Improving the Realism of Synthetic Clinical Benchmarks Under Utility Constraints
By Omid Bazgir · Paper · cs.AI
Synthetic clinical benchmarks for enterprise AI agents can pass existing utility checks and still remain structurally unrealistic, especially in privacy-sensitive healthcare settings where operational data are hard to access. We study how to improve such benchmarks without breaki