Live Trading Runbook
This runbook documents the minimum production operating model for ta4j-based live systems.
Scope and ownership
ta4j provides:
BarSeries and ConcurrentBarSeries for market state- indicators, rules, and strategy evaluation
BaseTradingRecord for position/fill state and downstream analytics
Your system still owns:
- market-data ingestion and historical backfill
- order routing and retries
- reconciliation against broker account state
- persistence, recovery, and alerting
Standard control loop
- Ingest bars/trades into series
- Evaluate strategy on close (or your chosen deterministic checkpoint)
- Emit order intent
- Update trading record from broker-confirmed fills
- Publish metrics and risk snapshots
Reference docs:
Startup checklist
- Load latest strategy configuration
- Rebuild or load the latest
BarSeries snapshot
- Load persisted
TradingRecord snapshot and broker mapping keys (orderId, correlationId)
- Reconcile open orders/positions with broker before evaluating new signals
- Verify clock skew and trading session boundaries
- Confirm alerting and logging pipeline health
Runtime controls
Data integrity
- Alert when expected bars are missing or delayed
- Track duplicate/late bars and apply deterministic handling
- Detect timezone or session-boundary drift
Execution integrity
- Never treat an order intent as a fill
- Only mutate
TradingRecord from confirmed fills (operate(fill) or operate(Trade.fromFills(...)))
- Track reject/cancel/partial-fill rates per venue and symbol
Risk integrity
- Maintain max position and max notional safeguards outside strategy logic
- Apply emergency kill-switch conditions (connectivity, stale feed, repeated rejects)
- Keep bounded exposure during reconnect/recovery periods
Persistence model
Persist at minimum:
- latest processed market timestamp/index
- latest strategy descriptor/config hash
TradingRecord snapshot (including fill metadata and fees)- broker-side IDs needed for idempotent recovery
Persist before marking events as durable in downstream systems.
Recovery and reconciliation procedure
- Pause new strategy-triggered submissions
- Reload series and trading record from durable state
- Fetch broker open positions and open orders
- Reconcile differences:
- missing local fills -> backfill into
TradingRecord
- unknown broker orders -> quarantine and investigate
- Recompute open-position metrics (
getCurrentPosition, getOpenPositions)
- Resume signal evaluation only after state convergence
Incident playbook
Stale or missing bars
- Quarantine strategy decisions for affected symbols
- Backfill bars from a trusted source
- Re-evaluate pending signals if policy requires
Repeated order rejects
- Disable affected strategy routes
- Capture reject payloads and classify root cause
- Resume only after routing/risk-policy correction
Divergent position state (local vs broker)
- Stop automated exits/entries
- Force reconciliation cycle
- Re-enable automation after explicit state match
Observability minimum
Track and alert on:
- bar lag and ingestion gap counts
- strategy signal count vs confirmed fill count
- rejected/cancelled/expired order rates
- reconciliation mismatch count
- unrealized and realized PnL deltas
References