Trick or Treat: Putting Peer Prediction to the Test