Yes, that was the issue. Before I realised that was the problem I tried removing the focuser from my optical chain to prevent it and I got a message (sorry, I didn't capture the logs) that I think said something about autofocus not working, so perhaps it was just a slightly unclear message. Anyway it's resolved now

Hi Paul,

I've just merged MR902 with the following fixes related to this issue...
1. Curve fitting failed on the 11th and last datapoint. I was able to reproduce this and have fixed the issue.
2. Having failed on that datapoint, there is an error in the logic that resulted in focus trying to set the focus position to -1. Hence the large inward movement you observed. This is now fixed.

If you are able to build Kstars/Ekos yourself then pull bleeding-edge (or just git patch MR902 on top of the official 3.6.5 release). If you are not able to build the software yourself then you can take the next nightly build which will contain the fixes.