metaphilabs/credit-underwriting-preview

Credit Underwriting Benchmark A 1387-task RL environment for financial document understanding. Agents are provided with business documents (bank statements), and asked to assess the risk associated by using a context-rich classification taxonomy for each transaction . The agent's output is programmatically graded against the ground truth labelled by domain epxerts with over 10+ years of experience in the risk assessment field. Scale 1387 qualifying cases… See the full description on the dataset page: https://huggingface.co/datasets/metaphilabs/credit-underwriting-preview.