Reliability

Watch the workflow: Scope reliability signals and inspect requests · Authored dummy data

Use Show all environments in Tensile to inspect authored demo signals after the Production view has no tagged events. Read the 12 of 15 tool-call success signal, the separate proxy and the unmeasured outcome signals, then follow Inspect requests to the broader 40-event Raw Logs view with the same period and environment scope. Event success and an authored proxy do not establish customer success. This clip uses dummy data and runs no evaluation.

Read the walkthrough transcript

• 00:00 — Show all environments in Tensile. These event signals use authored demo data.
• 00:05 — Follow Inspect requests. Event success does not establish customer success.

Open Reliability to see how your agent performed in a time range. The page shows an escalation signal, a resolution signal, tool-call success, a task-success baseline and mean time to resolve (MTTR). Use it first, before you look at single failures.

Before you start, send events to Tensile. Use the Quickstart or connect a trace source in Settings. If your events carry no environment tag, tag them as described in Settings.

Choose the scopeCopy Anchor Link

1. Use the time control to set the date range.
2. Use the environment control to choose an environment. The page opens on Production. Choose All environments to include every tag.
3. Use the customer list to show one customer.

If no event has an environment tag, the page shows all environments and explains the coverage. It does not show an empty scorecard.

Read the cardsCopy Anchor Link

• Observed escalation signal and Observed resolution signal use only sessions that carry an outcome signal. The card says not measured when no session carries one. It never shows 0% for missing data.
• Tool-call event success counts tool-call events. A successful event does not verify the customer outcome.
• Task success baseline names its source in the label: judge, proxy or mixed. A proxy value is a deterministic check. It is not a verified customer outcome. The card shows unavailable when the verdict read fails.
• MTTR is the time from the first observation of an incident to its resolution. It covers all environments.

Select the summary line on a card to see the counts behind it.

Check the trendCopy Anchor Link

The Escalation rate chart shows one point for each UTC day. Open View daily counts and missing periods to see the observed, measured and escalated sessions for each day. A day with no measured outcome shows not measured.

When two environments carry traffic, Reliability by stage compares Production and Staging in one table.

Empty and failed statesCopy Anchor Link

• No sessions match the selected dates and filters. Widen the range or clear the customer.
• A coverage message appears when the selected environment has no events but other events exist. Use its links to change the environment.
• If a query fails, the scope controls stay on the page with an error banner. Reload to try again.

Next stepsCopy Anchor Link

Open Triage to review detected incidents. Open Evals to see the verdicts behind the task-success baseline.