Create a tool that simulates reader-grounded feedback loops to replace LLM-as-judge benchmarks for disinformation. This requires building a representative dataset of user perception.