The Computational Democracy Project


πŸ‘Ή Moderation

  • Moderation in Polis is available to and performed by the owners of each conversation via the admin interface
  • As each comment enters the system after being submitted by a participant, the owner will have the opportunity to approve or reject it, or leave it unmoderated.
  • Polis does employ comment routing to intelligently choose comments semi-randomly, to maximize participant time on meaningful tasks.
  • Polis will work without moderation.
    • You can, however, save a considerable amount of participant time and energy by moderating out nonsense or irrelevant comments like
      • t4wrgdfsa
      • I am on the bus and looking at the sky while reading this
    • These actions on your part are multiplicative: perhaps 1000's of participants would have decided to pass on a comment, and now those 1000's of votes are going towards meaningful and relevant issues.

Moderation Schemes

  • There are two moderation schemes schemes: strict moderation and permissive moderation
    • A conversation must choose one of the two moderation schemes, the default is strict
    • strict moderationΒ is a whitelist model: no comments are shown to participants unless they have been accepted into the conversation by the moderator
    • permissive moderation is a blacklist model:Β all comments wil be shown to users as soon as they are submitted, but can be moderated out after the fact
    • If you are embedding Polis on your site, it's a good idea to turn on strict moderation, as participants can, in effect, display profanity directly onto your digital property. It's a primary feature that what users type is turned right back around into what others vote on.
    • Switch between strict moderation and permissive moderation by toggling strict moderation on and off in the configuration screen in the admin interface for your conversation


  • It is a preferred practice to have a basic rubric for πŸ‘Ή Moderation. This helps in communicating expectations to participants (who do not get feedback about whether their comment was moderated in or out), and providing consistency. Here's a framework for creating one.
    • participants vote, on average, between 20 and 60 times, with some highly engaged conversations averaging over 100 votes per participant. There is a long tail on the distribution of votes: a vanishingly small percentage of participants will vote on upwards of 600 to 800 comments. Still, a large conversation of thousands of participants have 1000 or more comments.
    • It is helpful to moderate out comments which are spam
      • Assume comment 1 says dfs34f43sdf
      • Assume comment 2 says lol whatever
      • Reject both of these comments
    • It is helpful to moderate out comments which are duplicative.
      • Assume comment 1 says We need better traffic management
      • Assume comment 2 says Traffic management has been really bad
      • If comment 1 has been accepted into the conversation, reject comment 2. Keeping duplicate statements out of the conversation maximizes the time and votes of participants.
      • If the comments are nuanced, and subtly different in terms of what they mean and intend, then, of course, allow them both in.
    • It is helpful to moderate out comments which state more than one idea, because it will confound analysis. It is difficult to tell which part of a comment a participant was agreeing or disagreeing to.
      • Assume comment 1 says We really need to treat opioids as a health emergency rather than use the criminal justice system. Doctors who overprescribe should be imprisoned instead of addicts.
      • Reject this comment, copy it, break it up into its parts and re-submit it in seed comments as two separate comments.
  • It is a preferred practice to release the full data export as open data when the conversation is finished, and to let participants know this will happen. The data export does include all comments, and shows which were moderated out.

Closing commenting

  • If you are leaving your conversation open for weeks, after a while comments will get duplicative. It's a good idea to close commenting in configuration the admin interface at this time and let participants know you will be doing so. That gives participants time to come back and vote on all comments, and saves energy on the part of the moderators


  • Note that rejected comments cannot be moderated back into the conversation, but that you can copy and paste the text back in as a seed comments

The Computational Democracy Project

Β© 2024 The Computational Democracy Project, a 501c3 nonprofit.


Β© 2024 The Computational Democracy Project, a 501c3 nonprofit.