Moderation Postmortem

Moderation incident from May 2023 and inter-instance communication.

Hello Hachydermians! There has been a lot of confusion this week, so we’re writing up this blog post to be both a postmortem of sorts and a single source of truth. This is partly to combat some of the problems generated by hearsay: hearsay generates more Things To Respond To than Things That Actually Happened. As a result, this post is a little longer than our norm.

Moderation Incident

(A note to the broader, non-tech industry, members of our community: “Incident” here carries similar context and meaning to an “IT Incident” as we are a tech-oriented instance. Postmortems for traditional IT incidents are also in this section.)

A Short, Confusing Timeline

On 24 April 2023 the Hachyderm Moderation team received a request to review our Fundraising Policy via a GitHub Issue. The reason for the request was to ensure there was a well understood distinction between Mutual Aid and Fundraising. Although our Head Moderator responded to the thread with the constraints we use when developing new rules, some of the hearsay we started to see in the thread the user linked raised some flags that something else was happening.

In order to determine what happened, we needed to dive into various commentary before arriving at a potential root cause (and we eventually determined this was indeed the correct root cause). While a few of our moderators working on this, our founder and now-former admin Kris Nóva was requested, either directly or indirectly, to make statements on transgender genocide (she is herself openly transgender) and classism.

These issues are important, we want to be unequivocally clear. Kris Nóva is transgender and has been open about her experiences with homelessness and receiving mutual aid. The Hachyderm teams are also populated, intentionally, with a variety of marginalized individuals that bring their own lived experiences to our ability to manage Hachyderm’s moderation and infrastructure teams. That said: it was not immediately apparent that these requests were initially connected to the originating problem. In fact, it caused additional resources to be used trying to determine if there was a secondary problem to address. This resulted in a delay in actual remediation.

The Error Itself

On 27 Mar 2023 the Hachyderm Moderation Team received a report that indicated that an account may have been spamming the platform. When the posts were reviewed at the time of the report, it did trigger our spam policy. When we receive reports of accounts seeking funds, we try to validate the posts to check for common issues like phishing and so forth, as well as checking post volume and pattern to determine if the account is posting in a bot-like way, and so forth. At the time the report was moderated, the result was that the posting type and/or pattern was incorrectly flagged as spam and we requested the account stop posting that type of post. Once we became aware of the situation, the Hachyderm Moderation team followed up with our Hachydermian to ensure that they knew that they could post their requests for Mutual Aid, apologized for the error, and did our best to let them know we were here if there was anything else we could do to help them feel warm and welcome on our instance.

Lack of public statement

There have been some questions around the lack of public statement regarding the above. There are two reasons for this. First and foremost, this is because situations involving moderation are between the moderation team and the impacted person. Secondly, we must always take active steps to protect against negative consequences that can come with all the benefits of being a larger instance.

How Errors in Moderation are Handled at Hachyderm

Depending on the error, one or both of the following occurs:

  • Follow up with the user to rectify the situation
  • Review policy to ensure it doesn’t happen again

This is because of two enforced opinions of the Hachyderm Moderation team:

  • Moderation reports filed by users are reports of harm done.
    • To put it another way: if a user needs to file a report of hateful content, then they have already seen that content to report it.
  • All moderation mistakes are also harm done.

For the latter, we only follow up with the user if 1) they request and/or 2) we have reason to believe it would not increase the harm done. All Hachyderm Moderator and Hachydermian interactions are centered on harm mitigation. Rectifying mistakes in moderation are about the impacted person and not ego on the part of the specific moderator who made the mistake.

To put it another way, the Hachyderm Moderation team exists to serve the Hachyderm Community. This means that we will apologize to the user as part of harm mitigation. We will not as part of “needing to have our apology accepted” or to “be seen as apologizing”. These latter two go against the ethos of Hachyderm Moderation Strategy.

Inter-instance Communication and Hachyderm

Back when we took on the Twitter Migration in Nov 2022, we started to overcommunciate with Hachydermians that we were going to start using email and GitHub Issues (in addition to moderation reports) to accommodate our team scaling. As part of the changes we made, we also started making changes to reflect this in the documentation, as well as including how other instances can reach out to us if needed. It was our own pattern that if we needed to reach out to another instance, we used the email address they listed on their instance page. This is because we didn’t want to assume “the name on the instance” was “the” person to talk to. There are likely other instances like ours that have multiple people involved. This is reinforced by the fact that the “name” is actually populated by default by the person who originally installed the software on the server(s).

That said, we recently made a connection with someone who has grown their instance in the Fediverse for quite some time and has been helping to make us aware of the pre-existing cultural and communication norms in this space. In the same way it was natural for us to check for the existence of instance documentation to find their preferred way to communicate, it was unnatural for some of the existing instances to do so. Instead, it seems there are pre-existing cultural norms in the space we weren’t aware of, the impact of this is that some instances knew how to reach us and others did not.

We have been taking feedback from the person we connected with so that we can balance the needs of the existing communication patterns and norms of the Fediverse, while not accidentally creating a situation where a communication method ends up either too silo’ed or not appropriately visible due to our team structures.