I think we are misusing the term accuracy here. This particular machine does not have to be accurate or precise (within reason) but repeatability is paramount for the basic PnP functions. The thing will be traveling from point A to point B and it does not really matter how straight and true that motion is. What is important is that the business end hits the destination coordinates every time within 0.1mm. Does not have to be linear motion either - it's just cutting air so to speak.
I'm saying all this assuming that the machine that Eclipze is building will have some teaching function when you can define those points A and B manually for every part. Typically they use down looking camera with crosshair.
On the other hand, the industrial machines must be more or less accurate as they rely on motion linearity a lot. While many machines have manual teaching, the destination point of every placement motion is getting calculated by the computer based on the fiducial correction and part position correction factors.
I don't think Eclipze is going after fiducials with this machine or going to calculate the coordinates on the fly for every part. With this in mind, I'd suggest you target repeatability and not precision of the steppers. In other words, there is no reason to use 10x microstepping trying to get 0.01mm positioning precision if you cannot repeatably position the machine due to sloppy bearings, belts or weather fluctuations. Keep in mind that with high microstepping rates your torque reduces dramatically and the step loss probability increases. Try it with 1x or 2x stepping and see if repeatability is there. If it's not, solve that problem first and then go after higher precision.
For reference, on my machine I was playing with stepping parameters and ended up using 0.02mm/step and doing it throug "macrostepping" if there is such a term. Essentially, the servo drive has software gearing function when it can subdivide incoming step/dir pulses (microstepping) or it can multiply them. In my case I'm using 200x *multiplication* because the native travel distance of the linear servos is 100nm/step and that's a little bit too precise. That's nanometers...