thanks for the very high signal comment. Appreciated! I'll probably be picking this back up again over the weekend.
My plan is switch out the exponential decay I have now for a kinemetics based model (e.g. position = speed * t + acceleration * t/2) and to fix some of the bugs Warren mentioned (e.g., not dealing with changes in direction properly, not cancelling the scroll on single finger events).
After that we can re-evaluate.