The idea of using ORES for surfacing rated revisions in recent changes is reasonable. However, there is still excessive amounts of noise in the revisions, and recent changes is definitely has a lot of usability problems.
One option is to generate a report (e.g. Special:RevisionReport) that shows only the highest likely and and 100% content damage and vandalism, for example:
- Page blanking - there is rarely a need to blank a page, either it needs to be deleted or improved
- Huge amount of random characters - e.g. "asdasdASDZXczxc"
- Emoji: lots of emojis in an article
- Hoaxes: If it is even possible to detect that
- Repetitive text: The "buy viagra ... buy viagra" kind of thing
- Similar entries in multiple articles: Often vandals hit several articles adding nearly identical text
The idea is that people can simple focus on these, and clean them up gradually without looking into the whole recent changes. While there are improvements on the horizon for recent changes, it is still a tool attempting to fulfill too many use-cases and failing miserably at a lot of them.
Another idea, would be to flag potential bad actors. While this may be controversial for registered users, it might be perfect for IP addresses to identify if there is a particular pattern of behavior for an ip range.