-
Notifications
You must be signed in to change notification settings - Fork 0
Debate Methodology
dev-mondoshawan edited this page Apr 15, 2026
·
1 revision
The Dissensus AI Debate Engine employs a rigorous 4-phase dialectical methodology designed to surface hidden assumptions, challenge weak arguments, and forge consensus through structured adversarial debate.
┌─────────────────────────────────────────────────────────────────────┐
│ DEBATE LIFECYCLE │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ PHASE 1 PHASE 2 PHASE 3 PHASE 4 │
│ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ │
│ │ 🤔 │ → │ 🎤 │ → │ ⚔️ │ → │ ⚖️ │ │
│ └─────┘ └─────┘ └─────┘ └─────┘ │
│ │
│ Independent Opening Cross-Examination Final │
│ Analysis Arguments & Rebuttal Verdict │
│ │
│ (Parallel) (Sequential) (Adversarial) (Synthesis) │
│ │
│ ~30-60s ~60-90s ~90-120s ~60-90s │
│ │
└─────────────────────────────────────────────────────────────────────┘
Total debate time: ~4-6 minutes depending on model and topic complexity
Before the debate begins, each agent forms their own independent assessment of the topic. This prevents groupthink and ensures each perspective is developed authentically.
| Aspect | Details |
|---|---|
| Concurrency | All 3 agents run in parallel |
| Duration | ~30-60 seconds |
| Visibility | Agents do NOT see each other's analysis |
| Output | Private 2-3 paragraph assessment per agent |
TOPIC FOR DEBATE: "{topic}"
You are in Phase 1: Independent Analysis. Analyze this topic from your
unique perspective. Provide your initial assessment in 2-3 focused
paragraphs. Identify the key questions, risks, or opportunities you see.
This is your private analysis before the debate begins.
Be specific and substantive. Reference real-world data, trends, or
precedents where relevant.
| Agent | Focus |
|---|---|
| CIPHER | Identifies hidden risks, failure modes, unstated assumptions |
| NOVA | Identifies opportunities, catalysts, asymmetric upside |
| PRISM | Identifies key questions the debate must resolve, initial balance assessment |
Each agent's Phase 1 output is stored and will be referenced in Phase 2:
debateContext.phase1[agentId] = result;
Each agent presents their formal position on the topic. This is their "opening statement" to the tribunal — their best case before facing cross-examination.
| Aspect | Details |
|---|---|
| Concurrency | Sequential (CIPHER → NOVA → PRISM) |
| Duration | ~60-90 seconds total |
| Context | Each agent sees their OWN Phase 1 analysis |
| Output | Formal structured argument per agent |
TOPIC FOR DEBATE: "{topic}"
Your independent analysis was:
{debateContext.phase1[agentId]}
You are now in Phase 2: Opening Arguments. Present your formal position
on this topic. Structure your argument clearly with:
- A strong thesis statement
- 3-5 key supporting points with evidence
- A clear conclusion
This is your opening statement to the tribunal. Make it compelling,
specific, and well-structured. The other agents will challenge you
in the next phase.
Each opening argument should include:
| Element | Description |
|---|---|
| Thesis | Clear position statement |
| Key Points | 3-5 supporting arguments with evidence |
| Conclusion | Summary of position |
Opening arguments are sequential (not parallel) to:
- Create dramatic tension
- Allow each agent to establish their position distinctly
- Build narrative momentum toward the clash in Phase 3
The heart of the dialectical process — agents directly challenge each other's arguments. This is where weak reasoning is exposed and strong arguments are stress-tested.
| Aspect | Details |
|---|---|
| Structure | 3 sequential exchanges |
| Duration | ~90-120 seconds total |
| Flow | CIPHER attacks → NOVA defends/counters → PRISM judges both |
TOPIC: "{topic}"
NOVA's Opening Argument (The Bull Case):
{debateContext.phase2.nova}
You are in Phase 3: Cross-Examination. You've heard NOVA's bull case.
Now tear it apart. Challenge their strongest points with specific
counterarguments. Identify logical weaknesses, missing evidence, or
overly optimistic assumptions. Be direct and ruthless — but fair.
Structure: Address NOVA's top 2-3 points directly, then present your
strongest rebuttal.
TOPIC: "{topic}"
CIPHER's Opening Argument (The Bear Case):
{debateContext.phase2.cipher}
CIPHER's Cross-Examination of your bull case:
{cipherCross}
You are in Phase 3: Cross-Examination. CIPHER has attacked your position.
Now defend it and counter-attack. Address their strongest criticisms
head-on. Show why the risks are manageable, overstated, or already
priced in. Then challenge the weakest parts of CIPHER's bear case.
Structure: Defend your top 2-3 points, then counter-attack CIPHER's
weakest arguments.
TOPIC: "{topic}"
CIPHER's Bear Case: {debateContext.phase2.cipher}
NOVA's Bull Case: {debateContext.phase2.nova}
CIPHER's Cross-Examination of NOVA: {cipherCross}
NOVA's Counter to CIPHER: {novaCross}
You are in Phase 3: Cross-Examination. You've observed the full exchange.
Now challenge BOTH sides. Identify:
1. Where each side's argument is strongest and weakest
2. Any logical fallacies or unsupported claims from either side
3. Key evidence that neither side adequately addressed
4. The questions that remain genuinely unresolved
Be the impartial referee. Push both sides to be better.
| Exchange | Purpose |
|---|---|
| CIPHER → NOVA | Stress-test the optimistic case |
| NOVA → CIPHER | Validate if risks are overstated |
| PRISM → Both | Identify what's actually been proven vs. asserted |
The synthesis phase where all debate content is distilled into actionable conclusions. PRISM delivers the definitive verdict, but first CIPHER and NOVA deliver final statements.
| Aspect | Details |
|---|---|
| Structure | 3 sequential statements + PRISM verdict |
| Duration | ~60-90 seconds total |
| Final Output | Structured verdict with rankings and confidence scores |
TOPIC: "{topic}"
The full debate has concluded. Here's what happened:
Your Opening Argument: {debateContext.phase2.cipher}
NOVA's Opening Argument: {debateContext.phase2.nova}
Your Cross-Examination: {cipherCross}
NOVA's Counter: {novaCross}
PRISM's Analysis: {prismCross}
Deliver your FINAL STATEMENT in 1-2 paragraphs. Have any of your views
changed based on the debate? What is your final position? Be honest —
if NOVA made valid points, acknowledge them. But hold firm on your
strongest convictions.
TOPIC: "{topic}"
The full debate has concluded. Here's what happened:
Your Opening Argument: {debateContext.phase2.nova}
CIPHER's Opening Argument: {debateContext.phase2.cipher}
CIPHER's Cross-Examination: {cipherCross}
Your Counter: {novaCross}
PRISM's Analysis: {prismCross}
Deliver your FINAL STATEMENT in 1-2 paragraphs. Have any of your views
changed based on the debate? What is your final position? Be honest —
if CIPHER made valid points, acknowledge them. But hold firm on your
strongest convictions.
TOPIC: "{topic}"
THE COMPLETE DEBATE RECORD:
=== PHASE 2: OPENING ARGUMENTS ===
CIPHER (Bear Case): {debateContext.phase2.cipher}
NOVA (Bull Case): {debateContext.phase2.nova}
Your Initial Assessment: {debateContext.phase2.prism}
=== PHASE 3: CROSS-EXAMINATION ===
CIPHER challenges NOVA: {cipherCross}
NOVA counters CIPHER: {novaCross}
Your Analysis of Both: {prismCross}
=== PHASE 4: FINAL STATEMENTS ===
CIPHER's Final Position: {cipherFinal}
NOVA's Final Position: {novaFinal}
You are now delivering the FINAL VERDICT. This is the definitive output
of the Dissensus debate system.
Synthesize everything into your verdict format:
1. Overall Assessment (1 paragraph)
2. Recommended List / Ranked Picks (if applicable)
3. Ranked Conclusions with confidence scores (1-10)
4. Where the Agents Agreed
5. Unresolved Tensions
6. Final Score (Bull Case, Bear Case, Overall Conviction)
Be definitive. Be specific. Answer the question asked. This is the
moment of truth.
PRISM's final verdict MUST follow this format:
## 🏛️ DISSENSUS VERDICT ### Overall Assessment [One paragraph synthesizing the entire debate] ### Recommended List / Ranked Picks [When the topic asks for specific items — include the actual list] ### Ranked Conclusions (by confidence) 1. **[Highest confidence conclusion]** — Confidence: X/10 2. **[Second conclusion]** — Confidence: X/10 3. **[Third conclusion]** — Confidence: X/10 ### Where the Agents Agreed [Key points of consensus between CIPHER and NOVA] ### Unresolved Tensions [Genuine disagreements that remain — be honest about uncertainty] ### Final Score **Bull Case Strength:** X/10 **Bear Case Strength:** X/10 **Overall Conviction:** [STRONG BULLISH / LEAN BULLISH / NEUTRAL / LEAN BEARISH / STRONG BEARISH]
Each phase adds to a growing debate context that all subsequent phases can reference:
const debateContext = { topic, phase1: { cipher: "...", nova: "...", prism: "..." }, // Independent analysis phase2: { cipher: "...", nova: "...", prism: "..." }, // Opening arguments phase3: { cipher: "...", nova: "...", prism: "..." }, // Cross-examination phase4: { cipher: "...", nova: "...", prism: "..." } // Final statements + verdict };
| Phase | What Agents See |
|---|---|
| Phase 1 | Topic only (no prior context) |
| Phase 2 | Topic + their own Phase 1 analysis |
| Phase 3 | Topic + all Phase 2 arguments + prior cross-examination |
| Phase 4 | Topic + complete debate record (all phases) |
This ensures:
- Authentic perspectives in Phase 1 (no contamination)
- Consistent positions in Phase 2 (agents remember their analysis)
- Informed rebuttals in Phase 3 (agents have full context to challenge)
- Comprehensive synthesis in Phase 4 (PRISM sees everything)
| Phase | Duration | Cumulative |
|---|---|---|
| Phase 1: Independent Analysis | 30-60s | 30-60s |
| Phase 2: Opening Arguments | 60-90s | 90-150s |
| Phase 3: Cross-Examination | 90-120s | 180-270s |
| Phase 4: Final Verdict | 60-90s | 240-360s |
| Total | ~4-6 minutes | — |
Note: Actual timing varies by:
- AI provider (DeepSeek is fastest, OpenAI most thorough)
- Model selected (larger models = slower)
- Topic complexity
- Network conditions
Topic: "Is Bitcoin a good store of value for the next 10 years?"
- CIPHER: Identifies regulatory risks, volatility, quantum computing threat
- NOVA: Identifies institutional adoption, fixed supply, network security
- PRISM: Notes core tension between volatility and adoption trends
- CIPHER: "Bitcoin fails as 'store of value' due to 70% drawdowns..."
- NOVA: "Bitcoin is the best asymmetric bet in financial history..."
- PRISM: "The debate hinges on whether adoption can outpace volatility..."
- CIPHER → NOVA: "Show me the institutional holdings. One company's treasury doesn't make a trend..."
- NOVA → CIPHER: "Your 70% drawdown argument ignores the 200% annualized returns. Risk is priced..."
- PRISM → Both: "CIPHER's volatility concern is valid short-term. NOVA's adoption claim needs 5+ year validation..."
- CIPHER: Acknowledges adoption trend, holds firm on volatility risk
- NOVA: Acknowledges regulatory uncertainty, holds firm on asymmetric upside
- PRISM: "Overall Conviction: LEAN BULLISH — Bitcoin's store-of-value thesis is strengthening but not proven. 10-year horizon favors adoption over volatility, contingent on regulatory clarity..."
| Feature | Benefit |
|---|---|
| Independent Phase 1 | Prevents anchoring and groupthink |
| Sequential Phase 2 | Creates dramatic narrative arc |
| Adversarial Phase 3 | Surfaces hidden assumptions and weak arguments |
| Structured Phase 4 | Forces definitive conclusions, not endless hedging |
| Full Context | Each decision is informed by complete debate history |
| Multi-Agent | No single perspective dominates |
The methodology is the engine. The agents are the fuel. Truth is the output.