At least one useful take-away: k-modes clustering. I initially thought it was a typo/miscommunication, but the soybean reference seemed bizarrely specific. It apparently actually exists, and is an extension of k-means to categorical data, using Hamming distance instead of Euclidean distance.