Getting better at blocking bad activity on wikis

Happy new (nonprofit fiscal) year! 🥳

July 1 marked the start of the FY24-25 annual plan for Wikimedia Foundation. Together with colleagues, I've been working the last couple of months on lining up some interesting projects for improving anti-abuse capabilities on the wikis under the WE4.2 key result.

There are four projects hypotheses planned for this quarter:

• WE4.2.1 Unique device model: We currently rely heavily on IP address blocking to mitigate abuse. Can we start on infrastructure foundations to allow for mitigating actions based on a unique device identifier, to reduce our reliance on IP blocks that harm good faith contributors?
• WE4.2.2 Account reputation scores: To review edits from accounts, one currently has to build a mental model using a variety of tools. Is it possible to provide a useful account reputation score or bracket to categorize an account, such that we reduce the mental burden on patrollers and functionaries?
• WE4.2.3 CAPTCHA evaluation framework: Our existing CAPTCHA has questionable efficacy and accessibility problems. Can we define a framework for assessing how effective the existing CAPTCHA is, so that we can make a meaningful comparison with potential replacements?
• WE4.2.9 Contextual IP reputation data on Special:Block: Can we reduce the potential for unintentional collateral damage blocks by providing information about an IP reputation and the potential impact of blocking an IP, in-context on the Special:Block page?

All of these hypotheses bubble up to the general umbrella of work of providing signals for use in anti-abuse workflows, to improve the precision of actions on bad actors. All of the work is experimental; my hope is that we learn through trying and by the end of the year have some building blocks to make life easier for people doing anti-abuse work, while raising the level of difficulty for bad actors.

I'll be speaking about this work (virtually) at Wikimania in August, hope to see you there!