OfficeQA Pro v2 is a grounded reasoning benchmark by Databricks for evaluating model and agent performance on end-to-end reasoning over real-world documents.
The benchmark consists of question–answer pairs that require reasoning over two centuries of U.S. Federal Accounts of Receipts and Expenditures reporting (1793–2024) — Combined Statements of Receipts, Outlays, and Balances of the United States Government, together with earlier… See the full description on the dataset page: https://huggingface.co/datasets/databricks/officeqa-pro-v2.
