D3-Gym

# D3-Gym: Constructing Real-World Verifiable Environments for Data-Driven Discovery D3-Gym is the first automatically constructed dataset of **verifiable environments** for **Data-Driven Discovery**. It contains 565 tasks derived from 239 real-world multi-disciplinary scientific repositories. Each task includes: - a natural language instruction, - an executable environment with pre-installed dependencies, - input datasets and artifact previews, - a reference implementation, - and an automatically generated evaluation script. The present dataset contains the metadata for each task. To access the actual training environments, please use our <a href="https://hub.docker.com/repository/docker/hananemoussa/d3-gym/general">Docker Hub Repository</a>. Full instructions can be found on our <a href="https://github.com/OSU-NLP-Group/D3-Gym">GitHub repository</a>. ## Disclaimer Repositories used in the creation of D3-Gym are under permissive licenses. We provide a full breakdown of licenses below. There are also 39 repositories that do not provide any license information; we assume these permit use for research purposes. ### License Distribution | License | Count | |------------------------|-------| | MIT | 99 | | GNU (GPL, AGPL, LGPL) | 43 | | None | 39 | | BSD | 29 | | Apache | 22 | | CC | 4 | | ISC | 1 | | Custom | 2 | | **Total** | **239** | ### Custom-Licensed Repositories - BrainIAC - DeepDelta --- ## Citation If you find our paper or resources useful in your work, please cite us: ```bibtex @article{d3gym2026, title = {D3-Gym: Constructing Verifiable Environments for Data-Driven Discovery}, author = {Hanane Nour Moussa, Yifei Li, Zhuoyang Li, Yankai Yang, Cheng Tang, Tianshu Zhang, Nesreen K. Ahmed, Ali Payani, Ziru Chen, Huan Sun}, journal = {arXiv preprint arXiv:2604.27977}, year = {2026}, url = {https://arxiv.org/abs/2604.27977} }