I call this Random Normal as it effectively generates randomized unit vectors uniformly distributed. Instead of using x = random(2)-1, y = random(2)-1, z = random(2)-1 then normalized.. which is rather poor in many reasons. We compute the euler angles for theta and phi. It works on a nice property of arc-cos which is a uniform distribution of the 'y' axis just by using -1 to 1. as you can see by the rotating sphere.. no polar bias. lovely.