It's like "reverse training" poisoning optimal solution with random values, so yeah, simulating.
Decided to go for a "feeling" of trackmaina training video over a resource-intensive re-creation, so that it runs well in the browser. Coolest surefire solution would be using actual samples from training, but that would skyrocket loading times or feels to same-ish and you can't edit tracks after that
Training against optimal trajectory in every position (instead of distance / lap times) results in same effect as now with a lot more CPU usage