Tracing

Following one particular request the whole way through, with a timestamp at every stage. This exact question arrived at 10:42, the search for relevant documents took a second and a half, the AI thought for four, this answer went out. One job, followed end to end.

Tracking a single parcel, rather than reading the courier's monthly performance report.

It's the single-thread view inside observability: observability is seeing the whole system, a trace is one thread pulled out of it.

Why it matters: when this answer was wrong, or that one took forty seconds, the trace tells you which step did it. Without one you end up re-arguing the whole system on a hunch. You'll never read a trace yourself — the useful version is the question: if a customer complains about one specific answer, can you pull up exactly what happened to it?

Related: Observability · Logging