Our paper DARE-bench: Evaluating Modeling and Instruction Fidelity of LLMs in Data Science is accepted to ICLR 2026. [arXiv] [code]