An operator wants to keep a failed instance available for later recovery. Which action should be avoided?

1. Inspecting its error details and activity stream.
2. Confirming that the instance is recoverable.
3. Checking whether its target dependency is healthy.
4. Discarding the message before the recovery decision.