Replication package (Level 1, public) for "Auditing successor selection in software repository mining: Evidence from filtering AI-attributed commits"

Abstract

Level-1 (public) replication package of the article Auditing successor selection in software repository mining: Evidence from filtering AI-attributed commits (H. Jeong, Hanyang University ERICA; manuscript submitted to the Journal of Systems and Software). Contents: the audit tool (Python, standard library only) with its 47 automated tests and a synthetic example; the registered analysis programs, byte-identical to the SHA-256 values registered in OSF hyf36 (cohort A) and OSF 7jasg (cohort B); the exploratory programs with the dated plan of 2026-10-08 and the record-level conformance checker; aggregate results for every table and figure; the deviation logs of both registrations; the 22-study literature evidence matrix; the cohort-B sampling frame and collection programs; and the generators of the supplement and the figures. Run python3 reproduce.py (Python 3.10 or later with numpy and matplotlib). It checks the registered hashes, runs the tests and the synthetic example, and regenerates the supplement tables and the figure data from the aggregate results. With --level2 it also recomputes, from the restricted per-unit files of the Level-2 archive (https://doi.org/10.5281/zenodo.23226980), the registered aggregates and the exploratory aggregates that do not require the raw commit logs, and compares every regenerated file with the shipped result; on Python 3.10.12 with numpy 2.2.6 every comparison passed (numerical values within 1e-9, text files byte-identical). The package contains no account names, e-mail addresses or commit histories. Code: MIT License. Results, documents, supplement and figure data: CC BY 4.0 (see LICENSE-DATA, which also notes the third-party origin of the cohort-B sampling frame).