Differential wiring, if the hardware allows, is much more noise-resistant, which can help avoid false stepping and other interference issues. In practice, standard single-ended connections usually work OK, but I would always go differential if it's possible. Why the step pulse polarity is different on one axis (if it is...) is a question only you can answer! But more seriously, it's so quick to check that it's worth having a look.

I found this a frustrating fault. I had been using my CNC router very happily with no problem until I did a bit of detailed 3D engraving with it, and found that the cutter slowly seemed to sink further into the work. All the time, Mach3 was recording the correct depth. Obviously, the fault had always been there but with simple profiling cuts where Z is over-depth anyway, a tiny drift was never going to show up but with lots of direction changes, it was a different story. What you have shown looks very familiar but in Y.