Running against a labeled set (categorize support questions), jev gets 95% answers correct. Layla 48%