Instead, adjust the asserts such that they lock down the current (wrong) behavior, and add a clear // TODO: comment explaining what would be the correct result. This prevents such tests from rotting and also catches cases where the behavior is fixed by an unrelated change.

Don't do that, because they are still comments and even worse, the tests are lying. In pytest, there is a way to tell that those tests are expected to fail: xfail, so it will show in the test results every time and it's clear it's just a temporary things and should be fixed.

https://docs.pytest.org/en/7.1.x/how-to/skipping.html#xfail-...