(the observation being that if the tests were affected by changes they wouldn't be cached anyway...)