Sorry I wasn't very clear, so I am using the UCCNC software as my controller (not MACH 3) but I used a MACH post processor when generating the code on my CAM software.

Within the UCCNC software there is a setup area where you zero the axis position, and you enter a value to move, then it moves what it thinks is that distance, then measure the actual table travel and enter it into the software and it calculates the distance per revolution. I had to repeat this a load of times on each axis to get it to down to where i wanted it. it worked out at something like 885 steps per revolution or something with a load of decimals after..

that is literally the only setup I have done though. So I may have missed some other important settings, but nothing jumped out at me.

all the axis use belts and are geared up/down ( i haven't actually calculated the ratios though)

On another note, the UCCNC software allows you to write some code directly into it and run it. When I enter a rapid or feed movement on any axis and run it, it travels the distance i have entered. So the issue only occurs when running a program, which is why I thought it may be to do with the CAM post processor... but if that was the case I guess I would see the same discrepancy on each axis and not random.

I will do a bit more investigating and try what you said about swapping the X and Y signals and see what happens