When reading the F1.com Bahrain 2023 GP strategy guide I started wondering what is the margin of error in the race pace estimations. Before thinking about error margins however I needed to replicate the estimations reported in the article first. So I downloaded FP1, FP2 and FP3 data from FastF1 and started looking for gaps between RedBull and the rest...The thing is, I could not really match it.

F1.com as a benchmark

F1.com reports the following gaps to RedBull

Estimating Race Pace

Assuming Race Pace estimations are based on the long run during FP2, I started looking for 0.46 gap between RedBull and Ferarri. This one I found pretty quickly as a gap between average times of PER and SAI.

But this brings me also to the first question. What is the definition of the race pace gap estimation?. It seems it should be at least something comparing two runs of the same length, with the same tyre life at the start of the stint. Which means estimating tyre degradation to be able to simulate/extrapolate missing data points. But somehow following this approach I could not get close to the gap reported in the article.
So I went for a naive approach: Average of lap times of a long stint during FP2. Without taking into account the length of the stint, and the tyre age at the beginning of a stint.
This led directly to the 0.47s gap between RedBull and Ferrari But further the things started to be tricky. For Aston Martin only ALO did a long run on soft tyre, STR tested on Mediums. And ALO was very fast. Even faster than PER ! So here it seems that the pace of STR on Mediums is compared to PER and not ALO on Softs. Is it the case? Or is the estimation approach used in the F1.com article different?

The next two estimates following naive approach looks close to the reported ones. Alpine seems tricky again. Both drivers did long runs on softs. Which one should we look at? And the gap seems to be much different either way. The last three teams are again very close to the f1.com estimation following naive approach.

Bottom Line

There is not enough information in the article to replicate the estimates, and the approach itself is not well defined leaving room for some creativity. But using simple averages gives similar gaps for 8 of 10 teams, indicating this was the way.
Aston Martin and Alpine cannot be matched this way, and I hope someone can point me in the right direction.

Looking forward to hear from you. You can find me on twitter as @f1scope And now back to margin of errors...

Asked ChatGPT today about 1997 F1 Season and got a little bit disappointing answer

New version includes career summary. It is based on the idea of driver distillates from the previous versions of f1scope. Enjoy.

After some time of inactivity F1Scope is back again (*). This time, the ambition is not (only) to visualize but to analyze and liberate F1 data. F1Scope distillates/visualizations will be probably back at some point. But not yet.

It is also the first solution based on a "soseki.io":http://soseki.io platform. Meaning it is a place where "soseki.io":http://soseki.io will be put to a test itself.

(*) for the 5th time since 2014 when the first dynamic visualization of a GP was published