Paper with code diffusion on GitHub: Disruption or consolidation?

Papers with code have become increasingly prevalent in disciplines such as computer science, yet the relationship between code diffusion and scientific disruption or consolidation remains insufficiently understood. This study examines 29,900 computer science papers published between 2009 and 2024 that contain GitHub links, employing repository indicators such as Stars, Watching, Forks, Issues, and Pull Requests to capture code diffusion in terms of collection, attention, application intent, discussion, and maintenance. The number of contributors is used to represent the size of the development team. We constructed a citation network to calculate the CD index and assessed the relationships among all variables using logistic regression and interpretable machine learning. The results show that code collection, attention, application intent, and discussion are primarily associated with scientific consolidation. In addition, the negative association between application intent and disruption becomes weaker when code maintenance is taken into account. Contributor counts exhibit only a limited effect on consolidation, operating mainly through team size. Overall, the study highlights the value of incorporating repository-based metrics into research evaluation frameworks, offering a more nuanced understanding of how open-source code relates to the trajectory of scientific progress.