Currently we mostly parse the categories from titles, based on certain keywords. It’s not an exact science. The one exception is “informational”, which comes from the index.json FEP listing. If there were more categories listed in there, I could update it more precisely, but for now I am going based on name keywords.