Replication Package for "Hidden Dependencies in LLM-Enabled Code Generation: An Empirical Study of Regressions and Escaped Failures"
Authors/Creators
Description
Replication package accompanying the ESEM 2026 paper "Hidden Dependencies in LLM-Enabled Code Generation: An Empirical Study of Regressions and Escaped Failures" (Leibniz International Proceedings in Informatics, LIPIcs, Vol. 394).
The archive contains the experiment configurations; frozen task-selection records and rendered experimental conditions; raw model generations; evaluation records for the three oracle layers (O1, O2, and O3); archived pre-correction evaluator snapshots retained for sensitivity analyses; analysis and integrity-checking scripts; result tables; the figures used in the paper; run-family provenance records; and the post-acceptance audit records PA-00-PA-06 together with their recorded outputs and hashes.
Start with README.md. REPRODUCE_CORE_RESULTS.md gives the shortest path to the main quantitative results, REPRODUCE.md describes the full reproduction workflow, and CLAIMS_TO_ARTIFACTS.md maps each quantitative claim in the paper to the corresponding scripts, inputs, and outputs.
Source code is released under the MIT License. Derived data are released under the Creative Commons Attribution 4.0 International License (CC BY 4.0) unless a more restrictive upstream license applies. Materials derived from HumanEval, MBPP, CodeRAG-Bench, and EvalPlus remain subject to their respective upstream licenses. Model weights are not redistributed; public model identifiers and the available provenance records are included in the archive.