Glad people are still enjoying this. I found it really difficult to get started making this sort of stuff, and was glad to be able to locate sources for previous similar projects by others (mouaccel and moufiltr). I've always made the source for the available in the hopes that others can improve on what I've basically hacked together through trial and error.
There are always going to be precision limitations, especially with regards to angle manipulation. Input angle is calculated on each report, so low CPI and/or slow movement (eg: 0x1, 1x0, 1x1) are going to give extremely 'wrong' results. Configuring your mouse to use the highest useful CPI and lowest acceptable polling interval will help. A cleverer person could probably make it sample numerous reports over some small time window to narrow down the angle a little better, but I'm sure there's some kind of horrible downside to that too.