underneath the race model sits a stage-by-stage crash-risk map – see below – every sector of every stage graded for hazard, crossed with each driver’s incident history and skill ratings by surface, speed and type.
As you can see it had Katsuta as the highest-risk driver in the field – an index of 4.4 per 10,000 km-events, nearly double anyone else’s in the leading pack. He rolled his GR Yaris at the first corner of the first stage, which validates the numbers.
However, as you can see, the same map scored Neuville at 0.87 – the lowest risk among the front-runners – and he was gone within an hour of Katsuta; a puncture wrecking his rear suspension. Ogier, also one of the lowest at 1.06, clipped a bank and followed.
Neither was really a “crash” in the sense the model prices: the story here was rubber, not bravery or skill. Tires delaminated up and down the order all weekend, and a crash model has no opinion about tires. Scored honestly: one direct hit, two losses to a failure mode we don’t model – yet…
The stage map, though, turned out to be relatively prescient. The two most hazardous sectors we graded anywhere on the route – 3.55 per 10k, half again as spiky as anything else – sat one in the Yerbateras test (SS3 and SS7 above) and one in Artigas (SS19 and SS21).
Yerbateras is where Evans’s rear-right tire delaminated on both passes, the second costing him 65 seconds and, as it turned out, the rally. And Artigas is where Katsuta – back out under Super Rally, still the riskiest driver on our map – got it wrong a second time on Sunday: two wheels through a fast right-hander and into a concrete pole, his second and far bigger accident of the weekend.
Both crew members walked away; the stage was red-flagged and cancelled, and the rest of the field drove the map’s most dangerous stage as a road section. We graded the route; the route bit back. When the two sectors the terrain model flags in advance produce the incident that decided the rally and the crash that stopped a stage, the grading is doing its job.
The podium the carnage built
Sami Pajari winning is a result the model can take credit for – not because he was our favorite (he was 10%, fifth on the published board), but because of what the weekend subtracted.
The crews that retired were carrying 51% of our pre-event win probability. Re-weight the published board across the six surviving Rally1 crews and it reads: Evans 31%, Fourmaux 26%, Pajari 21%, Solberg 19% – and then Evans’s tyre took the top line out of the fight too. Pajari winning a rally like that isn’t a longshot landing; it’s the third pick in a cluster – the quick car that avoided every rock, bank and delamination.
You can watch it happen below: half the board’s win probability evaporates inside the opening stages, and from SS7 the rally is Pajari’s to lose.
Hayden Paddon and Adrien Fourmaux in second and third looks worse for us at first glance – until you hold the win odds and the podium odds apart. Fourmaux started the weekend at 33% to podium on the published board; Tuesday’s piece flagged two Hyundais in our top four, and Sunday’s podium had two Hyundais on it – just not the one we led with.
Paddon’s podium price was 11%: a live outsider, not a miracle. Once half the board’s win probability was out of the race, somebody’s outsider ticket was getting paid. A 1-in-9 shot coming in on a weekend of survivors is not a result the model has to be too concerned about.
Fourth place was a tire, not a temperament
A tempting read of Evans fourth – leading Friday afternoon, but then finishing 45 seconds down, 5.1 seconds off the podium – is that the championship leader drove like a man protecting a lead. The stage times say something simpler: he had an incident, and the incident is the whole story. Across Friday’s other seven stages he was quicker than Pajari; the entire deficit is one shredded tyre on Yerbateras. On Saturday he matched Pajari’s cumulative time to the tenth of a second over nine stages, and on Sunday he outpaced him by eleven – Fourmaux, with less to protect, found more still and took the final podium place off him on the last morning.
There is no caution debate worth having here. Take away one delamination on one pass of one stage and Evans probably wins the rally. Banking fifteen points, at a personal risk index of 0.90, on a weekend that put four crews off the road, is exactly the rally our model would have advised.
A five-horse race becomes two
Here’s why. In our previous article, Evans’s title probability was 78%. Then came a weekend in which half the field’s win probability crashed out, his closest championship rival won, and he lost the podium by five seconds – and his title probability now reads 76.9%.
All that pandemonium moved the number that matters by one point. He holds a 20-point lead with three rounds left, and a lead is simply worth more than it feels – that was the whole argument in the prior article, and Paraguay just stress-tested it with a flamethrower.
Perhaps the more interesting point is who is still in the running now – and what they need – not the size of the favourite. Ogier’s 1.9% went to effectively zero, eight Super Rally salvage points and all. Katsuta’s 2.5% is gone entirely. Solberg’s 2.7% is down to 1.5% despite sweeping Sunday. Nearly all of it flowed to one man: Pajari, 14% to 21.5%.
Before Paraguay, five drivers held a live – albeit small – title chance. After it, only two do – and one of them holds more than three-quarters of it. A weekend of chaos resolved the season into the simplest question it has asked all year:
“Can Pajari out-score Evans by 21 points across Chile, Sardegna and Saudi Arabia?”
He just out-scored him by 10 in a single weekend, so it is no longer a hypothetical. But he needs three more weekends like the last, and Evans needs three ordinary ones. If Evans wins even one of the three, it is effectively over – in the simulated worlds where that happens, his title chance is 96%. Even winning none of them leaves him at 66%.
The door Pajari kicked open in Paraguay is real; it just isn’t very wide yet.
