I wonder if it would be possible to do a repeatability check over a longer distance? Set the DTI to zero against a fixed block (allowing for backlash), move off a fair distance, then move back. Repeat many times and see if there is any drift. I'm wondering if there are any lost steps due to pulse timing issues (dir/step pulse timing, in particular) or maybe even triggering on the wrong pulse edge which can give cumulative errors after significant numbers of reversals or start/stop motions. As you didn't build it yourself, then someone else might have just fiddled until it kind-of worked without really understanding what they were doing.

Interesting - mechanically-minded folk go for mechanical issues, electronics types go for the sparky bits! Anyway, that's a relatively quick one to try out with a bit of hand-written gcode.