Thanks for the update! Your workflow looks correct, but I think I might have found a (very subtle!) problem:
If your patterns were created with the label INSULTS, but the textcat.teach recipe specifies the label INSULT, you will not see any suggestions from the patterns (because Prodigy will only look for patterns for the label INSULTS). This would also explain the behaviour you're seeing: all suggestions you get are suggestions from the model, which are pretty random, so it's not actually learning much as you annotate. Pattern matches will show the pattern number (line number) in the bottom left of the annotation card, and also highlight the matched word (a new feature we added after the video was created, to make it more obvious where the match comes from).