Researchers propose fine-grained confidence calibration methods for Large Language Models (LLMs) in automated code revision tasks, improving calibration error and enabling more trustworthy model usage.