Replication Package for "Think Less, Code Better? An Empirical Study of Reasoning Efficiency in LLM-Based Software Engineering"

Replication package for the paper Think Less, Code Better? An Empirical Study of Reasoning Efficiency in LLM-Based Software Engineering. The package contains the pipeline code, SLURM templates, benchmark manifests for HumanEval Pro, MBPP Pro, and BugsInPy, all generation records of the five RQ1 runs, the RQ2 codebook and annotation label tables, the RQ3 stop points and regeneration records, and the pre-computed statistics used in the paper. All reported statistics can be recomputed from the bundled records with numpy and scipy. See README.md.