Hello Ben! Yeah, the engine (at least I hope so!) handles well simple strokes like back, diagonals, etc. If we take a little care to not define gestures very similar in nature, I believe that it can tell well the user gesture without many mistakes.

Just note, however, that if the user designates an area (like drawing a square in a Paint program -- that's what I understood.. right?) he can make the same area via various different paths, which would generate different gestures (like he goes to left and then down, or down and left, or all the way diagonal and then adjusts up on the end...)