This is interesting. Is nobody training LLMs to repeat resolving issues by patches from real projects where refactoring is required, judging the refactor similarity and not just tests passing?