First Backtest

With the dev stack running and market data seeded, you can run a backtest and inspect results in gordon-console.

The backtest = live invariant

Gordon has one strategy evaluation code path. The same Strategy::evaluate function runs inside gordon-bot's live loop and inside gordon-manager's backtest replay engine (BacktestExecution). There is no shadow engine, no Python reimplementation, no separate simulation model. A CI byte-parity gate (tests/it_backtest_live_byte_parity.rs) enforces this at every push.

This means backtest results are structurally trustworthy: what you see in the backtest is what the bot would have done.

Run a backtest via the console

Open http://localhost:3000 and navigate to Research in the sidebar. Select a strategy, symbol, and timeframe, then click Run Backtest.

Results appear in the runs list once the replay completes. Click a run to see:

• Performance stats (Sharpe, max drawdown, CAGR, win rate)
• Equity curve with drawdown overlay
• Roundtrip table with individual trade PnL, duration, and fees

Run a backtest via the BFF API

The manager BFF exposes a REST endpoint directly:

curl -X POST http://localhost:8083/backtest \
  -H "Authorization: Bearer dev-operator-token" \
  -H "Content-Type: application/json" \
  -d '{
    "strategy": "supertrend",
    "symbol": "BTCUSDT",
    "timeframe": "1d",
    "from_ts": 1704067200000,
    "to_ts": 1735689600000,
    "params": {
      "atr_period": 10,
      "multiplier": 3.0
    }
  }'

The response body contains a run_id. Query the run directly:

curl http://localhost:8083/runs/{run_id} \
  -H "Authorization: Bearer dev-operator-token"

Available strategies

Strategy	strategy value	Validated timeframes
Supertrend 3.0	supertrend	1d, 8h
EWMAC ensemble	ewmac	6h, 4h, 2h, 1h
PSAR tight	psar	1w, 1d

The 11 strategies added in 4.6.x (basis_revert, bollinger_breakout, funding_arb, hurst_regime_switcher, opening_range_breakout, pairs_mr, rsi_mean_reversion, tsmom, volume_spike_breakout, vpin_impulse, vwap_stretch_mr) are registered in the StrategyRegistry and available for backtesting, but their walk-forward validation is still in progress.

Interpreting results

• Sharpe ratio is annualized. Above 0.5 is acceptable for trend-following; above 0.9 is the current production threshold.
• Max drawdown shows the worst peak-to-trough decline. Trend-following systems routinely see 20-40% drawdowns — this is expected.
• Fees are included. The backtest applies Binance VIP 0 taker fees (4 bps) and estimated slippage per asset. Results are net of costs.
• Walk-forward validation separates real edge from curve-fitting. A single full-history backtest is not sufficient — see Walk-Forward Testing for the proper methodology.

Database tables written

Each backtest run writes to:

Table	Content
runs	One row per run — strategy, symbol, timeframe, parameters, timestamps
trades	Individual fills (entry and exit)
roundtrips	Matched entry-exit pairs with PnL, duration, fees
equity_points	Account balance over time
performance_snapshots	Aggregate metrics

Query directly if needed:

make dev-psql

SELECT id, strategy, symbol, timeframe, sharpe, max_drawdown_pct
FROM runs
ORDER BY created_at DESC
LIMIT 10;

Next step

Now that you have results, read Project Structure to understand how the v7 workspace is organized.