Our current monitoring setup does not notify us when the ORES workers go down. This is a problem because it is out most common source of downtime.
We have a dummy model for testwiki that would be perfect for this. E.g. requests to this URL should always return 200 OK:
http://ores.wmflabs.org/scores/testwiki/reverted/<revid>/
And you can put whatever integer you want for the <revid>. I think a unix timestamp would work great to make sure we don't pull from the cache. E.g. it's 1450302568 right now so
http://ores.wmflabs.org/scores/testwiki/reverted/1450302568/
will return
{ "1450302568": { "prediction": true, "probability": { "false": 0.14, "true": 0.86 } } }