This is a work in progress and is not to be interpreted as hard and fast rules or laws, but rather just guidelines and observations from the code I've been working on.
Example: a test that I was reading made an assertion that expected an audit count to increase by seven (7). It wasn't obvious at all why the SUT would generate seven audit records. I suppose this exposed another code smell, since a lot of the auditing occurred magically (implicitly via ActiveRecord callbacks), so I would recommend: "don't test implicit, hidden behavior" by testing other things.
This could be filed under Martin Fowler's extract method refactoring.