The answer to the title question is to do exactly what you're doing-- argue your case. It's quite likely that I've misunderstood a bug or mistaken it for another.

In this case, the date the bug was entered made me think it was entered against the test database before transition, but right after I had restored an old dump, which would have produced symptoms similar to those you mention. In the case of the live database, others have answered you: the time cutoff and result count are parallel limits. You only get the changes that meet both criteria. If there's a way you think this might be better documented, feel free to open a documentation bug or feature request.