Optimization

Optimization tests every combination of parameter values to find the set that scores highest on a chosen metric. It runs a full backtest for each combination (grid search).

How It Works

For each parameter marked optimize: true, a range of values is generated from its min, max, and step
The Cartesian product of all ranges produces the full grid of combinations
Each combination runs a complete backtest
Results are ranked by the optimization target metric
The best-scoring combination is returned along with all results

Grid Size

The total number of combinations is the product of step counts:

fastLength:  min=5,  max=25, step=5  → 5 values  (5, 10, 15, 20, 25)
slowLength:  min=20, max=60, step=10 → 5 values  (20, 30, 40, 50, 60)
rsiPeriod:   min=7,  max=21, step=7  → 3 values  (7, 14, 21)

Total combinations = 5 x 5 x 3 = 75 backtests

Limit: The maximum grid size is 50,000 combinations. If your ranges produce more, reduce the number of parameters or increase the step sizes.

Optimization Targets

Target	What It Maximises
Sharpe Ratio (default)	Risk-adjusted return
Total Return	Percentage return on capital
Total Profit	Absolute profit amount
Win Rate	Percentage of winning trades
Profit Factor	Gross profit / gross loss
Calmar Ratio	Return / max drawdown
Max Drawdown	Minimises maximum drawdown
Composite	Blended: 40% Sharpe + 30% win rate + 30% profit factor
All	Returns the best result for each major metric

Fixed Capital

Optimization always uses fixed (non-compounding) capital for each backtest. This prevents early-period results from snowballing via compound growth, which would bias the parameter search toward strategies that got lucky early.

Execution Modes

Mode	Behaviour
Synchronous	Blocks until complete — good for small grids
Asynchronous	Returns a job ID immediately; progress is streamed via SSE (Server-Sent Events)

Large optimizations automatically run asynchronously. You can monitor progress in real time and the results are saved for 7 days.

Tips

Start with coarse steps (large step sizes) to explore broadly, then refine with finer steps around the best region
Use Sharpe Ratio as the default target — it balances return and risk
If you see the best result at a boundary (min or max), extend the range
Watch out for overfitting: a perfect-looking result on one date range may not generalize. Use Walk-Forward Analysis to validate.
Keep the grid under 5,000 combinations for fast iteration; save larger grids for final validation