Use LunaBot / 7 min read
Research & backtests
Set up a historical test, compare changes and read the results with the settings that produced them.
On this page
Start a backtest
- Save a strategy first, then open Research → Add research.
- Select the intended strategy, market, timeframe and version.
- Choose Single replay for a first run, or Walk-forward windows to test successive later windows.
- Enter Closed bars to load (200–10,000) and Fixed test volume · lots.
- Review assumed spread, slippage, commission and the other displayed valuation settings.
- Select Run backtest.
Test volume and assumed costs describe the run. They do not change the account’s live trading rules or place orders.
LunaBot 1.0.1-rc.1 · current interface in an isolated Paper profile. Simulated data; no broker trade.
Open a run
The list shows strategy, market, method, requested bars and status. Choose View details to read progress or a completed result. An active run can be cancelled using Cancel run.
Open Run configuration to see the settings saved with that run. If it fails, read the error before changing the strategy or starting another test.
Read the results
- Net P&L
- The test’s profit or loss after its modelled costs.
- Closed trades and win rate
- How many trades completed and the share that ended with positive net results. A small trade count gives limited information.
- Profit factor
- Gross profit relative to gross loss. Read it alongside trade count and drawdown.
- Equity curve and drawdown
- How the modelled account value changed. Bar-close sampling can miss movements within a candle.
- Requested, supplied and scored bars
- Different counts can reflect available history and the candles needed to initialise indicators.
Read the assumptions and limits below the results. A qualification label refers to the app’s stated checks; it does not guarantee future results.
Change settings and compare
A completed run keeps its saved configuration. To test a change, select Add research and create another run. To change the strategy itself, edit it in Strategies, save, then select the intended version for the next run.
Change one assumption at a time where possible. Compare the same market, timeframe, bar range and costs before drawing conclusions from a difference.
Other research methods
- Train, then test
- Separate earlier and later portions of the data using the configured windows.
- Walk-forward windows
- Inspect successive training/test windows. Training prefixes are not automatically fitted or optimised.
- Compare a parameter
- Compare declared parameter variants.
- Stress trading costs
- Compare declared spread, slippage or commission scenarios.
Use the labelled controls first. Training ends at bar is an absolute boundary within the loaded history and includes warm-up. Keep warm-up below that boundary, and the boundary below the total loaded bars. For example: 2,000 loaded bars, 500 warm-up bars, training ending at bar 1,000 and 500 bars per test window. Each child needs at least 200 bars including warm-up. Incomplete final windows are not scored.
Expert options · JSON is closed by default. Opening it makes its JSON replace the guided settings; review it before submitting. Training prefixes reserve earlier data but do not automatically fit or optimise strategy parameters. The summary shows the window layout before the run.
Strategy research preview
In the 1.0.1-rc.1 candidate only. Open Research → Strategy research, import the supplied catalog, and choose a market. The 45 entries are starting hypotheses, not ready-to-trade strategies. Open an entry to read its rules, competing explanations and missing requirements.
Use the process check to test the research selection machinery on synthetic data. Its saved settings and attempt count stay attached to the run, including cancellation. Read the dedicated process-check result, not a trading P&L: synthetic detection is not a market win rate.
Historical screens use the existing backtester only where rules and data are supported. Missing clock alignment, session data or costs are blockers, not automatic passes. Neither a completed job nor a passing software test activates a strategy. No new candidate has qualified for Live trading.
Review a trading problem
In the Research review area, start a review of an account, strategy, Scout, proposal or order. LunaBot saves the available records, then Run checks compares them for inconsistencies.
“OK” means a particular record check passed. “Not enough data” means the saved records cannot answer that check. Creating a review does not start an AI session; use a connected assistant if you want further analysis.
Core workflow with 1.0.1-rc.1 preview noted separately · Updated 4 October 2026 Suggest a correction ↗