Skip to main content

From Backtest to Live Trading

This playbook connects research, validation, and live operation without tying the process to a particular strategy, exchange, or account. Replace angle- bracketed values with settings from your project.

1. State the Experiment​

Before running commands, record the hypothesis, strategy version, parameter range, symbols, timeframe, data window, fees, slippage, fill model, and acceptance criteria. Define the out-of-sample window before selecting a result.

2. Run a Small Reproducibility Check​

Start with a few symbols and a limited parameter set:

npx @tradejs/cli backtest \
  --user <user> \
  --config <StrategyName:configName> \
  --connector <connector> \
  --timeframe <minutes> \
  --tickers <SYMBOL1,SYMBOL2> \
  --tests <small-limit> \
  --parallel <workers> \
  --cacheOnly

Confirm that candles, timestamps, trade direction, sizing, fees, entries, and exits match the intended rules before scaling the run.

Expand only to the predefined symbols, time window, and parameter grid. Keep the command, configuration, package versions, and output together. Avoid adding parameters after seeing results unless you start a new experiment.

Inspect the result set and symbol coverage:

npx @tradejs/cli results \
  --strategy <StrategyName> \
  --coverage \
  --user <user>

4. Validate the Candidate​

Evaluate the selected configuration on untouched data and across relevant market regimes. Stress fees, slippage, entry delay, spread, and nearby parameter values. Review the full trade distribution, including drawdown, recovery time, exposure, turnover, and concentration. Aggregate profit alone is not enough.

Reject the candidate if its result depends on a narrow parameter point, unrealistic fills, a few symbols, or a small number of trades.

5. Freeze the Live Settings​

Copy one complete reviewed configuration into the target deployment in tradejs.config.ts. Pin the exact strategy package and lockfile, then review the account, connector, symbol selection, and risk limits. Do not add a manual version; Project checks parse the complete config and compute both runtime identifiers.

Verify the resolved settings:

npx @tradejs/cli runtime-control verify \
  --user <user> \
  --deployment <deployment>

Record the returned strategyRevision and deploymentCompositionId with the candidate evidence.

6. Replay the Exact Deployment​

npx @tradejs/cli replay \
  --user <user> \
  --deployment <deployment> \
  --days <days> \
  --cacheOnly

Investigate missing entries, extra entries, timestamp drift, price drift, and differences in gates or market context. Do not edit historical records to make them agree.

7. Observe Current-Market Evaluation​

Run without order placement first:

npx @tradejs/cli signals \
  --user <user> \
  --deployment <deployment> \
  --cacheOnly

Check candle freshness, resolved symbols, strategy decisions, notifications, and monitoring. Confirm that pausing new entries does not prevent management of existing positions.

8. Start a Bounded Live Rollout​

Enable order placement only with explicit authorization and venue credentials. Use a small fixed risk allocation, define maximum loss and operational stop conditions, and retain a tested pause/rollback procedure. Compare decisions, orders, and fills with replay regularly.

Related reading: