Per the parent task, we are just about ready to launch support for the multilingual revert risk in Automoderator. We need to communicate with the wikis supported by this model that already have Automoderator deployed to let them know it is available and worth trying out.
Message
Translations in comments.
Hi - I want to let you know about an upcoming feature update to Automoderator ([LOCAL NAME]) which will cause it to stop working on your Wikimedia project until it is turned back on by an administrator. Please read on to find out what steps are required.
On Wednesday, November 19th we will be rolling out an update to Automoderator which will enable you to choose one of two machine learning models to power the software. Currently, Automoderator uses the [[:meta:Machine_learning_models/Production/Language-agnostic_revert_risk|Language-agnostic Revert Risk model]] on all Wikipedias. This update enables you to instead use the [[:meta:Machine_learning_models/Production/Multilingual_revert_risk|Multilingual Revert Risk model]] to determine which edits should be reverted. We believe that this model should perform better - it will revert more edits with a lower false positive rate.
For technical reasons, when this change is deployed, the configuration page for Automoderator will move from [[Special:CommunityConfiguration/AutoModerator]] to [[Special:CommunityConfiguration/AutomoderatorMultilingual]]. As a result, Automoderator will turn off, until the new configuration page is set up by an administrator. The old configuration can still be viewed at [[MediaWiki:AutoModeratorConfig.json]], so it can be used as a starting point for reconfiguring the software. This configuration page can be safely deleted when you no longer need it.
Because the Multilingual model performance varies more by language as compared to the language-agnostic model, the configuration is slightly different, if you choose to use it. Instead of selecting one of a few predefined thresholds, administrators can instead define the threshold precisely, as a number. This number can be any value between 0.85 and 1. We recommend that you start with a high value (for example 0.99), and reduce it over time as you evaluate the false positive rate. If you would like to access some data on past edits and their multilingual revert risk scores to help you define a threshold to use, please visit [[:mw:Moderator Tools/Automoderator/Multilingual testing|Moderator Tools/Automoderator/Multilingual testing on MediaWiki]].
If you would like to continue using the current configuration, that's not a problem - Automoderator still supports the same language-agnostic model that you've been using until now. If you have any questions or concerns, please let me know.