When you wonder why your algorithm refuses to perform better than 90% on one set of data and 75% on another, and you realize that the concepts are supposed to be numbered from 0 instead of 1. Yes, the first dataset had 10 concepts and the second 4. #OffBy0