Is your feature request related to a problem? Please describe.
The commentary is very serial right now; that is, it picks an event from the stack, reports it, then moves on to the next. This is evident when multiple cars are stopped on track at the same time, for example. The commentators will call off each stoppage separately, whereas real commentators would almost certainly report it as a single event if they were all on the same part of the track.
Describe the solution you'd like
Commentators should be aware of all of the events on the stack. This, of course, will require more careful stack cleanup, but the payoff of the commentators recognizing related events is worth it.
Describe alternatives you've considered
Alternatively, rather than letting the text generation determine when events are related, methods could be written to group events together. For example, a method could be written to detect two cars stopped at the same corner. I think this approach should only be taken if allowing the text gen to determine it is significantly terrible, though.
Additional context
This, along with image prompting, could also help with things like crash detection. Even if a crash isn't directly detected, if two cars are stopped, and the screenshot shows damaged cars, the commentary may very well fill in the blanks.
Is your feature request related to a problem? Please describe. The commentary is very serial right now; that is, it picks an event from the stack, reports it, then moves on to the next. This is evident when multiple cars are stopped on track at the same time, for example. The commentators will call off each stoppage separately, whereas real commentators would almost certainly report it as a single event if they were all on the same part of the track.
Describe the solution you'd like Commentators should be aware of all of the events on the stack. This, of course, will require more careful stack cleanup, but the payoff of the commentators recognizing related events is worth it.
Describe alternatives you've considered Alternatively, rather than letting the text generation determine when events are related, methods could be written to group events together. For example, a method could be written to detect two cars stopped at the same corner. I think this approach should only be taken if allowing the text gen to determine it is significantly terrible, though.
Additional context This, along with image prompting, could also help with things like crash detection. Even if a crash isn't directly detected, if two cars are stopped, and the screenshot shows damaged cars, the commentary may very well fill in the blanks.