This isn't really a metric though. It's a formal existence proof that the bug exists. The key difference IMO is that you have to create a test which A) looks (to the maintainer) like it should pass, while simultaneously B) not passing. It's much harder to game.

There are other cases where Goodharts Law fails as well: consider quant firms, where the "metric" used to judge a trader is basically how much money you pull in. Seems to be working fine for them