Documentation

AI Debate System

Understanding how our multi-agent AI consensus system generates trading signals

What is the Debate System?

The Debate System is a sophisticated multi-agent AI framework that uses 10 specialized AI agents to analyze cryptocurrency markets and reach trading consensus through structured debate rounds.

Why Debate Instead of a Single AI Model?

Single Model Problems

  • • Blind spots: One model may miss patterns another would catch
  • • Overconfidence: No challenge to flawed reasoning
  • • Single point of failure: If model hallucinates, no backup
  • • Limited perspective: Only one analytical lens

Multi-Agent Debate Benefits

  • • Cross-validation: Agents verify each other's analysis
  • • Error detection: Challenge round exposes weak reasoning
  • • Diverse expertise: Technical, sentiment, macro, risk all covered
  • • Calibrated confidence: Consensus strength reflects true certainty

10 AI Agents

Specialized Analysts

3 Debate Rounds

Initial → Challenge → Final

1 Decision

BUY / SELL / HOLD

Core Concepts

1

Specialization

Each agent is an expert in one domain (Technical Analysis, Elliott Wave, Smart Money, etc.). They don't try to know everything - they go deep in their specialty.

2

Structured Debate

Agents don't just vote - they engage in 3 rounds where they can see, challenge, and respond to each other's analyses. This mimics how a real trading desk works.

3

Weighted Consensus

Not all agents are equal. Head Analyst assigns weights based on historical accuracy and current market conditions. A 80% accurate agent has more influence than a 40% accurate one.

4

Risk-First Architecture

Risk Manager has veto power. Even 100% bullish consensus can be blocked if risk checks fail. Capital preservation is the top priority.

Multiple LLMs

GPT-4o, Claude, Gemini, Qwen, DeepSeek, and more

Consensus Based

Weighted voting from all specialists

Real-Time Data

Live market data from Binance

What Happens in a Debate Session?

  1. 1Data Injection: System fetches live price, OHLCV, indicators (RSI, MACD, MAs), and pre-calculated patterns (Elliott Wave, Harmonic, SMC zones)
  2. 2Round 1 (INITIAL): Each agent analyzes independently without seeing others. Outputs: direction, confidence, price levels
  3. 3Round 2 (CHALLENGE): Agents see each other's R1 analyses. They critique, question, and can adjust their position if convinced
  4. 4Round 3 (FINAL): After challenges, each agent provides final recommendation with refined confidence and specific trade levels
  5. 5Head Analyst: Synthesizes all 8 specialist opinions, assigns weights based on accuracy, calculates weighted consensus
  6. 6Strategy Director: Makes final BUY/SELL/HOLD decision with exact entry, stop loss, and take profit levels

Total LLM Calls per Debate: 8 agents × 3 rounds + 1 Head Analyst + 1 Strategy Director = 26 LLM calls