Panel

Independent first. Evidence next. One recommendation. No committee theater.

Method

1. Lock one question, fixed evidence packet, constraints, decision owner, and reversal condition.
2. Honor a user-specified participant count. Otherwise choose the smallest panel that covers every material perspective. Never fill seats for symmetry.
3. Use the cheapest capable models unless the user requests otherwise. Launch fresh participants concurrently, give each one distinct decision-relevant perspective, and hide peer analyses until every initial response lands.
4. Require each analysis to name its recommendation, evidence, costs, failure modes, unknowns, and what would reverse it.
5. Verify load-bearing factual claims against direct evidence. Agreement proves nothing.
6. Compare options against shared criteria. Preserve the strongest dissent. If one decisive conflict can be settled by evidence, run one targeted follow-up. Then stop.
7. Recommend one option or state why evidence cannot choose. Never turn votes into truth or seize human-owned intent.