Questions:
- If I went back and ran the program a few times in a row, what are my tolerances for returning to zero that should make me feel like I can do some nice work with this thing? At what tolerance do I walk (or is this a poor way to judge)?
The only really reliable test that i've seen used is the "circle-square-diamond" test. (sometimes known in a different order like circle-diamond-square, etc.) Couple of big companies of my employ used that test for many years.
In short, you bolt down (near the middle - you'll be using the perimeter) a piece of material - about 12" square and an inch thick, yes, aluminum.
Then you cut a 12" circle about .250 deep. Then you cut a circumscribed diamond (square at 45deg) at .500 deep. Then a square (oriented XY) at .750 deep. Each layer clearing out excess material.
This gives you about a dozen points to carefully check alignment and repeatability. Is the circle exactly tangent to the square and diamond? Is it exactly 12" diameter in every direction? Is the square and diamond exactly 12" across in both directions down both sides? Are the steps exactly .250? Etc. etc.
It's a very thorough and telling test of tracking, columnation, and repeatability.
Obviously, it can be done with any sized piece, but the bigger, the better.
Trouble is, you probably won't get to run anything like that prior to purchase, plus there's a little bit of G-code involved that has to come from somewhere. But were I to get it, that'd be the first chips I'd make so I'd know what i have and not be chasing it all over.
After all, you don't care if it returns to zero faithfully. You care if it cuts where it says it's gonna cut, faithfully.