I'm making a human cannonball simulation in A6.31 by applying a central velocity (phent_addvelcentral) to an entity. For some reason the velocity applied seems to be somewhat inconsistent. Most of the time it is OK but sometimes he massively overshoots the mark.

I have extracted trace data to a file and found that often the velocity that I ask for is not applied. If there is no fricton and no damping, then if I ask for a particular horizontal velocity, that velocity shouldn't change, right? So why isn't it exactly what I asked for? Most of the time it is slightly lower, but sometimes it is much higher, causing him to overshoot horribly. This is amusing to watch, but my human cannonball is getting hurt, and that costs in health insurance premiums. Am I doing something wrong? Is "phent_addvelcentral" unreliable?