To answer: “Is a variable reward during generalization something I should rethink?”

I don’t think that is something to rethink. I think that rule makes the most sense most of the time, unless there is a specific reason not to follow it.

Sometimes there may be choices that are particularly dangerous to the dog in certain scenarios, where you may want to give the dog a chance to discover a bad consequence in a controlled way. This provides insurance that they will be less likely to choose disobedience in that situation in the future, especially if rewards are not available or are less frequent than the schedule the dog may be expecting.

This can happen even after the initial avoidance conditioning lesson, depending on the chief complaints and concerns a client may have.