Observability is a design property, not a dashboard installed after release. Instrumenting user-critical flows and operational bottlenecks helps teams understand whether architecture choices are working.
Connect signals to a question
Metrics show trends and thresholds, traces show a request across boundaries, and logs preserve discrete context. Collecting all three without a diagnostic question creates cost rather than clarity.
Carry context across boundaries
A correlation identifier should follow work through API, queue and worker boundaries. Domain identifiers can add useful context, but sensitive data and credentials do not belong in telemetry.
Instrument the product outcome
A successful HTTP response does not prove that a statement reconciled, a notification reached its destination or an upload became usable. Record meaningful domain outcomes alongside technical health.