The metadata-updates.sh test script doesn't really test anything particularly interesting, and appears never to fail unless all of the tests in that session fail, which indicates to me that anything of interest that it might have tested has already been caught by some earlier test. It has not been updated for anything except test-framework changes in many years. The only minor part of interest is write_disjoint, but that is also tested by parallel-scale (along with more MPI tests).