Investigating inconsistently failing tests

Nice talk on the subject: