# Diagnose a failed run

Locate the failed boundary without discarding accepted work.

Source: cd46e0e78d73497aec99e3e8d33c1fef632f36f9
URL: /troubleshooting

| Symptom | Next observation |
| --- | --- |
| Cannot connect | Check `berry info`, the selected runtime and its connection details. |
| Missing operation or model | Inspect `berry catalog` on that runtime; discovery and installed selection differ. |
| Build or preparation refused | Read its diagnostic and correct the source, binding or explicit permission. |
| Start reply lost | Retry the same request ID and input; retain any already returned identities. |
| Wait timed out | Inspect the execution. A wait timeout does not stop work. |
| Execution waiting for input | Inspect its thread and supply the required input. |
| Execution failed | Inspect the outcome and child calls before deciding whether a new operation is appropriate. |
| Update conflicts | Read the installation's current revision and reconcile the intended selection. |

## Keep useful evidence [#keep-useful-evidence]

Retain the selected runtime, agent, installation revision, request ID, session,
execution and thread IDs. Include the original diagnostic and the source revision
when reporting a documentation problem. Exclude credentials and private content.

Resetting runtime History is not a diagnostic step. Installation updates,
continuation and uncertain external effects have their own recovery decisions.

The [CLI reference](/reference/cli) describes inspection commands. The
[HTTP reference](/reference/http) separates host-only and remotely available operations.