For power cycle events, it would be proper if the host is unreachable to do the folowing:

• Check the status of the instance
• If the instance is stopped, attempt to start it
• If the start fails then mark this as a system failure

This could be done in evergreen.yml in the post and timeout phases. For the powertest.py script, it can also retry to start if it detects a stopped instance, however it would be more difficult to mark this as system failure, unless a dedicated exit code is used,