For tests, when in doubt, I don't touch the original tests and just add my own separately. That way I know that the old tests still validate the originally-intended behavior, and my new tests validate what I expect them to do.

If the tests aren't broken, is there ever a good reason to make sweeping changes to them?