What's the opposite of skipping steps?

My 3 axis router developed an interesting issue a while ago and after a lot of experimentation I'm pretty sure the problem is with the 7i96s.
The first thing that went wrong was that the X axis started moving more than it should. According to math the steps/mm should be 480, but suddenly something decided that about 455 is the correct number. On it's own, that wouldn't be much of a problem, but it seems to occasionally switch to a different number and loose its position.
I have tried the following:
- different acceleration settings (it happily did 3000mm/s2 for a while, returning to within 0,01mm, but sometimes lost steps at 250mm/s2) 
- different stepper drivers, both open and closed loop
- different wiring between the 7i96s and driver
- new linuxcnc installation
- flashed the firmware to the 7i96s again
- completely new configuration through pncconf
- moved x axis from stepgen 0 to stepgen 4

The other axis work perfectly everytime and for about a year so did X, at 480 steps/mm. Z axis has the same motor, driver and encoder as X.
I have access to a 7i96 and I will try it out on friday, but I still want to ask if anyone has a different idea for what could cause this?

If the problem stays with the X axis with motor driver and 7I96S channel changes,
it really sounds like the issue is more likely to be mechanical than electrical.

I thought about that, so I marked the position of the pulley on the motor and ballscrew, but they didn't move and the belt didn't skip. Also the ballnut mounting and the bearings on the fixed side of ballscrew have no backlash.
I think the primary problem is the 455 steps/mm. The microstepping is set to 3200, the motor pulley is 20 teeth, the ballscrew pulley has 30 teeth and the ballscrew pitch is 10mm. That should be 480 steps/mm, but if I set it to that it moves about 5% farther than it should. I should somehow count the pulses coming out of the 7i96s, but I haven't been able to convince my oscilloscope to count reliably. Is there something in halscope that would show that?

If you want to, you can loop back a step/dir channel to the 7I96S's encoder
Step to A and DIR to B  with the encoder set to UP/DOWN  =  counter mode

This is how the hardware Step generator was originally tested

Thanks, that sounds like exactly what I need. Do I need to connect it electrically or can it all happen in hal?

Needs to be done with physical wiring