Maelstrom was much more difficult to test than to write, because few of us would enjoy releasing a tool of such powers to do damage to our network during testing! Thus testing required writing a simulator whose input was a known set of task precedences, to see if Maelstrom could sort them out without foreknowledge of their precedences. It does this as expected.
Quite obviously, the success of Maelstrom depends upon the quality and reliability of the scripts it dispatches. To date, we have not deployed Maelstrom in production, because we are not yet confident of our scripts' convergent properties. The scripts that we envision using in production are largely targeted at restoring specific mission-critical services, and not at addressing systemic failures or network connectivity as yet. A typical script tries a netcat from a remote device to see if a service is working from outside a server and then attempts to repair any problems inside the server, as is possible by using Babble.