In many cases you cannot see the correct abstraction without introducing the duplication back. When working with particularly messy code I often do sort of https://en.wikipedia.org/wiki/Karnaugh_map of important variable states to see what actually happens before I can refactor it.

This is basically introducing the duplication back.

Whether you keep the duplicated code or refactor it in a different way is another question, what matters for the "duplication is cheaper than wrong abstraction" to be true is just the fact that by introducing abstraction early you wasted time refactoring one way and back. Refactoring isn't free. So in fact leaving the duplication there would have been cheaper - Q.E.D.

It doesn't mean you should never risk it, but it does mean you should think hard before you do it.