RecallRadar Intelligence published the harness and question set for Luna at max, with Luna at max spending so long thinking it ran out of room to answer and four blank responses.

The recall that takes you down didn't start with a recall. It started six months earlier in a warning letter to a supplier you forgot you used. Or a 483 observation on a contract manufacturer two states over. Or an EPR deadline that triggered a packaging change nobody flagged. That data is public. Also buried across FDA, USDA, EPA, CDC, and 50 state portals. None of them talk to each other. Our research agents pull from all of them simultaneously. They cross-reference warning letters, import alerts, 483 observations, enforcement actions, and state rule changes across 5.9 million records. They find the connections a manual search would miss and surface the sources that matter to your operation. Then a human reads every word before it reaches you. No hallucinated citations. No orphaned links. Every source verified, every finding checked against the original regulatory text by an analyst who's worked the floor, the lab, and the land. We tested it. Ranked #20 globally on the DeepResearch Benchmark, ahead of OpenAI, Gemini, and Tavily. AI speed. Human verified. Recalls don't blindside teams watching the right signals. RCRI is taking clients. rcri.io