Real captured outputs and agent demos from AI models on healthcare ops prompts. Every score has a receipt.