The 2026 Regulation Reset
How new rules broke every prediction model — including ours
“When the regulations change, the past becomes a different country. The data is still there — but its meaning has shifted.”
The Biggest Reset Since 2014
Formula 1's 2026 regulations represent the most comprehensive technical overhaul the sport has seen since the turbo-hybrid era began in 2014. New power unit architecture. New aerodynamic philosophy. Active aerodynamics for the first time. Revised weight distribution. A fundamentally different car concept.
For fans, this means unpredictability. For prediction models, it means crisis.
Every data-driven prediction system — ours included — relies on the assumption that past performance contains information about future performance. When a driver finished P2 at Silverstone last year, that tells us something about their likely performance at Silverstone this year. When a team dominated high-speed circuits in 2025, that informs our expectations for Monza in 2026.
The 2026 regulations break this assumption. Not partially — fundamentally. The car that dominated in 2025 was designed for a different set of rules. The aerodynamic concepts, the power unit mapping, the tyre interaction — everything changes. A team that understood the 2022-2025 ground-effect formula perfectly may have misunderstood the 2026 active-aero formula entirely.
This is not speculation. We have seen it before.
Lessons From Previous Resets
The historical record is instructive:
2009: Brawn GP. A backmarker team under a different name (Honda) produced a car so well-suited to the new regulations that they won the championship in their first season. No prediction model based on 2008 data would have placed them anywhere near the front.
2014: Mercedes. The turbo-hybrid regulations transformed the competitive order. Mercedes went from occasional race winners to utterly dominant — winning 16 of 19 races in the first season. Red Bull, the four-time consecutive champion, fell to P2 and stayed there.
2022: Ground Effect. The return of ground-effect aerodynamics reshuffled the midfield dramatically. Ferrari emerged as genuine title contenders after years in the wilderness. Mercedes, dominant for the previous eight seasons, struggled to understand their car's behaviour for months.
The pattern is clear: regulation changes redistribute competitive advantage in ways that historical data cannot predict. The question is not whether the 2026 reset will surprise us — it is how much.
How Our Model Adapts
We handle regulation resets with a three-part strategy:
1. Elo Dampening
When a major regulation change occurs, we pull all Elo ratings toward the mean by 80%. A driver rated 1800 (elite) drops to approximately 1560. A driver rated 1300 (struggling) rises to approximately 1340. This compression reflects genuine uncertainty — we do not know who will be fast under the new rules.
Simultaneously, we increase the K-factor (the rate at which ratings update) by a factor of 1.8-2.0. This means early-season results carry outsized influence on ratings. A strong result at R01 moves the needle more than the same result would in August.
2. Dynamic Weighting Shift
Our standard weighting gives the ensemble model (which uses historical features) 40% influence by mid-season. During the first two races under new regulations, we reduce historical feature influence and increase the weight of current-season signals: qualifying pace, practice times, and team performance.
In practice, this means our Round 1 predictions under new regulations are wider — closer to a uniform distribution — than they would be in a stable-regulation year. This is correct. Expressing high confidence when the rules have just changed would be intellectually dishonest.
3. Feature Re-Evaluation
Some features become less predictive under new regulations. Circuit-specific history, for example, carries less weight when the cars behave differently at every circuit. Qualifying gaps, by contrast, remain informative — regardless of regulations, the driver who qualifies fastest on Saturday is more likely to win on Sunday.
We re-evaluate feature importance after every race in the first half of a regulation cycle, pruning features that have lost predictive power and up-weighting those that remain informative.
What the Early Data Shows
Two races into the 2026 season, the regulation reset has produced exactly the kind of disruption the historical pattern predicted.
Mercedes — who struggled through much of 2025 — have emerged as the team to beat. Their understanding of the new active aerodynamic regulations appears to be significantly ahead of the field. George Russell and Kimi Antonelli have both stood on the podium in every race so far.
This is a dramatic reversal from 2025, where Mercedes struggled with an underperforming car concept. Our pre-season model, which relied partly on 2025 performance data, initially underestimated Mercedes. The Elo dampening meant we did not rank them last — but we did not rank them first either.
After two races, our model has rapidly adjusted. The high K-factor means Mercedes' Elo has climbed sharply, and the recent-form signal now heavily favours both their drivers. Our Round 3 predictions reflect this updated reality.
Conversely, some teams that performed well in 2025 have struggled under the new rules. This, too, is consistent with the historical pattern: the correlation between performance under old regulations and new regulations is significantly weaker than the year-to-year correlation within a stable regulation period.
New Faces, New Challenges
The 2026 grid includes drivers in new situations: Arvid Lindblad makes his F1 debut at Racing Bulls, while Valtteri Bottas and Sergio Perez return to the grid with the new Cadillac team after a season away. Several young drivers still early in their F1 careers — Antonelli, Hadjar, and Bortoleto in their second seasons, plus Bearman and Colapinto with two years of experience — face the added challenge of a regulation reset during a formative stage.
Drivers with limited F1 history are challenging for prediction models. Our Elo system uses fewer data points to estimate their strength, and the regulation reset makes their 2025 performance data less predictive of 2026 results.
Antonelli's case is instructive. After a solid first season at Mercedes in 2025, the regulation change appears to have suited his driving style. His maiden victory in China was the fastest a second-year Mercedes driver has won since the turbo-hybrid era began. Our model is updating rapidly — his Elo has climbed sharply after two races, reflecting the outsized K-factor during regulation-change seasons.
The lesson for our model: performance under new regulations can diverge significantly from the prior season. Drivers and teams adapt at different rates, and the model must be flexible enough to update quickly when the data demands it.
Prediction Accuracy Under Uncertainty
How do you measure prediction accuracy when the world has just changed?
Our Brier score after two races is informative but must be interpreted carefully. Early-season predictions under new regulations are inherently wider — we assign more probability mass to more drivers. This means our Brier score will naturally be higher (worse) than during a stable mid-season period when we can confidently identify the top three.
The correct benchmark is not "how does our Brier score compare to mid-2025?" but "how does it compare to what a pure grid-position model would produce, and what a championship-standings model would produce, under the same uncertainty?"
By this measure, our model has performed well. The regulation-adapted predictions — with their Elo dampening, dynamic weighting, and rapid updating — have outperformed both the grid baseline and the championship baseline in the opening races. We are making fewer confident wrong predictions than a model that naively extrapolates 2025 performance.
This is the value of building a model that explicitly handles regime changes rather than pretending they do not exist.
What Comes Next
The first five races of a regulation cycle are the most volatile. By Round 6, the competitive order typically stabilises as teams understand their cars and development trajectories become clearer.
For our model, this means:
- —Elo ratings will gradually converge on a new equilibrium that reflects 2026 performance rather than historical reputation.
- —The K-factor will remain elevated through Round 5 before decaying toward its standard value.
- —Feature importance will shift as we accumulate circuit-specific data under the new regulations.
- —Prediction confidence will increase race by race as the model learns the new competitive landscape.
We expect our best prediction accuracy to come in the second half of the season, once the model has ingested enough data under the new rules to reliably separate signal from noise. Until then, we will continue publishing honest probabilities — wider when uncertainty is high, sharper when the data supports it.
The 2026 regulations have reminded us of a fundamental truth about prediction: the most valuable thing a model can do is not to be right — it is to be honestly uncertain when the world is genuinely uncertain. The confidence will come. The data will accumulate. The model will learn.
In the meantime, we publish every prediction and score every result. That is the deal.