Eureka Case 25 min Decision & Risk

The Poisoned Pages

You are Marta Zielińska, Director of Community and Trust at the OpenSavoir Foundation in Warsaw. Over one weekend, coordinated accounts have poisoned 1,200 of your encyclopedia's most sensitive articles. Your crisis team meets at nine, and its first call is how to tell the attack apart from the volunteers.

The situation

Savoir is a free, multilingual encyclopedia written entirely by volunteers: 3.1 million articles in 14 languages and 210 million readers a month. The foundation behind it is small on purpose: 90 staff and a €24 million budget that comes almost entirely from small donations. It builds the tools and sets the terms of use; the editorial rules are written by the community. Of the 48,000 contributors who edit each month, about 1,900 hold permissions to approve, revert or protect.

Since Friday evening, automated and paid accounts have made 41,000 edits to 1,200 politically sensitive articles: subtly false claims, fabricated citations, framing bent a few degrees, written to survive the glance a tired volunteer gives a change at midnight. The quality index on those pages (0 to 100) has fallen from 82 to 61. A national newspaper has run the headline "The encyclopedia anyone can poison". The national election is six weeks away.

On 3 June the board expects the index back above 78, and two institutional grantors, worth €3.8 million (16% of the budget), expect a credible integrity plan. The board has made you chair of the crisis team and released a €600,000 emergency budget. The review queue was already 14 days behind, one experienced reviewer checks about 40 edits a month, and burnout threads have appeared on the forums. Those same pages draw about 4,000 good-faith edits in a normal week. Your standing with the 1,900 rests on never having overruled them.

Your job in this case: weigh the size of the attack against the attention your reviewers have, work out what each separation method leaves in the queue and on the pages, and choose how the foundation will separate coordinated abuse from good-faith editing.

The decision

How will you separate coordinated abuse from good-faith editing?

  • A · Triage by behaviour, then a human decides
  • B · Revert all 72 hours, restore good edits later
  • C · Treat every recent edit as suspect until read

The rest of the case is locked

The data, the calculations, the decision itself and what happens next open with your institution's access code.

Have an access code?

No code yet?

Instructors can request access for a course. Tell us who you are and which course it is for, and we will send your institution a code.

Run it live in class: every student decides and justifies, and the whole class sees how it splits. How Eureka Case works →