ZK-AgentGuard Synthetic Medical Triage Dataset for Policy-Compliant AI Evaluation

This dataset is designed for evaluating policy-compliant AI decision systems using zero-knowledge proofs. It simulates a medical triage scenario where an AI agent predicts patient urgency levels (Low, Medium, Emergency) based on structured clinical features such as age, heart rate, systolic blood pressure, and a derived risk score. The dataset is constructed to support the enforcement of safety and compliance policies. In particular, it enables validation of constraints such as restricting emergency classification for low-risk patients. The dataset is fully structured and suitable for arithmetic circuit encoding, making it compatible with zero-knowledge proof systems. This dataset is used in the evaluation of the ZK-AgentGuard framework for verifiable AI decision-making.