As suggested by @Cycologista here, we could aggregate all button press lengths and check if we're seeing something bimodal and whether it's clear enough to use as a cut-off. The problem is that of 2 overlapping distributions (one for long presses, one for short presses). If the overlap is significant, we'll have big classification errors.
Even if this might not to serve as a reliable tell, it might at least increase the chances that our guess between oncoming/overtaking is right.
As suggested by @Cycologista here, we could aggregate all button press lengths and check if we're seeing something bimodal and whether it's clear enough to use as a cut-off. The problem is that of 2 overlapping distributions (one for long presses, one for short presses). If the overlap is significant, we'll have big classification errors.
Even if this might not to serve as a reliable tell, it might at least increase the chances that our guess between oncoming/overtaking is right.