We should have common tests for clustering metrics including that labels can be permuted (e.g. 0 and 1 swapped) to achieve the same score. General properties such as scores reduce when clustering is not perfect can also be tested.