EVALS & OBSERVABILITY • 251 / 397
Measure whether the AI is good and trace where it fails.

Trace

A trace is the end-to-end record of one request or task across multiple steps.

Think of it like

Think of Trace like testing and telemetry for any production service: quality must be measured, not guessed.

Real life

Teams use Trace to decide whether a new release is safer, faster or more useful.

SRE lens

One incident investigation may have dozens of child operations under a single trace.

Remember thisA trace is the end-to-end record of one request or task across multiple steps.
AIForSREJump to a concept
Search is optional. The main journey is simply ↓ Next.