I noticed that too, just didn't have the Time to fix it yet, sorry.
I'm assuming its because I'm currently directly sampling the Mouse location, and when you stop at the End, it normalizes into seemingly random direction due to both values being tiny/near-0.