Consensus Formation

The heart of ConsortiumAI is its consensus mechanism — where multiple AI agents must agree before any capital moves.

The Consensus Process

Step 1: Independent Analysis

Each agent analyzes market data independently:

• No agent sees another's conclusions
• Each produces its own assessment
• Analysis covers opportunity AND risk

Step 2: Verdict Submission

Each model submits an independent verdict:

Verdict	Meaning
Execute	Proceed with the strategy
Modify	Adjust parameters, then proceed
Reject	Do not execute
Wait	Delay for more information

Step 3: Consensus Check

MODEL_ALPHA Verdict   MODEL_BETA Verdict   Result
───────────────────   ──────────────────   ──────────
Execute               Execute              ✅ Execute
Execute               Modify               ⚠️ Modify & Execute
Execute               Reject               ❌ No Action
Execute               Wait                 ⏳ Wait
Reject                Any                  ❌ No Action

Step 4: Cryptographic Signing

• All proposals are cryptographically signed
• Verdicts are immutably logged
• Decision trail is fully auditable

Consensus Threshold

Current Threshold

Consensus threshold: ≥95%

For the dual-agent system, this means both agents must substantially agree for execution to proceed.

Disagreement as Protection

When agents disagree:

• No action is taken
• Capital is protected
• The system waits for clearer signals

Design Philosophy

Disagreement is treated as a protective feature, not a failure. In uncertain conditions, doing nothing is often the right choice.

Example Scenario

Market Signal: Potential arbitrage opportunity

MODEL_ALPHA (Analyst):
  "High-conviction opportunity. 2.3% spread detected.
   Liquidity appears sufficient. → EXECUTE"

MODEL_BETA (Risk Auditor):
  "Slippage simulation shows 1.8% cost.
   Net profit only 0.5% — below threshold. → REJECT"

Consensus: NOT REACHED
Action: None — Capital Protected

This is ConsortiumAI's strength: capital that reasons before it moves.