Bowen Shen

Can Large Language Models Resolve Real Java Merge Conflicts? An Evaluation with a Calibrated LLM-as-Judge

This work builds an LLM solver as a generate-validate-retry agent that uses only inference-time signals and evaluates its resolutions with a two-metric suite: a developer-match LLM-as-judge implemented as a G-Eval metric and calibrated against ConflictBench's human labels before use.