How the Architecture Works

1. Orchestration: 5�8 LLMs (e.g., GPT-5, Claude, Gemini, DeepSeek, Perplexity, Mistral, Grok) are assigned roles and guardrails.
2. Execution: Each module follows its own recipe (debate, sequence, or simulation). Outputs can be anonymous and cross-checked.
3. Scoring: Contradictions, assumptions, risks and evidence are highlighted and, when relevant, reconciled.
4. Synthesis: A visible meta-layer aggregates the strongest signals into an auditable answer.
5. Persona layer: Super-AI, Super-Human, and User+ personas adapt the final framing to the intended decision style.

Every step remains observable so teams can understand why a conclusion was reached.