> nothing to do with how people are graded

If by "nothing to do with" you mean "uncorrelated", then this is very trivial and empirically untrue. There are a lot of tests that correlate with peer and manager-rated performance in software development (and many other roles).

Even the dumbest most ad hoc unrelated tests will have nonzero correlations, and the well-studied principled ones much more so.

The better argument is that tests are high effort and can feel demeaning. But it's not productive to pretend they don't work.