The problem there is you made it too simple, lemme try to squiggle out a drawing of what i mean:




Your example there is more of a job for the multiple buttons on each side algo; either way, you forget about the time dimension, at any given moment you always know at least one side for sure because of where the cursor moved to before.



Regarding the accuracy of the thumbstick while combined with buttons, i don't know how to program well enough to actually code a proof of concept to test the limitations, very preliminary tests using the accuracy thingy in Healthcheck seems promising, some sort of filtering will probably be needed, but it looks like it's worth further investigations.