Hopefully this explaines better how names are determined when running into test failures.

Mentioned in #2291.