I don't doubt it, but obviously something's going wrong between the human-generated training data and the SERP, else why are we getting utter crap back?

(Or, as I said, it's our idea of what constitutes a good result that's wrong).