council-of-high-intelligence: a structured multi-perspective deliberation framework for high-stakes decision making

council-of-high-intelligence: a structured multi-perspective deliberation framework for high-stakes decision making

What it solves

Council of High Intelligence provides a structured framework for making high-stakes decisions by preventing premature consensus and avoiding single-point-of-failure reasoning. It solves the problem of LLM "hallucinations" or biased single-path reasoning by forcing multiple, deliberately different analytical personas to deliberate and disagree before arriving at a verdict.

How it works

The system assigns a question to a panel of AI personas (based on 18 grounded analytical lenses, such as Socrates for assumption destruction or Nassim Taleb for tail risk). The process follows a strict protocol:

  1. Independent Analysis: Members analyze the problem blindly and independently.
  2. Cross-Examination: Personas force direct disagreement and challenge each other's positions.
  3. Synthesis: The system generates a final verdict that preserves dissent, unresolved questions, and "kill criteria" rather than forcing a generic prose consensus.
  4. Evidence Labeling: Claims are categorized as FACT, INFERENCE, ASSUMPTION, or UNKNOWN to maintain transparency.

It integrates as a plugin for tools like Claude Code, Codex, Gemini CLI, and OpenCode, and can route tasks across different LLM providers (OpenAI, Google, Ollama, NVIDIA NIM, Cursor) to ensure a single model family doesn't dominate the deliberation.

Who it’s for

Decision-makers, architects, and product managers who face irreversible choices or complex trade-offs where the cost of a wrong decision is high and a single AI response is insufficient.

Highlights

  • Multi-Perspective Deliberation: Uses 18 distinct analytical lenses to ensure broad coverage of a problem.
  • Adversarial Protocol: Prevents premature agreement through a fixed round budget and enforcement checks for missing dissent.
  • Multi-Provider Routing: Distributes seats across different LLM providers to avoid model-family bias.
  • Structured Verdicts: Produces outcomes that emphasize what remains unresolved and define concrete next steps and kill criteria.

Sources