Artifact of Hot Fixing in the Wild

This Zenodo artifact accompanies the paper “Hot Fixing in the Wild” and contains the scripts and analysis outputs used in our study of automated and human-assisted pull request (PR) relevance assessment for software hot fixes. Contents The artifact includes the following files: main.pyThe main analysis script that processes issues, pull requests, commits, and repository metadata. It implements the logic for classifying critical vs. non-critical issues, computing temporal metrics, and extracting PR- and commit-level statistics used in the paper. prompt_llm.pyCode for constructing and executing LLM prompts used to automatically assess PR relevance. This script supports the reproducibility of the LLM-based relevance classification described in the paper. HotFix_Master_Manual_Checking-anon.xlsxAn anonymised spreadsheet containing human annotations of PR relevance. These labels serve as ground truth for comparison against automated (LLM-based) relevance judgments. human_relevant_prs.san.pngA visualisation summarising the human PR bag of words. bot_relevant_prs.san.pngA visualisation summarising the bot PR bag of words. clean_csv.shA helper shell script used to clean and normalise CSV inputs prior to analysis. Data Sources The analysis relies on the publicly available AIDev dataset, including issues, pull requests, commits, and commit-level file changes. Large raw datasets are not redistributed in this artifact; instead, they are accessed programmatically (e.g., via Hugging Face datasets) as described in the paper and code. We also ommit including derived data directly for ethics consideration and instead include the scripts we used. Reproducibility To reproduce the results: Ensure a Python environment with the required dependencies (e.g., pandas). Run main.py after configuring dataset paths as described in the code comments. All scripts are provided in anonymised and self-contained form to support transparent evaluation while preserving data privacy. Intended Use This artefact is intended to reproduce the experimental results in the paper.