So, this take is sort of fashionable now, but it’s never really convinced me: what I’d suggest is that, when you discover that the extracted method is a bad abstraction, you do one of two things: (1) inline the method (IntelliJ is great at this for Java) or (2) duplicate the method and rename, adjusting the new version for the new use case. As hard as naming may be, adding the level of abstraction often ends up helping keep each method working at a single level of abstraction.

In my experience, the biggest messes I’ve inherited were a result of not being DRY enough: the cases of incidental duplication I’ve run across have been comparatively easy to unwind.