Whenever I get faulty results with composer (cursor rules not followed, concurrency not properly done) and point it out to the model, it tries one more time. When I then again point out a mistake in the suggestion by composer, it just stops answering, being ashamed, instead of reverting mistakes or fixing the errors.
Its kind of funny, if it weren’t for the efforts needed afterwards to untangle in some cases.