Backtest Settings and Reproducibility
Published Jun 22, 2026 · Updated Oct 16, 2026
To reproduce or compare a backtest, check the input snapshot of its run version. The same template, a similar curve, or the same strategy name does not establish that two runs used the same inputs. Go to Strategy → Strategy backtest.
If you do not have an experiment yet, start with Your First Strategy Backtest. This guide covers checking inputs, confirming saves, comparing versions, and using a deliberately selected version for Paper trading.
Which inputs should I record before running?
| Input (page field) | What to check |
|---|---|
| Strategy parameters, execution and exposure, risk control | Record the version number you compare |
| Market source, symbols | Confirm the source has history for these USDT perpetuals; if not, the page shows “Market check failed” and suggests available sources |
| Start date, end date (inclusive) | Dates use UTC; the end date includes the whole day, not midnight in your browser’s timezone |
| Simulated backtest candle interval | The candles used to replay the price path, 1 minute at the finest; indicators such as RSI still download at the strategy’s own timeframes |
| Initial equity (USDT) | Keep it consistent across compared runs; it does not become your Paper account balance |
| Account membership | Comes from your signed-in account and cannot be changed for a backtest; it affects simulated execution cadence (the page notes 10 s for Pro / Pro Max and 60 s for Free) |
The Web uses the default execution model. After a run, check the execution model, intra-candle price path, maker / taker fees, slippage, position mode, data quality, and engine version in the version snapshot under Saved backtest parameters. A snapshot records the run’s inputs; it does not mean every field can be edited in the Web form.
How do I confirm a run was saved?
After the backtest finishes locally in your browser, the page saves the new version to the server. Record the experiment name, version number, and run time, and find the version under Saved backtest parameters.
- If you see “Backtest finished, but saving to the server failed”, click Retry save. A visible curve does not confirm a save.
- An older version marked Curve not saved still keeps its parameters and summary metrics. Apply its parameters and run again to see full results; a missing curve does not mean zero return.
- Detailed trades and the saved summary are different records. To investigate a fill, also record its symbol, time, and version.
How do I reproduce and compare two versions?
Select the version to reproduce
Open the experiment from the backtest experiment list, go to Saved backtest parameters, select the version, click Use this version’s parameters, and confirm. The draft is restored to that version’s strategy configuration, market source, dates, initial equity, and simulated candle interval. Switching the result you are viewing does not replace the draft.
Account membership still comes from your current account, and the execution model is still the current default. After an upgrade or an expired trial, matching parameters do not mean matching run conditions, so compare both with the original snapshot.
Check the inputs, then run
Check the table above, then click the run button (shown as Run backtest, or with the next version number, such as Run V2). Every run creates a new version; even with matching parameters, record changes to data, the engine, or the execution model.
Compare two versions
Tick 2 versions in the list under Saved backtest parameters. Compare versions shows return, drawdown, Sharpe, trades, and other metrics side by side and lists the config differences. Once you decide, click Apply vN parameters.
Explain one change at a time: to compare a parameter, keep the dates and other inputs fixed; to look at another market period, keep the configuration fixed and change only the dates. For example, apply version A, change one strategy parameter, and run version B—that is a recorded parameter comparison. If you also switch the market source, dates, and equity, the difference cannot be attributed to that parameter alone.
The trade count records filled reductions or closes, not signals or entries. With zero trades, also check for open positions, pending orders, or data warnings. Historical returns do not guarantee later performance.
Deleting a run version or a whole experiment cannot be undone, and deleted version numbers are never reused. Record version numbers and snapshots first if you need them for comparison.
How do I use the selected version for Paper trading?
Put the target version back in the draft
Under Saved backtest parameters, select the version, click Use this version’s parameters, confirm, and check the current draft.
Open the create dialog
Click Create strategy from parameters. The dialog copies the latest parameters from the editor and creates the strategy directly with an account you choose; it does not open the three-step New Strategy form.
Choose the Paper account
Enter a Strategy name and pick your prepared paper account under Trading account (shown as “AlphaFox Paper”). The dialog also lists live accounts and shows a Live badge when one is selected, so check before you submit. If no account is available, add a trading account or stop the strategy using one.
Check the cost and create
Check the estimated cost in the dialog, then click Create and start. “Strategy created; start requested” means it worked; click View strategy to open its detail page.
Paper starting funds are set in Account → Trading account; backtest dates and initial equity never overwrite them. See First Paper Strategy for runtime status, orders, and stopping, and Backtesting vs Paper Trading vs Live for how historical replay differs from current-market simulation.
To use your own Agent, follow the AlphaFox CLI installation guide to install the CLI and its Skills. That guide covers installation and authentication; use CLI command help for individual command arguments.