I think you mean focusing in one FOV and then slewing to another FOV with the target in it. If so, that is the way FocusMax/ACP does it but they also need a “good” single star. My take is that involves too much mucking about and I was not fond of that method when I was using ACP. Focusing in the target FOV like SGP does is much better, IMHO.
That is why I suggested a user-selected subframe (and this selection mechanism already exists for “frame and focus”) and allowing the user to use this subframe during autofocus. As long as the subframe was nowhere near the object that is causing the focus to fail, dithers and such should have little or no negative effect and the user could select a region with an appropriate level of usable focus stars. The subframe would need to contain enough stars to be valid but this would be up to the user and would be better, in any case, than having an object in the focus region that causes fails.
It was pretty bad last night on two globulars (M3 and M5 at 1000 mm with my STT-8300), nearly impossible to get a reliable autofocus regardless of what I modified. I tried modifying pretty much everything one can including sample size, nebula filtering, filter used for focus, and exposure time. It maybe gave decent focus one out of 5 times and it was not predictable or reliable. As soon as I moved the cluster just out of the FOV, it started working great again so I know that was the issue.
Of course one could also take a reverse approach and allow the user to select an “exclusion zone” where SGP would not look when doing focus, but it seems more logical to use the existing subframe mechanism.