The reviewer said "keep." The swarm still selected nothing. In a test of Brainstorming Swarm V2, separate agents explored and challenged two designs for an invented fairground. One route looked promising, but it needed shade. The supplied notes proved the location was staffed and the route was step-free. They did not prove it was shaded. A positive critique could not supply the missing fact. So the decision stayed open. That is the method's useful constraint. More agents can give you more angles. They cannot turn agreement into evidence. Try it on a small problem you can check: 1. Give each discovery lens the same facts and a distinct question. 2. Ask for competing approaches, their assumptions, and the conditions each needs. 3. Have a different reviewer challenge each approach. 4. Mark every necessary condition supported, unverified, or contradicted against the source. 5. If a necessary fact is missing, name the next check instead of choosing a winner. The public skill includes the full seven-stage method, templates, a first-run guide, and an optional structural checker. It can run with independent agents where a host supports them. A single-agent run should say so. Separate the lenses. Audit the facts. Keep the human gate. Full essay: https://aris-space.com/documents/workflows/the-swarm-that-refused-to-choose Skill: https://github.com/PUSHINGSQUARES/brainstorming-swarm-v2 https://grok.com/imagine/post/9abd6868-dda0-49d5-a328-6b5a974ab1ac //A<3