Instead of checking if data exists (and in correct state), we have been using a unique identifier for each test that can search for all related data in the db and blast it at the start of the test. Then we create exactly what we need in each table.