REAL-FL: Runtime Evidence Augmented LLM for Feature Location

This replication package accompanies the study of REAL-FL (Runtime Evidence-Augmented LLM-based Feature Location), an approach that combines runtime evidence from feature-invoking executions, information retrieval, and LLM-based repository exploration to identify source code entities implementing existing software features. The package contains: Experimental code: Scripts, Docker configurations, dependencies, and supporting tools used to conduct the experiments. Benchmark data: Feature instances, ground-truth information, and feature-invoking scenarios used in the evaluation. Agent results: Recorded experimental results for the Baseline agent, REAL-FL, and REAL-FL without runtime evidence. Information retrieval results: Code entity rankings produced by static and runtime trace-based retrieval methods, including VSM, BM25, and LSI where applicable. The package supports the inspection and reproduction of the experimental evaluation presented in the accompanying paper. Please refer to README.md for further information about the package and its usage, and LICENSE_AND_SOURCES.md for licensing and source attribution details.