One question, six voices, one decision brief.

1. Step 1 Ask One hard question — yes/no or contested claim.
2. Step 2 Debate Opening, rebuttal, then closer or battle — on your schedule.
3. Step 3 Crux Judge names the load-bearing assumption and mind-change triggers.
4. Step 4 Brief Truth Factor, Assumption X-Ray, and an exportable Decision Brief.

1. You ask one hard question

The product is not designed for chit-chat. It is designed for the question you would normally avoid oversimplifying. Yes-or-no framing is fine. So is a contested claim you want stressed.

2. Six speakers stream in parallel

• Mara Power Attorney · Strategy and Precedent

  High-stakes counselor. Stress-tests claims for enforceability and institutional fallout.

• Quinn Business Titan · Incentives and Execution

  Operator-investor lens: unit economics, ownership, incentives, what scales in markets.

• Reid Screen Icon · Culture and Persuasion

  Narrative strategist. Trust, legitimacy, and how ideas survive contact with audiences.

• Sloan Tech Magnate · Scale and Systems

  First-principles builder. Mechanism design, scaling failure modes, compounding advantage.

• Pax Contrarian Thinker · Incentives and History

  Cross-disciplinary skeptic. Surfaces hidden assumptions and second-order effects.

• Atlas Heavyweight Closer · Blunt First Principles

  Hard-nosed generalist. Pressures every claim until only testable assumptions remain.

3. Three rounds, on demand

• Opening — each voice stakes a position with concrete support.
• Rebuttal — every voice reacts to every other voice and concedes one earned point.
• Closer or Battle — closer pins the load-bearing disagreement; battle is ruthless line-by-line attack.

You can stop after the opening. You can reply to a single speaker. You can ask for the crux at any point after a round completes.

4. The crux

A neutral judge view names the single dispositive question — the one assumption that, if resolved, ends the debate — and what evidence would change each speaker's mind. The crux becomes the anchor for the final brief.

5. Assumption X-Ray and Decision Brief

After the crux, local analysis extracts what each side is betting on, the most fragile assumption, the evidence trigger that would change the debate, reversible and irreversible risks, and a practical next move. This is local argument analysis, not factual verification.

6. Truth Factor

Local 0–100 score for argument quality, broken into Support, Balance, Pressure, and Clarity. It is not a fact-check verdict — it measures the shape of the argument.

What runs in the browser, what runs on the server

• Six parallel SSE streams from one of Anthropic, Google, OpenAI, Groq, or xAI — selected per speaker.
• With no provider keys, local fallback mode streams deterministic debate text for $0 demos and visual QA.
• Decision Brief, Assumption X-Ray, and Truth Factor run locally and do not require extra provider calls.
• Browser-supplied system prompts are not accepted. The server builds prompts from a structured contract.
• Voice playback uses short-lived signed grants tied to the exact generated text.
• No database. Snapshots live in your browser unless you turn on Private session.