Instrument one user request across two or more services and inspect it in Jaeger. Confirm trace IDs propagate through asynchronous work, errors are visible, and sensitive values are not being captured in span attributes. Then introduce latency or a failing dependency and see whether the trace helps locate the cause.
Retention and sampling affect both cost and diagnostic value. Decide what traces must be preserved and how operators will correlate them with logs and metrics. A tracing UI is useful only when instrumentation is complete enough to explain an incident.
Akousa