OpenTelemetry Does Not Replace a Developer Error Inbox

OpenTelemetry is how you see latency, status, and dependencies for calls that reached a process you instrumented. A developer who failed npm install, or whose CLI exited before login, produced no span.

That gap is the silent part of an SDK evaluation. The dashboard of traces stays healthy because the people who failed never became traffic.

A 401 span does not say whether they were on staging, following the sample, or holding a token from the wrong workspace. One sentence at the terminal does.

Put the stable error code on the span and on the probe so a reviewer can jump from one to the other. Do not put the sentence into a trace attribute that gets sampled away.

Head-based sampling keeps the happy path. The one evaluator who hit a rare TLS error is the one you needed. Error probes should be separate from the trace sample rate.

Do not attach request bodies to either pipe. The code, the route template, and the redacted stack are enough.

Use traces to see why a route is slow. Use the developer inbox to see why someone never came back. A status page answers neither question on its own.

AethelLayer is the second system: it keeps the stack from the CLI or SDK, asks what they were trying to do, and drafts a GitHub issue a human still sends.