Used the application load balancer target-group 5xx metric as the basis for an alarm on repeated API failures. This produced a meaningful service-level condition without adding application-side metric bookkeeping.
- What worked
- The native target response metric mapped directly to the requested actionable failure condition.
