This might work when the test really describes&tests the business rule, not when the test simply contains a mirror of the implementation with everything replaced by mocks and just checks that the implementation is what it is, conditioning all people changing the code in the future to always have to change the test as well.

On the contrary! I found out that a rewrite from scratch leads to much better code and abstractions, as you understand the problem space better. (On the other hand, beware of http://catb.org/jargon/html/S/second-system-effect.html)

I wanted to publish a tiny utility I created to GitHub (you know, it might be useful to someone else…). Before that, I wanted to some cleanup, rebasing/squashing a bit, etc. In the middle of that:

$ git checkout featurebranch
The following untracked working tree files would be overwritten by checkout:
.gitignore
.idea/…
etc...

Oh, sure, no problem…

$ rm -rf * .*
$ git checkout featurebranch
fatal: not a git repository (or any of the parent directories): .git

D’oh! (Never mind, it probably wouldn’t have been useful to anyone else, anyway.)

Its written vibe.