Open Code, Hidden Collections: An Empirical Study of Academic Software Stewardship

Abstract

Academic institutions produce a substantial and growing digital collection of open source software artifacts, yet these repositories often lack the metadata, licensing, and community infrastructure necessary for discovery, reuse, and long-term stewardship, core concerns of digital libraries. University-based Open Source Program Offices (OSPOs) are natural stewards of this distributed collection, but lack systematic tools for cataloging what their institutions are building, assessing health, and identifying where support is needed. This paper presents a large-scale empirical study of open source stewardship practices across academic institutions, developed with input from academic OSPO practitioners and motivated by their operational needs. Using the 32 universities and research institutions of CURIOSS (Community for University and Research Institution OSPOs), a network of academic OSPOs spanning North America and Europe, as a case study, we build an automated pipeline that combines the GitHub API with LLM-based repository filtering and classification. Applied across all 32 institutions, the pipeline yields a curated collection of 216,920 affiliated repositories (90–92% accuracy) from 761,920 candidates. Analysis of this collection reveals significant stewardship gaps: only 23.4% of repositories carry machine-detectable licenses, making the majority legally unreusable regardless of code quality, a fundamental barrier to the core digital library mission of enabling access and reuse. Community engagement infrastructure is nearly absent: Contributing Guides appear in fewer than 2% of repositories and Codes of Conduct in under 1%, despite 84% having a README. These gaps persist even among high-visibility projects. Project type classification reveals the collection is dominated by software development (45.6%) and educational projects (39.4%). The pipeline is open source; the resulting dataset has been shared with CURIOSS member OSPOs and released to the research community, alongside an interactive dashboard enabling per-institution collection health assessment.