Trace Inspection
The platform trace page is the operational center for Captar v1. It shows the span tree, timing, payload retention, spend events, and any blocked or failed execution that happened in the session. Use it as the first stop when you want to answer questions like:- What started this run?
- Why was a call blocked?
- Which tool caused the side effect?
- How much budget was reserved and committed?
What to inspect first
- Root session metadata
- Request and tool span hierarchy
- Prompt and response payloads that were retained
- Guardrail violations
- Spend reservation and committed usage
Strong traces are what make dataset export and manual eval review valuable. If
a trace is noisy or incomplete, it is a weak source row later on.
Good trace hygiene
- Keep metadata short but specific.
- Retain prompts and responses only when they are useful to review.
- Give tools stable names that match what the team says out loud.
- Export runs that are representative, not just the happiest path.