Goal

We want to introduce a tool to assist patrollers and checkusers in sockpuppet detection work by automating the bulk of the work they do when determining if two or more users are sockpuppets of each other. We imagine that this tool will drastically reduce the amount of time and effort patrollers and checkusers currently spend on detecting sockpuppet accounts and allow them to work more efficiently.
In addition to that, it would serve to benefit many patrollers when IP addresses are masked by providing insight into sockpuppet activity without IP information.

Product Spec

This project will build on the research work done by the Wikimedia Foundation Research team on detection of sockpuppets. The script is hosted on one of Wikimedia’s Kubernetes instances as an API (“similar-users”).
Relevant links:

• Technical walkthrough for the Kubernetes hosted instance
• Parameter documentation for the API
• Response documentation

V1 goal: Improve prediction script accuracy

• Port the existing tool into a MediaWiki extension
• Limit access of the tool to checkusers
• Build functionality to enable checkusers to provide feedback on the script predictions (through QuickSurveys)
• Share page with limited pool of checkusers
• Research to use the feedback to improve the prediction script
• Limited to english wikipedia (as the script support is limited)

V2 goal: Build a feature to allow editors to detect sockpuppets more quickly and easily

• With the improved prediction script, build out a tool that allows quicker and easier detection of sockpuppets
• The tool scales to all languages and performs at a reasonable performance level
• Users with certain permissions will be able to access the tool

Design Spec

• Use existing HTML form and MW pagers
  • UI similar to system messages and blocked users
• TBD workflows
• Feedback mechanism: T297687: Allow users to submit feedback via a QuickSurveys survey
• TBD definitions for table headers

Technical Spec

1. Create the extension
2. Create the special page
3. Gate access
4. Display

Success metrics

TBD

Rollout plan

Phase 1: Access to checkusers only
Phase 2: Access to patrollers for checking IP socks
Timeline TBD