Mundi Orch — Multi-LLM Debate

Overview

Cross-model adversarial debate. Takes a proposition (architectural choice, hire, acquisition, etc.), assigns opposing sides to different providers (Claude + Gemini + GPT), and runs multi-round structured debate. Distinct from /mundi:debate (Claude-only chat.json) in that the providers have genuinely uncorrelated priors.

Related to mundi-orch-multi-llm-consensus — consensus is "agree-or-flag-divergence", debate is "actively attack each other's reasoning."

When to use

• Evaluating a contentious proposal where you need the strongest case for and against
• Pre-mortem on a decision Claude already recommended ("Gemini: argue why this fails")
• Architectural tradeoff reviews (microservices vs monolith, build vs buy, etc.)
• Legal / ethics / compliance edge-cases
• Board prep: "what will the skeptical director ask?"