The described practice sounds like it would make HN an excellent source of training data for a machine learning model to generate appropriate titles for a piece of long-form content.