_INTELLITRADE

Consensus-driven BTC trading agent. Runs 31 parallel prediction paths on 5-minute candles. Trades fire only when 28/31 models agree. Below 26 votes the trade dies instantly. Position sizing via Kelly criterion.

Voice Notification

When executing a workflow, do BOTH:

1. Send voice notification:

   curl -s -X POST http://localhost:31337/notify \
     -H "Content-Type: application/json" \
     -d '{"message": "Running WORKFLOWNAME in _INTELLITRADE"}' \
     > /dev/null 2>&1 &

2. Output text notification:

   Running **WorkflowName** in **_INTELLITRADE**...

Full documentation: ~/.claude/PAI/DOCUMENTATION/Notifications/NotificationSystem.md

Workflow Routing

System Architecture

┌─────────────────────────────────────────────────────────┐
│                   Intellitrade Orchestrator               │
├─────────────────────────────────────────────────────────┤
│   Data Layer  │  Model Layer  │  Consensus  │ Execution │
├───────────────┼──────────────┼─────────────┼───────────┤
│ Binance/CCXT  │ Model 1..31  │ Vote Engine │ Order Mgr │
│ 5-min candles │              │ 28/31 thresh│ Kelly     │
│ OHLCV streams │ parallel     │ 26 kill     │ Slippage  │
└───────────────┴──────────────┴─────────────┴───────────┘

Core Principle

Not prediction. Consensus. Each model independently predicts direction (long/short/flat). Only when 28+ of 31 converge does a signal fire. Below 26 votes the signal dies instantly. Most days the system stays flat.

Key Parameters

Examples

Example 1: Run the live trader

User: "Start Intellitrade"
→ Invokes Run workflow
→ Connects to exchange, begins 5-min candle loop
→ Runs 31 models in parallel on each candle
→ Outputs: signals, P&L, current positions

Example 2: Check trading status

User: "How is Intellitrade doing?"
→ Invokes Monitor workflow
→ Shows: current signal, open positions, daily P&L, model vote breakdown

Example 3: Backtest on historical data

User: "Backtest Intellitrade on last 30 days of BTC data"
→ Invokes Backtest workflow
→ Downloads 30 days of 5-min candles
→ Runs full simulation
→ Outputs: equity curve, win rate, max drawdown, Sharpe ratio

Gotchas

• MiroFish is a social simulation platform, not a crypto trading simulator. This system implements the parallel-prediction pattern independently.
• The 31 models use different parameterizations of the same underlying indicators — diversity comes from parameter spread, not fundamentally different model architectures.
• API keys are stored in ~/.intellitrade/config.json — never hardcode them.
• Exchange rate limits must be respected. The data fetcher uses adaptive throttling.
• Kelly criterion can suggest very large positions during high-conviction periods — use a fractional Kelly (0.25-0.5) in production.
• The system has no opinion on most candles — expect 80%+ flat time.
• Backtest results are NOT guaranteed to predict live performance. Market regimes shift.

INTELLITRADE — Drtlemon Elite Tech Conglomerate
Engineering the asymmetric edge. Systems over outcomes. Mathematics over emotion.
Proprietary. Unauthorized distribution is a violation of intellectual property.