Writing a spec for something that will be wrong

The acceptance criteria box stops making sense. What replaces it is examples, thresholds, and a plan for the data you do not have yet.