Adjusting steps-per-mm for accurate part dimensions

I'm afraid I don't know the answers for the M92 command. I always save the values in the eeprom via the Repetier menu.

One thing I do know for sure is that the nickle test is way too small to be a good test for X and Y steps. It is hard to make accurate measurements of dimensions that small. I got my best calibration doing something like punching holes and cutting notches in a 3x5 card, scanning it, turning it into a 3D model with inkscape and scad, then printing a replica and scanning that at the same DPI on the same scanner. Then I could see exactly how much I had to scale the scan of the printed object to match the scan of the original 3x5 card.

I doubt the issue is steps/mm. At 700 microns over 20mm, thats likely backlash.

Yup... I remember a discussion similar to this with the z axis in this thread if you're interested.

Basically, you need to check at two points (say, a 20mm and 60mm cube), not just one, so you can determine both the constant and proportional components of your error. Backlash, wall width and the usual other extrusion derps will appear as a constant error, while an inaccurate steps-per-mm will be a proportional error.

If you check at one point (say 20mm), find it's a bit undersized, then increase the steps-per-mm to achieve exactly 20mm, then when you print a 50mm cube I guarantee you it will be oversized now!

I think most people should find that the constant error is much larger than the proportional in pretty much all cases... Those belts are not lying about having a 2.03mm pitch.

Also, calibrating using internal geometry is probably not the best idea, as things tend to come out a bit undersized due to the plastic pulling around the corner.