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