Compare two setups

A setup is one way of producing labels: a human questionnaire version, or one model under one prompt recipe. Choose two and see how many of the same 3,000 tweets end up labeled differently.