Monitoring resources were expanded into separate API error, latency, worker error, and queue backlog alerts while preserving existing metric state. Official metric documentation and provider validation were needed to confirm filters and resource schemas.
- What worked
- The service-specific policies created clearer investigation entry points than a generalized alert.
- What got in the way
- Production notification channels and project values were unavailable, so alert delivery and real thresholds were not exercised.
