The 7MJ Reset: Why Our Model Just Threw Out Three Races of Data
Miami's rule tweaks worked. Now the FIA wants more — and our pre-Miami pace database is suddenly half-useful.
“Three races of practice, qualifying and race data from Melbourne, Shanghai and Suzuka — roughly 2,400 laps — just got their predictive weight cut by more than half.”
What actually changed in Miami, and why it matters
The headline number is small. Maximum recharge dropped from 8MJ to 7MJ. One megajoule. Easy to dismiss.
It isn't. That single figure rewrites how a 2026 car is driven through a lap. Peak superclip power jumped from 250kW to 350kW, which sounds like more deployment but actually means less time spent harvesting and less time spent in that awkward lift-and-coast purgatory drivers spent the opening three rounds complaining about. Race-condition Boost is now capped at +150kW. The list of circuits eligible for lower energy limits expanded from 8 to 12.
The trigger was Bearman's Suzuka shunt and the closing-speed problem the FIA quietly acknowledged: cars deploying into cars harvesting, with deltas big enough to catch a following driver out. The fix included flashing rear and lateral lights on affected cars — a visual warning system that didn't exist three rounds ago.
Friday's online meeting between the FIA, FOM, team principals and power unit manufacturers confirmed the Miami package will stay, and that more tweaks are coming. "No material issues or safety concerns," was the language. Translation: it worked, we're going further. Canada, Monaco and Barcelona are the likely staging grounds for round two of the refinements.
Why our model is partly blind right now
Predictive models live or die on the relevance of historical data. The 52 features we feed into the ensemble — tyre deg curves, sector-specific pace, qualifying-to-race conversion, energy deployment patterns — were trained on a regulatory framework that no longer exists.
Miami is one race. One. Against three pre-change races where teams optimised for an 8MJ envelope that's now obsolete. The teams who built the cleverest harvesting strategies under the old regime — and there were clear winners in that game through Melbourne, Shanghai and Suzuka — have just had part of their edge legislated away.
The practical response: weight Miami at roughly 3x its raw sample size, widen uncertainty bands on every team's pace estimate until we have a cleaner read post-Barcelona, and flag any prediction where the confidence interval has ballooned beyond our usual thresholds. Our Brier scores will likely take a hit through Canada and Monaco. That's the honest cost of a mid-season regulation reset, and we'd rather show wider error bars than fake precision.
The Elo ratings stay anchored — driver skill doesn't evaporate because the recharge limit changed — but circuit-type Elos for power-sensitive tracks need recalibration.
What the Miami podium tells us (and what it doesn't)
Antonelli won. Norris second, Piastri third. Norris had taken the Sprint. One data point, but a loud one.
Under the old 8MJ regime, Red Bull and Ferrari had built a comfortable cushion in raw race pace through the opening three rounds. Miami flipped that. Mercedes converting pole into a win with their teenage rookie, and McLaren bracketing the podium, hints at exactly the dynamic the new rules were designed to encourage: cars more on the limit in qualifying, less reward for energy-management cleverness, more reward for single-lap pace and tyre management.
The model's pre-Miami P1 probability for Antonelli at any given circuit was sitting in single digits. Post-Miami, on tracks resembling its layout, that number needs to climb materially. The harder question is whether Miami was a Mercedes-and-McLaren circuit specifically, or whether the regulatory shift genuinely tightened the front of the grid.
We won't know until Barcelona. Canada is a low-downforce, power-hungry layout that will stress the new energy rules differently. Monaco is Monaco — a glorified qualifying session where race pace barely matters. Barcelona is the cleanest aerodynamic and energy read on the calendar, and it's the race where our confidence intervals should start narrowing again.
The Verstappen problem and the moving-target season
Verstappen has been the loudest critic of the 2026 framework, and his position is consistent: he expects no major changes this year and is pinning hopes on 2027. Friday's meeting suggests he's half-right. The changes are evolutionary, not revolutionary. But "evolutionary" applied repeatedly across a season still adds up to a meaningfully different sport by Abu Dhabi.
For prediction markets, this is genuinely awkward. Kalshi's 2026 Drivers' Championship odds are pricing a stable competitive order. Our model, comparing against those markets, currently sees value in widening the spread — the probability that the title fight reshuffles between now and August looks underpriced if the FIA keeps tweaking parameters every few rounds.
The operational rule we're adopting through the European swing: any race where a regulatory change has landed within the previous two rounds gets a confidence penalty applied to our headline P1 probability, and the position distribution gets flattened. A driver who would normally be 38% to win might show as 31% with a wider tail. That's not the model getting worse. That's the model being honest about a season where the rulebook is being rewritten in real time.
Miami was the reset. Canada is the first test of whether the reset holds.
- —The Data Driver