Consensus Mechanisms
The final verdict of an Aegis-Graph audit is the result of a Multi-Agent Consensus Handshake. This chapter details the mathematical and logical process by which the MARS swarm resolves conflicts and issues a sovereign proof.
⚖️ The Weighted Voting Model
Each agent in the MARS swarm (Vision, Graph, Logic) contributes a Confidence Vector (v) and an Evidence Weight (w) to the final decision.
| Agent | Core Metric | Base Weight (W) |
|---|---|---|
| Vision (VF) | Artifact Fidelity | 0.30 |
| Graph (GN) | Institutional Standing | 0.35 |
| Logic (LA) | Temporal Consistency | 0.35 |
🤝 The Handshake Protocol
- Agent Discovery: When an audit is initialized, the local node spins up a temporary MARS swarm.
- Independent Audit: Agents perform their specialized checks in parallel, generating internal evidence logs.
- Conflict Detection: If the Vision agent flags a "High Risk" but the Graph agent finds a "High Authority" institution, a Consensus Conflict is triggered.
- CoT Resolution: The Logic Auditor initiates a Chain-of-Thought reasoning path, querying both agents for their raw evidence. It then assigns a higher weight to the layer with the most robust primary-source alignment.
🔒 Settlement & Finality
Once the consensus threshold (T > 0.90) is met, the system generates a Sovereign Audit Proof.
- Finality: Once a proof is signed by the MARS quorum, it is anchored to the institutional node's ledger.
- Immutability: The reasoning trail (minus PII) is preserved to allow for future audits or appeals if new data enters the global graph.
Return to Documentation Home