Worst negative pattern I've seen is hyper defensive programing. E.g.

  try:
    something_that_should_not_fail_and_if_it_does_our_assumptions_are_all_wrong()
  except:
    fallback_that_will_not_result_in_correct_behavior_but_make_failure_hard_to_detect()

But, of course, it depends a lot on which models you're using and how you instruct them.