I enjoy running custom GT3championships in rFactor 2 (plug: https://forum.studio-397.com/index.php?threads/custom-ai-championship-tutorial.64916/), but one of the issues that I've run into is that it is quite hard to setup a properly competitive field across different GT3 cars. To address this issue I spent the morning running a bit of analysis to help understand the differences in car performance. The test setup was to run a field of AI GT3 cars at different tracks with different AI settings (BOP data tab at this link: https://docs.google.com/spreadsheets/d/12Fdgze_BNyPbx4apReWVAPI2phUwspdwkCQiqC7S5sw/edit?usp=sharing). What this allowed me to do is run a regression against AI difficulty, car and track to determine two values: For a 1 point increase in AI rating how much faster does a driver get. Independent of track and driver skill how fast is each car. If you look at the image below the Estimate field represents the adjustment in seconds for each track and car (excluding Aston Martin which has an adjustment of 0.0). What we learn from this are three things: This simple model is able to explain 99.8% of the variance in the fastest lap times. A 1pt increase in AI skill reduces lap time by 0.034 seconds (AI coefficient). The Radical is a full 3.9 seconds faster than the slowest car (Porsche Cup GT3) To have a field that is consistently racing with one another you would need an Radical driver with skill of 0 and a Porsche driver with skill of 110 to have these two cars put down the same lap times. It seems to me that BOP here is broken (though it has come to my attention that the cup car is probably a bad comparison). This can be seen in the table below where the "Necessary AI Adjustment" indicates the difference in AI needed for the each car to match the fastest car. However, if you remove the Radical and Porsche from the GT3 mix things start looking much better despite the Mercedes AMG GT3 looking comparatively slow. What is most surprising to me is that the 2019 Audi R8 is slower than the 2018 Audi R8. I hope this is helpful for anyone looking to balance these cars as it has been for me. I'll be interested to see what happens with the upcoming BOP adjustments. Hopefully the cars are more in-line and easier to balance with the AI. Edit: As a follow up I setup the AI using the BOP adjustments here and had the AI run a bunch of laps around Imola. The results were quite good in that the cars seemed to be a bit mixed up in the final rankings. If you run enough laps they will probably converge to some order as we are seeing the McLaren 650s and Bentley start to, but once you randomize the AI skills it should mix it up enough to give you a good spread. AI setting per car below: The high medium and low just gives me a spread for when I setup drivers. There are a bunch of other metrics I end up manipulating, but this is probably a good starting point if you want to run a championship.
Yes, the Cup Porsche is not the same as the straight GT3 porsche. Mercedes has something wrong, it has right from the start, FFB is not similar to the other cars and the AI tends to understeer badly.
