과도한 편집은 모델이 필요한 범위를 넘어서 코드를 수정하는 현상

The phenomenon of Over-Editing highlights that accuracy alone is insufficient for AI-driven maintenance, as excessive code changes can degrade codebase quality and hinder developer review processes. Future benchmarks and model training must prioritize minimal, high-fidelity modifications to ensure AI tools are truly effective in complex, existing software environments.