AR3-GP wrote: ↑14 Dec 2025, 00:44
Emag wrote: ↑14 Dec 2025, 00:00
To give some context, a 10.0 in this system represents a "special" season where a driver maximizes positional value, dominates their teammate, and statistically outperforms the machinery. Outside of Max (who achieved this every year from 2018 to 2025), this score was achieved by only three other drivers in that entire bracket:
*Lewis Hamilton (2018, 2019, 2020, 2021)
*Fernando Alonso (2023)
*Charles Leclerc (2025)
It really highlights how difficult perfection is to achieve. Yet Max scored it every single year since 2018. I have been impressed by Max ever since his rookie season, so this didn't exactly surprise me, but I'll be honest, seeing the data visualize that level of sustained performance really unlocked some perspective.
I think the rating system is interesting. Is there any version that does not result in a perfect score for any driver? It seems just qualitatively strange when we know that drivers do not drive perfect seasons. I've always been interested in things like race starts, racecraft, in-season penalties, and so on that could add another quantitative layer , but that would be difficult without reviewing every single race in every single season again (but we could take shifts
).
I checked some other seasons. Qualitatively Lando Norris was the 3rd best driver of 2021, but your ranking system places him 4th, behind Pierre Gasly. No one who watched that season would put Gasly ahead of Norris in 2021. What would need to be fixed here? On the other hand, it rates Lando Norris as 2nd overall in 2024. I could not disagree more with that
That’s a totally fair reaction, and it's good that it's asked here because I have the word capacity to explain what the model is and isn’t trying to do.
First, on the 10.0 thing:
The score does not mean “this driver drove a perfect season in every sense.” It’s more of a statistical saturation point. In some seasons the inputs (median results, car-adjusted overperformance, teammate gaps, consistency) add up to more than the scale allows, so the score just caps at 10. In other words, it’s saying given the variables I’m measuring, this driver completely maxed them out, so it doesn't really matter in the statistical sense if they did better. So basically, the claim is not that they did no mistakes, no bad starts, no penalties, etc.
Now, on Pierre 2021 vs Lando 2021, this is where the philosophy of the model matters a lot.
Gasly 2021 is heavily rewarded because:
*He finished extremely high relative to AlphaTauri’s expected performance
*His median finish position was excellent for that car
*He absolutely crushed his teammate all year
*There were very few low-end results dragging down his median
So from a “value extracted from the seat” perspective, Pierre looks insane statistically. That doesn’t mean the model is saying “Gasly is a better driver than Norris in absolute terms.” It’s saying Gasly got more out of an AlphaTauri than Norris got out of a McLaren, especially once Lando’s late-season drop-off pulls his median and consistency down.
This is also why Lando falls behind Pierre in 2021:
*McLaren had a higher baseline expectation than AlphaTauri
*Lando’s end-of-season dip hurts a median-based model more than people expect
*His teammate gap, while good, wasn’t as lopsided as Gasly vs Tsunoda
If you asked “who would I pick for a top seat in 2022?”, basically everyone would probably say Norris at the time. The model is not answering that question though, because that's where it turns a bit subjective and as humans we have a bit more context as well. This is just purely number-based and it's answering “who extracted the most relative performance from their machinery over this season?”
As for the 2024 results Lando being 2nd comes from the opposite effect:
*He outperformed McLaren’s early-season baseline
*Strong teammate dominance
*Much stronger consistency across the year
So again, it’s the same logic applied symmetrically, even if it clashes with gut feeling.
The model also brakes down the reasoning for the scoring. You will see that Lando gets a boost for dominating over his teammate "harder" than Charles did that season.
As for adding things like race starts, racecraft, penalties, you’re absolutely right, that would add another layer. The problem is that once you go there, you add some subjectivity, which is not in itself a bad thing, but there's a data limitation as well. Starts can depend on grid slot and strategy, penalties are inconsistently applied, and racecraft usually requires subjective judgement or manual tagging of events. At that point, your model is not fully data-driven unless you’re willing to annotate every race by hand, which unfortunately requires time I do not have.
So in short I will frame it this way:
This ranking isn’t a “who is the best driver?” list, it's rather a "who extracted the most relative performance from their machinery over this season" list. It’s a seasonal value-extraction model. Disagreements are to be expected, because it’s deliberately answering a narrower, more mechanical question, but let's be honest here. Even if it was the perfect model, there's always going to be disagreements.
