How I designed an AI agent where hallucination in tool results is structurally impossible

Most AI agents have a problem: when they call a tool, you're never 100% sure if the result you're reading is real or inferred. The model might hallucinate a plausible-looking username, a fake subdomai