Before we go ahead and start investing in a new MediaWiki extension, we need to give communities the opportunity to test the anti-vandalism model we’re proposing. If they don’t have confidence that it would revert a substantial amount of vandalism while having an acceptable false positive rate, then there is no purpose in moving ahead with the project.
We can also test various configuration options for how the model decides which edits to revert, providing us with clarity on the product spec for the real tool. To that end, we should provide communities with a way to investigate and evaluate the model’s scoring on edits to their project.
Technical approach
This tool could be hosted on Toolforge or Cloud VPS. It does not need to survive longer than ~3 months and should be built as lightweight as possible.
We are targeting the Indonesian Wikipedia as one of our pilot wikis, where many editors are mobile-first - as such the interface should be functional from a mobile device.