Board Thread:Game Discussion/@comment-35745775-20180602191223/@comment-31433854-20180604131151

I played around a while ago to get a sense of how it worked as I was curious and expecting a Bayesian solution, but it's not. It seems to be mostly a rolling estimate of where you fall in the distribution, based on time/speed/distance and NOT finishing position. But the distribution is skewed and so you get a ceiling effect. There is also a modificiation for the track, e.g. silverstone gets fast bots and melbourne gets slow ones.

The problem with what you are trying to do is what others have elluded to: the AI is much slower than a human player. I'm often in P1 afer a few turns, or certainly before lap 2 against rattlesnack et al, or even TSMs that are at the limit of what the game can model. Now, if you have fast TSMs, they will be way far ahead of the top bots on non-modelled events (speed and distance events).

And Amrosa, for the record, I didn't wreck Marcelsmate. A bot wrecked him and then he wrecked me! It actaully made for an interesting race as then I was down on top speed at panorama and the front bots got a big leap on me. Despite this (are you listening AH), I was in the lead at the elbow on lap 3 (of 4) and posted a lap time several seconds faster than the top, undamaged bot, all w/o cutting, etc.