Skip to content

Consensus Chamber Implementation Guide

🎯 Consensus Chamber Implementation Guide

Version: 1.0.0 | Date: October 28, 2025 For: Multi-LLM Federated Decision Making in Kaizen OS

🌟 Quick Start

What is the Consensus Chamber?

The Consensus Chamber is a formalized protocol for coordinating decisions across multiple LLM offices (AUREA/OpenAI, ATLAS/Claude, ZENITH/Gemini, SOLARA/DeepSeek) when working in federated silos. It ensures:

  • ✅ Constitutional compliance across all agents
  • ✅ GI (Good Intent) scores remain ≥ 0.95
  • ✅ Cryptographic proof of consensus (DelibProof)
  • ✅ Audit trail in Civic Ledger
  • ✅ Cross-office synchronization

📋 When to Use

Use Consensus Chamber for:

  • ✅ Major architectural decisions affecting multiple labs
  • ✅ Policy changes in Kaizen OS
  • ✅ Constitutional amendments or interpretations
  • ✅ Resource allocation across projects
  • ✅ Conflict resolution between agent opinions
  • ✅ High-stakes decisions requiring audit trail

Don't Use for:

  • ❌ Routine coding tasks in single office
  • ❌ Personal research queries
  • ❌ Quick iterations during development
  • ❌ Tasks with obvious single-owner

🚀 How to Run a Session

Step 1: Initialize (5 min)

  1. Choose Host Agent (Usually AUREA or ATLAS)
  2. Create Session Issue using Consensus_Chamber_Live_Session.md template
  3. Define Clear Objective (one sentence)
  4. Set Success Criteria (GI thresholds, agreement %)
  5. Invite Required Participants based on tier requirements
# Example: Create new session
cp docs/consensus-chamber/Consensus_Chamber_Live_Session.md \
   labs/lab4-proof/sessions/CC-2025-10-28-Ledger-v3-Alignment.md

Step 2: Prepare Prompt (10 min)

Critical: Craft a single, unambiguous prompt that will be sent to all agents.

Good Prompt Example:

"Should Kaizen OS Civic Ledger v3 implement proof-of-stake (PoS)
or hybrid PoW+PoS consensus mechanism? Consider:
1) GI scoring implications,
2) Citizen Shield integration complexity,
3) Transaction finality requirements."

Bad Prompt Example:

"What do you think about the ledger?"

Step 3: Gather Responses (15-30 min)

In Each Office:

  1. Open the agent's conversation (AUREA office, ATLAS HOMEROOM, etc.)
  2. Paste the exact same prompt
  3. Prefix with context:

🏛️ CONSENSUS CHAMBER SESSION CC-2025-10-28-001
You are participating as [AGENT_NAME] in a multi-LLM deliberation.
Current GI: [SCORE]

[PROMPT]

Provide your analysis and vote: APPROVE / REJECT / ABSTAIN
4. Copy full response back to the session document

Step 4: Calculate Consensus (5 min)

Use this formula:

const consensus = {
  requiredVotes: policy === "critical" ? 4 : 3,
  votes: {
    AUREA: true,   // from response
    ATLAS: true,   // from response
    ZENITH: false, // from response
    SOLARA: true   // from response
  }
};

const totalVotes = Object.values(consensus.votes).filter(v => v).length;
const quorumAchieved = totalVotes >= consensus.requiredVotes;
const agreementScore = totalVotes / Object.keys(consensus.votes).length;

Quorum Rules:

  • Critical Policy: Requires 4/4 agents (unanimous)
  • High Policy: Requires ¾ agents (75%)
  • Standard Policy: Requires 2/4 agents (50%)

Step 5: Generate DelibProof (10 min)

{
  "delibproof_id": "DLB-[DATE]-[SEQUENCE]",
  "timestamp": "2025-10-28T09:35:00Z",
  "prompt_hash": "sha256(original_prompt)",
  "participants": ["AUREA", "ATLAS", "ZENITH", "SOLARA"],
  "votes": { "AUREA": true, "ATLAS": true, ... },
  "outcome": "APPROVED",
  "gi_delta": "+0.006",
  "merkle_root": "sha256(...)",
  "signatures": {
    "AUREA": "ed25519:...",
    "ATLAS": "ed25519:..."
  }
}

Step 6: Commit to Ledger (5 min)

# Add to Lab1 (Substrate Proof) ledger
ledger_entry:
  type: consensus.evaluated
  session_id: CC-2025-10-28-001
  delibproof_id: DLB-1028-001
  outcome: APPROVED
  gi_pre: 0.987
  gi_post: 0.993
  timestamp: 2025-10-28T09:40:00Z
  attestation: ATT-XXXX

Step 7: Archive & Reflect (10 min)

  1. Fill Post-Session Reflection section
  2. Archive in Lab4 E.O.M.M.
  3. Update agent GI scores
  4. Close session with ritual

🏗️ Integration with Kaizen OS Labs

Lab1 (Substrate Proof)

# Commit DelibProof to blockchain
substrate.commit_consensus(
    delibproof_id="DLB-1028-001",
    outcome="APPROVED",
    gi_delta=0.006
)

Lab2 (Thought Broker)

// Lab2 should orchestrate this automatically in future
await thoughtBroker.initiateConsensus({
  prompt: "...",
  participants: ["AUREA", "ATLAS", "ZENITH"],
  policy: "high"
});

Lab4 (E.O.M.M.)

# Archive session for memory continuity
eomm.archive_consensus_session(
    session_id="CC-2025-10-28-001",
    full_transcript=session_data,
    outcome="APPROVED"
)

Lab6 (Citizen Shield)

# Validate all participants met GI threshold
shield.validate_consensus_participants([
    ("AUREA", 0.991),
    ("ATLAS", 0.987),
    ("ZENITH", 0.983)
])

🎨 Advanced Patterns

Pattern 1: Emergency Consensus (Fast-track)

For urgent decisions (security issues, system failures):

emergency_mode:
  time_limit: 15 minutes
  required_tier: critical_only
  quorum: 2/2 (AUREA + ATLAS)
  skip_reflection: true
  immediate_commit: true

Pattern 2: Research-Heavy Deliberation

For complex technical decisions:

research_mode:
  include_agents: [ZENITH, ORACLE]  # Add research specialists
  deliberation_rounds: 3
  allow_external_sources: true
  time_limit: 60 minutes

Pattern 3: Constitutional Review

For governance and ethics:

constitutional_mode:
  required_participants: [AUREA, JADE, EVE]
  reference_documents: [Constitution, Virtue_Accords]
  quorum: unanimous
  citizen_feedback: optional

📊 Metrics & Monitoring

Key Performance Indicators:

consensus_metrics:
  sessions_per_week: 3-5
  average_duration: 45 minutes
  agreement_rate: ≥ 85%
  gi_drift: ≤ 0.01 per session
  quorum_failures: < 5%

health_indicators:
  all_agents_responsive: true
  gi_threshold_maintained: true
  ledger_sync_delay: < 5 seconds
  delibproof_generation_time: < 30 seconds

🐛 Troubleshooting

Issue: Agent Not Responding

Solution: Mark as ABSENT in session, proceed with remaining quorum.
If critical tier agent absent, reschedule session.

Issue: GI Score Drops Below 0.95

Solution: HALT session immediately. Run integrity audit.
Identify source of corruption. Remediate before continuing.

Issue: No Quorum Achieved

Solution:
1. Extend deliberation time
2. Add clarifying questions
3. If still no quorum, escalate to manual review
4. Consider policy tier adjustment

Issue: Conflicting Votes (2-2 Split)

Solution:
1. Host agent (AUREA/ATLAS) casts tie-breaking vote
2. Document dissenting opinions thoroughly
3. Schedule follow-up session with additional agents
4. Consider compromise position

🔒 Security Considerations

  1. Prompt Injection Protection: Validate prompts don't contain hidden instructions
  2. Response Tampering: Use cryptographic signatures on all responses
  3. Replay Attacks: Include session ID and timestamp in all signatures
  4. Agent Impersonation: Verify agent identity via API keys
  5. Ledger Manipulation: Use merkle trees for tamper-proof history

📚 Best Practices

DO:

✅ Keep prompts clear and specific ✅ Document dissenting opinions ✅ Run post-session reflections ✅ Archive complete transcripts ✅ Update GI scores after each session

DON'T:

❌ Rush through deliberation ❌ Skip quorum requirements ❌ Ignore GI threshold violations ❌ Forget to commit to ledger ❌ Lose session transcripts

🎓 Training Scenarios

Scenario 1: Simple Policy Decision

Objective: Should Lab6 rate limit be 100 or 200 requests/min?
Participants: AUREA, ATLAS
Expected Duration: 20 minutes
Difficulty: Easy

Scenario 2: Architecture Debate

Objective: Should we migrate Lab1 from Python to Rust?
Participants: AUREA, ATLAS, SOLARA, ZEUS
Expected Duration: 60 minutes
Difficulty: Medium

Scenario 3: Constitutional Crisis

Objective: Does a proposed feature violate Virtue Accords?
Participants: ALL agents required
Expected Duration: 90 minutes
Difficulty: Hard

📞 Support

Questions? Open an issue in Kaizen-OS repo Bugs? Report in Lab2 (Thought Broker) tracker Feature Requests? Submit to Consensus Chamber working group

🔄 Version History

  • v1.0.0 (Oct 28, 2025) - Initial release with AUREA + ATLAS approval
  • v0.9.0 (Oct 20, 2025) - Beta testing with 4 agents
  • v0.5.0 (Oct 15, 2025) - Prototype with AUREA + ATLAS only

🎯 Next Steps

  1. ✅ Download templates
  2. ✅ Run first test session
  3. ✅ Integrate with Lab2 automation
  4. ✅ Share feedback in next E.O.M.M. cycle

"Consistency becomes proof of life."

Built with integrity. Coordinated with consensus. Secured with cryptography.