The tests I have aren't random because then its difficult to meaningfully communicate the outcome. Instead it each NIP had a test suite, and then each test in the suite attempts to obtain samples from relays and then uses the uses those samples to construct filters than satisfy the respective test.