Skip to main content

Diagnose Live Trading Behavior

Diagnose an unexpected result by following the decision and order lifecycle in order. First define the exact strategy, account, symbol, candle timestamp, and expected outcome.

1. Verify the Deployed Setup​

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

Confirm the strategy package version, computed strategyRevision, deploymentCompositionId, complete configuration, account, connector, timeframe, symbol selection, enabled state, and pause state. Compare them with the reviewed revision, not with a local draft. In the app, use the strategy's Revisions action and chart markers to locate changes recorded in the trade window.

2. Verify Market Data​

Check that the decision candle is closed, present once, timestamped correctly, and identical across the live record and replay source. Verify warm-up history and any required derivatives, spread, global-market, or on-chain context.

If the strategy was not evaluated, investigate symbol selection, candle freshness, timeframe, process health, and data gaps before inspecting entry logic.

3. Trace the Strategy Decision​

Determine whether the strategy returned skip, entry, or exit and inspect the recorded reason and inputs. Recalculate the relevant indicators and state at that timestamp. Do not use data from later candles during diagnosis.

If replay and live evaluation disagree, compare package/configuration versions, warm-up length, timestamp alignment, and external context first.

4. Trace Filters and Risk Controls​

When an entry exists but no order was submitted, inspect each subsequent decision:

  • strategy and project hooks;
  • account and portfolio exposure limits;
  • pause state and order-placement permission;
  • AI/ML or other policy filters;
  • sizing, minimum notional, price deviation, and liquidity checks.

Record which control rejected the entry. A missing order is not necessarily a strategy mismatch.

5. Trace the Order Lifecycle​

If submission was attempted, follow the client order id through request, acknowledgement, rejection, partial fill, fill, cancellation, and reconciliation. Compare local state with the venue before retrying an unknown outcome. Check permissions, rate limits, precision, margin, position mode, and connector errors.

6. Reproduce the Window​

Use a narrow replay for the same strategy, symbols, and timestamps. For an entry-only comparison:

npx @tradejs/cli runtime-parity \
  --user <user> \
  --connector <connector> \
  --strategy <StrategyName> \
  --tickers <SYMBOLS> \
  --startTime <ms> \
  --endTime <ms> \
  --details \
  --cacheOnly

For exact version and execution comparison, collect a runtime record and follow Validate live decisions with replay.

7. Close the Diagnosis​

Document the observed behavior, expected behavior, exact versions and configuration, evidence, root cause, and corrective action. If the cause is uncertain or can create uncontrolled exposure, pause new entries while continuing to manage open positions.