I was going to suggest performing plate-solving when the guiding star is lost for more than 2-3 frames. IMO adding plate solving to the beginning of the job is not going to solve this problem, as it happens AFTER the job is already started and in progress (in the above example it happened from img20 to img25), and plate-solving EACH captured frame is so resource-intensive.
I don't understand this behavior of the internal guider (IG). I believe what is happening is that the IG was tracking the guide star, then lose it for a frame or two, then another star pops up from the cloud in the periphery of the captured image, IG search and find this new star and ASSUME it is the same guide star it was tracking and tries to bring it to its original location. If we treat the newly auto found star as a new guide star and just keep guiding based on the new parameters (just trust the mount), I think this will fix many problems. Adding a plate-solving option after losing the guide star will assure accuracy.