Dual-LLM Consensus Engine

At launch, ConsortiumAI operates with a Dual LLM Adversarial System, expandable to N-agent consensus.

The Two Agents

MODEL_ALPHA — Market Analyst

The opportunity-seeking agent focused on identifying profitable strategies.

Attribute	Value
Role	Opportunity Discovery
Risk Profile	Aggressive
Focus	Maximizing returns

Responsibilities:

• Analyzes macro trends
• Evaluates sentiment and momentum
• Studies price action and volume
• Proposes high-conviction strategies

MODEL_BETA — Risk Auditor

The protective agent focused on capital preservation.

Attribute	Value
Role	Capital Protection
Risk Profile	Conservative
Focus	Minimizing risk

Responsibilities:

• Simulates slippage and liquidity stress
• Analyzes volatility and drawdown risk
• Evaluates sandwich and MEV exposure
• Modifies or rejects Alpha proposals

Adversarial Design

The two agents are intentionally designed with opposing risk profiles:

            MODEL_ALPHA              MODEL_BETA
            ───────────              ──────────
Risk:       Aggressive               Conservative
Goal:       Find opportunities       Protect capital
Bias:       Toward action            Toward caution
Role:       Propose                  Challenge

This adversarial relationship ensures:

• No single-model bias — Alpha's optimism is checked by Beta's caution
• Robust decisions — Only opportunities that survive scrutiny proceed
• Built-in skepticism — Every proposal faces rigorous challenge

Verdict System

Each model submits an independent verdict:

Verdict	Alpha Meaning	Beta Meaning
Execute	"High conviction opportunity"	"Risk acceptable"
Modify	"Opportunity with adjustments"	"Reduce exposure"
Reject	"No opportunity found"	"Risk too high"
Wait	"Need more data"	"Conditions unclear"

Consensus Requirements

For execution to proceed:

1. Both agents must submit verdicts
2. Verdicts must align (both Execute, or Execute + Modify)
3. Consensus threshold (≥95%) must be met
4. All verdicts are cryptographically signed

Future: N-Agent Expansion

The dual-agent system is designed to expand:

• Additional specialized agents (MEV detection, volatility analysis, macro trends)
• Weighted voting based on agent expertise
• Dynamic consensus thresholds
• Agent marketplaces

Roadmap

See Roadmap for the multi-agent expansion timeline.