/evolve

description: Run the Evolution Engine to discover and validate trading strategies
argument-hint: "[symbol] [timeframe] [generations]"

Evolve Strategy

Trigger the Evolution Engine to autonomously discover trading patterns from raw OHLCV data. The engine generates candidate strategies via LLM, backtests them vectorized, validates out-of-sample, and graduates survivors.

Workflow

Step 1: Configure Evolution

If parameters are provided, use them. Otherwise ask:

• **Symbol**: e.g., BTCUSDT, ETHUSDT (Binance pairs)
• **Timeframe**: 1h, 4h, 1d
• **Generations**: How many evolution cycles (default: 3)
• **Candidates per generation**: How many strategies to test (default: 10)
• **Data period**: How many days of historical data (default: 90)

Step 2: Fetch Market Data

Use the `evolution_fetch_market_data` MCP tool:

evolution_fetch_market_data({
  symbol: "BTCUSDT",
  timeframe: "1h",
  days: 90
})

Step 3: Discover Patterns

Use the `evolution_discover_patterns` MCP tool:

evolution_discover_patterns({
  symbol: "BTCUSDT",
  timeframe: "1h",
  num_patterns: 10
})

The LLM analyzes price data and generates candidate trading rules (entry/exit conditions, position sizing, stop loss).

Step 4: Run Evolution Loop

Use the `evolution_evolve_strategy` MCP tool:

evolution_evolve_strategy({
  symbol: "BTCUSDT",
  timeframe: "1h",
  generations: 3,
  candidates_per_gen: 10
})

Each generation: 1. **Generate** — LLM creates N candidate strategies 2. **Backtest** — Vectorized backtesting with Sharpe, win rate, max drawdown 3. **Select** — Top performers survive, bottom eliminated 4. **Mutate** — LLM evolves survivors with variations 5. **Validate** — Out-of-sample test on held-out data

Step 5: Report Results

For each graduated strategy:

| Metric | In-Sample | Out-of-Sample | |--------|-----------|---------------| | Sharpe Ratio | X.XX | X.XX | | Win Rate | X% | X% | | Max Drawdown | X% | X% | | Total Return | X% | X% | | # Trades | N | N |

Plus:

• Strategy description (entry/exit rules in plain language)
• Graveyard summary (why eliminated strategies failed)
• Confidence assessment (how robust is the OOS performance?)

Important Notes

• Requires `ANTHROPIC_API_KEY` for LLM-powered pattern discovery
• Uses Binance public API for OHLCV data (no API key needed)
• Each generation takes 30-60 seconds depending on candidate count
• Always validate OOS before live trading — in-sample results are not reliable alone

Example

User: /evolve BTCUSDT 1h 3