Federated social platforms like Mastodon and Bluesky are growing alternatives to centralized social media. Their moderation actions cross servers, lists and labelers in ways users cannot easily see. When something visibly changes, users turn to public forums like Reddit to ask why. We read these discussions as folk theories of federated moderation: accounts users build from observation and discussion with each other. We pair manual qualitative coding with LLM-assisted labeling using Gemma 4 (for extending the analysis across a larger set) to map how users explain moderation across five federated-platform subreddits. Two cases stand out: defederation on Mastodon, where users settle on a few recurring reasons, and list propagation on Bluesky, where users report being mass-blocked via a stranger’s subscription list. We close with two design changes that would surface these explanations on the platform itself rather than on Reddit.

Federated social media has grown noticeably over the past few years. Mastodon and Bluesky are the most visible cases, alongside Lemmy as a Reddit alternative, with millions of accounts either migrating from centralized platforms or signing up directly [6]. Unlike centralized platforms, federated platforms divide moderation work across many independent instances and, on Bluesky, across user-subscribed lists and labelers. This division makes moderation outcomes harder for ordinary users to see: an account can be defederated by a server its owner has never heard of, or added to a block list curated by someone they have never spoken to. A new Mastodon user usually does not know which other servers their instance has cut ties with. A new Bluesky user often does not know which labelers their app trusts by default, or that a stranger curating a list can add their account and quietly lower their reach. When something visibly breaks (an account goes dark, replies fail to land), people often take the question to Reddit. The answers that come back vary, from accurate to confidently wrong.

…

The contrast between the two cases also shows that folk theories of federation are not generic. The dimensions users reach for in each case track what their protocol makes visible to them. On Mastodon they reach for the admin and the question of how far admin power extends, because the admin is the named human role in their experience of the platform. On Bluesky they reach for the list curator and the mechanics of subscription, because lists and labelers are the named user-side mechanism the protocol exposes. Where prior folk-theory work has studied a single algorithm that users cannot see [5, 7], we look at two architectures that are equally invisible from the outside but lead users to ask different questions. This protocol-shaped variation is what makes the federated setting useful for extending folk-theory work beyond centralized platforms.

…

One might ask whether good in-product documentation removes the need for a venue like Reddit. We think not: much of what users explain is exactly what a single platform cannot narrate, a defederation decided by another server, or a block inherited from a stranger’s list, and even for its own actions, the threads do what documentation cannot, letting users compare protocols, correct each other in real time (our D4 subset), and read an event through people who have met it before. In-product explanations would give this cooperative reading a starting point, not replace it.

Our population is not federated-platform users in general but a self-selected slice: people engaged enough with fediverse governance to write about it in public. Unlike the users caught off guard by feed curation in Eslami et al. [7], ours had already chosen to explain moderation policy, so they are more technically literate than a median Mastodon or Bluesky user and their folk theories likely sit closer to the actual protocol mechanism. Our claims are thus about fediverse-engaged Redditors, not federated-platform users at large; native conversations on Bluesky, Mastodon, Discord and GitHub are likewise out of reach here. Reddit threads show public reasoning, not private experience and not the ground truth of what the platform did. Our claims concern how users describe federated moderation, not platform behavior we tested directly. The next step is to widen the sample with a held-out audit and to extend the codebook to a third protocol family (e.g., Nostr) so the two-case story can be tested against a setting we have not yet read.