Research questionHow can LLMs edit another model’s code while preserving functionality and minimizing unnecessary changes?Models trained on different data can have different coding styles, causing an LLM to rewrite more of foreign code than the requested change requires. The challenge is to limit these unnecessary edits without causing build or test failures.