What to collect before you escalate

The ids, run header, failed tool calls, connection test and audit entries to gather before you ask for help.

How-to · Updated 2026-10-17 · Describes release 0.1.164

When the Troubleshooting hub and the page for your area have not solved it, gather these before you contact your platform administrator or support, so whoever picks the request up starts from evidence.

Before you start: open the incident in question; the audit entries need report.view. You should see every item below in hand before you send the request: each one is a question the person who picks it up would otherwise ask first.

What to collect before you escalate

• The incident id and investigation id, from the incident page URL and the investigation header.
• The investigation header line itself: model, token count, cost and any error. mock there answers most "why is the analysis generic" questions immediately.
• The tool timeline rows with status denied or error.
• The datasource name and its last test result, from Data sources.
• The relevant Audit log entries, which need report.view. Tenant-scoped audit is described in Audit log.
• Whether a colleague with a different role reproduces it. That one check separates a permission problem from a platform problem.

Next steps

• Troubleshooting hub — the symptoms, their causes and what to do.
• Troubleshoot incidents — symptoms specific to incidents.
• Status and error reference — every status word and HTTP error the product returns.