Strategy Backtest & Parameter Sweep
Published Aug 12, 2026
The strategy backtest workbench validates parameters before you create a classic strategy: pick a built-in template, configure symbols and parameters, and run a historical backtest locally in your browser. It can also sweep a batch of parameter combinations for you and use risk diagnostics to find ranges that are both profitable and robust. The entry is Strategy development → Strategy backtest in the sidebar; sign-in is required.
Division of labor with the AI workspace: the workspace backtests chat-generated strategy code; this page backtests classic template parameters and can hand the winning parameters straight into the creation flow.
Run a backtest
Pick a strategy and symbols
Choose a built-in template, then a market data source and trading pairs. Skip templates marked “backtest not supported yet”.
Configure parameters
Three groups: execution & exposure (leverage sets both target notional exposure and order leverage), strategy parameters, and risk controls (stop loss, take profit, trailing exits—all feed the backtest and join the sweep list).
The exposure preview shows target total notional, gross leverage, and estimated initial margin as a share of starting funds in real time.
Run and read the results
The backtest runs locally in your browser. When it finishes, read returns, drawdown, trades, and positions—the same way as in Run Your First Backtest.
Create the strategy with these parameters
Click Create strategy with these parameters to carry them into the New Strategy flow—no copying by hand.
Before running, the workbench checks market data integrity (missing markets, candle gaps, and similar). In strict mode, data issues stop the run; you can switch data sources, adjust the range, or drop the affected pairs from the config.
Parameter sweep and risk diagnostics
Trying parameters one by one is slow. The sweep scans them for you:
Choose parameters to scan
Pick from the numeric parameters. Free accounts can select up to 5 parameters with a cap of 125 combinations; Pro removes the parameter limit and raises the cap to 2000 combinations, sampling automatically beyond that.
Choose the scan range
Neighborhood (around current values) suits fine-tuning; full range suits exploring from scratch. Fast sweep does a coarse pass first, then refines around the best region. Free accounts run serially; Pro can run parallel workers (very high concurrency may make the browser sluggish).
Read results—not just returns
Every combination reports return, drawdown, win rate, liquidation flags, and an island risk score. Lower is more robust; 50 and above signals overfitting risk—the parameters only look good in a narrow value range, and neighboring values may liquidate or collapse.
Apply the best
Once the best combination sits safely away from liquidation and failure points, click Apply best to merge it into the current config, then re-run a backtest to confirm. Sweep history keeps recent results for review.
The highest-return combination is often not the one to use. Check the island risk score and liquidation warnings first: a high-return parameter sitting on an “island” can flip over on the next market shift. Prefer a slightly lower return in a flat neighborhood.