PAL Consensus - Multi-Model Debate

Build consensus through systematic analysis and structured debate across multiple AI models.

When to Use

• Complex architectural decisions
• Technology selection
• Feature design trade-offs
• Risk assessment
• When you need multiple perspectives
• Validating important decisions

Quick Start

# Step 1: State the proposal and do your analysis
result = mcp__pal__consensus(
    step="Evaluate: Should we migrate from monolith to microservices?",
    step_number=1,
    total_steps=4,  # Your analysis + 2 models + synthesis
    next_step_required=True,
    findings="My initial analysis: Consider scale, team size, complexity...",
    models=[
        {"model": "openai/gpt-5", "stance": "for"},
        {"model": "deepseek/deepseek-v3.2", "stance": "against"}
    ]
)

# Step 2+: Process each model's response
result = mcp__pal__consensus(
    step="Recording pro-microservices perspective",
    step_number=2,
    total_steps=4,
    next_step_required=True,
    findings="GPT-5 argues: scalability benefits, team autonomy...",
    continuation_id=result["continuation_id"]
)

Stance Types

Required Parameters

Optional Parameters

Model Configuration

models=[
    {
        "model": "openai/gpt-5",
        "stance": "for",
        "stance_prompt": "Focus on scalability and maintainability benefits"
    },
    {
        "model": "deepseek/deepseek-v3.2",
        "stance": "against",
        "stance_prompt": "Consider operational complexity and team capacity"
    },
    {
        "model": "google/gemini-3-flash-preview",
        "stance": "neutral"
    }
]

Available Models

Top models for consensus (all score 100):

• openai/gpt-5 - Strong reasoning
• deepseek/deepseek-v3.2 - Thinking-enabled
• google/gemini-3-flash-preview - 1M context
• x-ai/grok-4.1 - 2M context
• bytedance-seed/seed-1.6 - Thinking-enabled

Workflow Pattern

Step 1: State proposal + your independent analysis
        ↓
Step 2: First model responds (for/against/neutral)
        ↓
Step 3: Second model responds
        ↓
Step N: Synthesize all perspectives

Example: Technology Decision

# Debate: GraphQL vs REST
mcp__pal__consensus(
    step="Evaluate: Should we use GraphQL instead of REST for our new API?",
    step_number=1,
    total_steps=5,
    next_step_required=True,
    findings="""
    Initial analysis:
    - Current team: 5 backend devs, familiar with REST
    - Use case: Mobile app with varying data needs
    - Timeline: 3 months to launch
    - Consider: Learning curve, tooling, performance
    """,
    models=[
        {"model": "openai/gpt-5", "stance": "for"},
        {"model": "deepseek/deepseek-v3.2", "stance": "against"},
        {"model": "google/gemini-3-flash-preview", "stance": "neutral"}
    ],
    relevant_files=[
        "/docs/api-requirements.md",
        "/app/api/current_endpoints.py"
    ]
)

Best Practices

1. Frame proposals clearly - Specific, evaluable statements
2. Provide context - Constraints, requirements, history
3. Use diverse models - Different strengths and perspectives
4. Balance stances - Include for, against, and neutral
5. Document synthesis - Capture key insights from all perspectives