That’s a great question, but it’s a difficult one. As I mentioned, we started with the harmonious side of the pairing spectrum because there are more rules, per se. We evaluated pairs with criteria like structure, x-height, proportions, etc. and then ranked the data: bad, somewhat bad, somewhat good, and good. Pairs that were ranked with four stars had a lot of harmony for example—a good pair. We’re evaluating which pairs are favorited and which get a thumbs down. I’m excited to see how the machine learning models and data results get better over time! — Terrance Weinzierl