Researchers often train and evaluate software engineering models on open-source code, which may not reflect the practices, dependencies, and scale of industrial systems. The Curated Industrial Developer Repository (CIDR) is a new dataset built from repositories contributed by industrial partners, providing a rare look at proprietary-style code at scale.

According to the paper, CIDR comprises 4,687 repositories spanning 84 programming languages. The authors describe it as a large-scale resource for software engineering research, with the goal of enabling studies that require realistic, industrial-grade source code rather than only publicly available projects.

Because the dataset is drawn from multiple industrial partners, it offers diversity in coding styles, project structures, and technology stacks. The paper does not provide details on licensing or access restrictions in the abstract, but the availability of such a dataset could help bridge the gap between academic benchmarks and production software development.