What did you think of AI’s view of the remainder of the season? About right? Miles off? If only there was a way we could stack it up against a human prediction – or at least a human driving one of the most advanced, data-driven predictor models.
Meet Graeme Acheson and say hello to Sports4Cast.
Sports4Cast builds calibrated prediction models for a range of sports around the world, applying the same discipline across football, rugby, tennis, cricket and other codes: clean historical data, skill ratings update event-by-event, and probabilities back-tested against reality.
The good news for us? Acheson’s built a model for the WRC. Like Toyota’s WRC27 car, it’s a prototype which is undergoing advanced testing right now. Fortunately, DirtFish has been granted access to Rally4Cast and the remainder of the season – and precisely the same questions we put to AI – is the ultimate test.
“Rally4Cast,” Acheson told DirtFish, “is our rally engine and our prototype for motorsport as a whole. Rally4Cast rates every WRC crew from every stage of every rally since the year 2000. It simulates each rally, stage-by-stage, including detail such as road order, retirements, crews re-entering after retirement, then replays the season 10,000 times to turn who’s looking hot into actual percentages.
Thankfully, in reality the drivers only have to contest the 2026 season once
“Looking at DirtFish’s response from an AI chatbot, credit where it’s due: AI made some genuinely sharp calls. So when we were offered the chance to have Rally4Cast spar with AI, we thought it was worth a go.
“We put the same questions to our rally model: driver ratings built from every stage time since 2000, a stage-by-stage assessment of each remaining event, and 10,000 simulated endings to the 2026 season. We refreshed all data after Finland, and we’re running on Rally del Paraguay’s real 22-stage, 357.6km (222-mile) route and this week’s entry lists.
“First, the scoreboard on the chatbot itself, because it earned it: it refused to naively add Hyundai’s per-rally chances – the correct instinct because they’re not independent – and its eyeballed 60-65% lands close to our simulated 69%.
“Its Robert Virves call on WRC2 sits exactly inside our number. AI also spotted Chile, not Saudi, decides the Junior title. And its projected final points for the top two – ~270 and ~240 – are within a handful of our simulated means.
Predicting success for Robert Virves wasn't the toughest call for AI or Rally4Cast
“Where we differ is mostly about magnitudes: how much a points lead is really worth, and how much momentum really carries. That’s precisely where 10,000 replays help.”
In Graeme’s words, here’s what Rally4Cast found.
I need help making some predictions ahead of the final FIA World Rally Championship rounds of 2026.
Who do you think is going to win this weekend’s Rally del Paraguay?
Rally4Cast: Sébastien Ogier AI: Sami Pajari
| Driver | Car | Win chance | Podium chance |
| Sébastien Ogier | Toyota | 24% | 67% |
| Thierry Neuville | Hyundai | 15% | 48% |
| Elfyn Evans | Toyota | 15% | 38% |
| Adrien Fourmaux | Hyundai | 13% | 33% |
| Sami Pajari | Toyota | 10% | 35% |
| Takamoto Katsuta | Toyota | 10% | 34% |
| Oliver Solberg | Toyota | 9% | 33% |
The chatbot’s most interesting paragraph was the one it filed under “dark horse”. Ogier won the inaugural Rally del Paraguay last year, and event-specific evidence is strong evidence. Our model agrees sufficiently to make him the outright favorite.
Pajari’s ratings did rise after Estonia and Finland, the model watched both wins happen, stage by stage. But two great events move a career-long rating by less than a hot streak moves a narrative. Class is persistent, form is noisy. The math sides with the nine-time champion.
The quieter headline in that table is two Hyundais in the top four. The chatbot nominated Fourmaux as its dark horse; our board goes further and makes Neuville second-favorite. If you’re looking for the Hyundai breakthrough the article wondered about, the model says it’s live this weekend – not in November.
Neuville is second favorite for Paraguay, according to Rally4Cast
Assuming your prediction is correct, who do you think will win Rally Chile two weeks later?
AI’s Chile answer was conditional: “If Pajari wins Paraguay, I’d pick Katsuta for Chile.” And this is where a simulation gets to do something a chatbot structurally can’t. We kept every one of our 10,000 simulated seasons, so we can simply go and look at the 946 worlds where Pajari does win Paraguay, and see what happens next in those specific worlds.
| Chile winner | All 10,000 draws | Given Pajari wins Paraguay |
| Sébastien Ogier | 20% | 21% |
| Sami Pajari | 15% | 19% |
| Adrien Fourmaux | 13% | 15% |
| Oliver Solberg | 13% | 13% |
A Paraguay win for Pajari nudges the Finn’s Chile chances up. The worlds where he wins are disproportionately the worlds where he’s simply faster than we thought, but it doesn’t crown him and it doesn’t summon Takamoto Katsuta, whose post-Finland rating has him just off this pace right now. Momentum, in the draws, is real but modest: it works through points and small rating updates, not magic.
Predictions may not account for unexpected physical exertion in Paraguay
The bigger conditional is the title itself. Pajari winning Paraguay is the chatbot’s launchpad scenario, and even inside it, Evans remains the title favorite at 60% as opposed to Pajari on 36%. The article guessed Evans would drop to “around 35–40%” in that world. The lead is simply worth more than it feels.
Do you think Sébastien Ogier will win a 10th world title?
Here’s the strangest, truest thing our simulation says about Sébastien Ogier: he is our favorite to win every single one of the four remaining rallies. That’s roughly 20-24% at each event, yet he still only carries 1.9% for the championship.
The chatbot’s “deliberately conservative” 5% was, if anything, generous for a record-breaking number 10 for Séb. In the 3,517 draws where Ogier wins Paraguay outright, Evans’ title probability barely moves: 78.4% to 78.9%.
Beating the field doesn’t beat the arithmetic: a 62-point deficit with 35 on the table per weekend needs the leaders to fail, repeatedly, and Evans fails rarely in our draws. The article’s verdict, “probably not, but don’t be shocked”, is exactly right, but we’d move the “probably” from 95 to 98.
| Driver | Points now | Title (model) | Chatbot said |
| Elfyn Evans | 201 | 78% | 45% |
| Sami Pajari | 171 | 14% | 30% |
| Oliver Solberg | 156 | 2.7% | 8% |
| Takamoto Katsuta | 160 | 2.5% | 12% |
| Sébastien Ogier | 139 | 1.9% | 5% |
Note the ordering: the chatbot’s ranking was almost perfect – it had Katsuta ahead of Oliver Solberg; our draws flip them by two-tenths of a point, genuinely too close to call. What the simulation adds is calibration: with four gravel rallies left, a 30-point lead converts to a title nearly four times out of five, sweeping or no sweeping.
Evans is a firm favorite for the title
Will Hyundai win any of the remaining rallies this year?
The article put “at least one Hyundai win” at 60-65% and nominated Saudi Arabia as the moment. Our draws say 69% – a near-bullseye for AI’s instinct – but they flatly refuse the Saudi romance: Hyundai’s per-round win chance is a remarkably steady 25% at all four events. No finale spike; the desert is not a cheat code. If anything, the best single chance is the first one: Neuville and Fourmaux combine for roughly 28% in Paraguay this weekend. Toyota, for completeness, is 30% on all four.
Which M-Sport Ford driver will finish the season on top?
Josh McErlean over Jon Armstrong, said the article, at 65/35. Our draws agree on the name and shrink the gap to a whisker: after Finland it’s McErlean 21 points to Armstrong’s 16, and their simulated season-end means land at 41.1 and 40.6 – closer to a coin-toss with a nose in front than a two-to-one favorite. The article’s own hedge, “one strong result could completely change this”, is doing the heavy lifting there, and it’s correct.
Will Robert Virves be crowned WRC2 champion?
Rally4Cast: Virves 93.6% AI: Virves 90–95%
Perfect agreement, and a small backstage confession that Club DirtFish readers will enjoy: until this week, our participation model refused to simulate part-season campaigners into future rounds, which produced the magnificently stupid output of giving the man who has won every rally he’s entered a zero percent title chance! We fixed it and future rounds now use the actual published entry lists and the fix comes with real intelligence attached.
Roope Korhonen, Virves’s nearest points rival, is not on the Paraguay entry list, while Virves, Yohan Rossel and Alejandro Cachón all are. Moral of the story, in both directions: “the model says” is only as good as the auditing behind it. The chatbot’s 90-95% was an excellent guess and it’s nice to be able to show the working.
Who will win Junior WRC at the season finale in Chile?
The article picked Calle Carlberg to win the Chile finale and gave him 55% for the title. Our draws split that call in a way we find more interesting than either answer alone: Carlberg is our title favorite at 48% with Ali Türkkan second on 39% and Gil Membrado third on 11%. AI said 55%, 38% and 7% respectively, which is basically the same shape. But he’s only about 8% to actually win the rally.
The model suggests Carlberg will be crowned JWRC champion - but he's not an overwhelming favorite
Identical Rally3 machinery makes Junior pace brutally flat: Raúl Hernández (19%), Membrado (19%) and Leevi Lassila (17%) head our winner board. Championships at this level are won by finishing ahead of one rival, not ahead of everyone and the double-points rule, which our simulation applies to base points only (exactly per the regulations) means Türkkan needs to beat Carlberg on the road, and not just be near him. The article’s fear of Türkkan was well placed; its assumption that the champion must win the rally wasn’t.
What do you think the final top 10 championship classification will look like?
| # | Driver | Model mean pts | 90% range | AI |
| 1 | Elfyn Evans | 262 | 236–289 | 270 |
| 2 | Sami Pajari | 235 | 206–265 | 240 |
| 3 | Takamoto Katsuta | 218 | 192–246 | 215 |
| 4 | Oliver Solberg | 217 | 188–247 | 210 |
| 5 | Sébastien Ogier | 216 | 188–244 | 200 |
| 6 | Thierry Neuville | 189 | 162–217 | 150 |
| 7 | Adrien Fourmaux | 188 | 160–219 | 180 |
AI’s top two are startlingly close to our simulated means, within eight points on Evans and five on Pajari. AI’s “Toyota locks out the top five” is also our modal outcome. Two refinements from the draws: positions three through five are effectively a three-way dead heat with 218-217-216 for Katsuta, Solberg and Ogier separated by less than the noise. And we have Neuville almost 40 points better than the article’s 150: consistency compounds, even without a win.
Evans doesn’t need to win a single one of the last four rallies in either of our worlds – his most likely path to the title, like the chatbot’s, is podiums and patience.
Evans just needs to keep doing what he's done all season - racking up strong results
How Rally4Cast got its numbers
The engine: per-driver skill ratings, long-term and recent-form, by class and surface, are built from every WRC stage time since 2000; a stage-gap model that simulates each rally stage-by-stage, including road order, retirements and SuperRally restarts. A 10,000-run Monte Carlo simulation of the remaining season with championship points, drop rules and the Junior double-points finale applied per the regulations.
Fresh inputs: ratings updated through Rally Finland; Paraguay simulated on its published 22-stage, 357.64km itinerary and this week’s real entry lists. Chile, Sardinia and Saudi Arabia run on season-program fields until their entries publish.
Honesty notes: event boards above use our calibrated win probabilities – calibrated against leak-free 2022–26 backtests – which are conditional, and manufacturer numbers come straight from the stored simulation draws, with sample sizes as quoted. Probabilities are rounded; they won’t always sum to exactly 100.
Where we’d happily be wrong: the model prices Paraguay on class, event history and the real route. It cannot see how Ogier feels stepping back in after his crash in Finland or whether Hyundai’s test program found something, or what a team asks of its drivers in later rounds.
That’s what the next four WRC Sundays are for.
