Broken correlation is the silent tax on microservice dependency tracing. Teams swear every service is instrumented, yet the graph still fractures at a gateway or a message hop.
Start with the seams
Audit the boundaries that change protocol: HTTP to gRPC, sync to queue, mesh sidecar to bare pod. Most correlation bugs live in the adapters, not the business handlers. We keep a “seam checklist” in Trace Lab that learners run before blaming sampling.
Baggage with restraint
Baggage can carry tenant or experiment context across edges, but it is easy to overstuff. Prefer a short, documented set of keys. Overgrown baggage becomes a privacy and performance liability — and still fails to fix a missing traceparent.
Prove end-to-end weekly
A synthetic journey that asserts a single trace id across the critical path catches regressions faster than another dashboard. Pair it with your dependency clinic so broken seams surface as process issues, not heroics.
Need help designing that ritual? Write to the studio.