Live systems fail quietly. A feed latches on a stale price and the strategy keeps trading a number that stopped moving twenty minutes ago; a token expires and orders silently reject; a process restarts and forgets it was holding a position. Monitoring exists to make those things loud.
What is monitored
- Process liveness and heartbeat interval per strategy
- Quote age and feed staleness, per instrument
- Order rejection rate, broker error clusters, latency spikes
- Position drift between the strategy's view and the broker's book
- Realised loss against the daily cap, with auto-flatten if you want it
Response
Alerts route to you and to us. During market hours we investigate, apply the runbook, and either fix or halt. Every incident gets a short written post-mortem: what happened, what it cost, what change prevents a repeat. We are not a call centre — this is engineering support with an audit trail.