Thanks! Yeah, I definitely chose the easy way over the 'right' way for this version with just setting simple thresholds.
I'm not sure how well it will work yet, but for a game controller I might try to have users record their own custom motions - as in, "show me a few of your chosen [high punch] motions to match against in-game."
Instead of MediaPipe try training a simple NN with a quickstart asking new players to assume the positions for fine tuning? :P