Do developers have agency? 7.3TB of GitHub data (66k projects) shows that the growth of large projects was resilient to external changes for decades.

An analysis of ~7.3TB of GitHub data across 66k projects found that large projects (>700 commits, 16.1% of the dataset) exhibit remarkably stable growth trajectories resilient to tooling changes, including early LLM adoption. The remaining 83.9% of smaller projects show less stable, decelerating growth, raising concerns about popularity bias when training LLMs or building learning materials from GitHub data indiscriminately.