I think what we’re testing for here is LPM output that hasn’t even been skimmed by a human, let alone paraphrased.