Replication Package -- Code-to-Prompt-to-Code: An Empirical Study of Prompt Recovery and Adaptation

The replication package accompanies a paper titled "Code-to-Prompt-to-Code: An Empirical Study of Prompt Recovery and Adaptation." It contains everything needed to reproduce the study's results: a main Jupyter notebook that regenerates all figures, tables, and statistical analyses; two Python scripts for recomputing CodeSage embedding-based similarity scores; and a data/ folder with four datasets. These datasets cover 200 sampled real-world Python functions, 8,400 experiment records (testing prompt recovery across 7 information levels, 2 models, and 3 repetitions), 400 records on prompt adaptation across 6 modification types, and precomputed CodeSage cosine similarity scores for both research questions. The figures/ folder holds the output plots comparing models and modification types across both RQs.