I'm on the team that implemented this, so I'm happy to answer questions. I'll give a brief technical overview.
It's broadly a system for publishing metadata on posts called "Labels". Application clients specify which labeling services they want to use in request headers. Those labels get attached to the responses, where they can then be interpreted by client.
This is an open system. Clients can choose which labelers they use, and while the Bluesky client hardcodes the Bluesky moderation another client can choose a different primary Labeler. Users can then add their community labelers, which I describe below. We aim to do the majority of our moderation at that layer. There are also "infrastructure takedowns" for illegal content and network abuse, which we execute at the services layer (ie the relay).
Within the app this looks like special accounts you can subscribe to in order to get additional filters. The labels can be neutral or negative, which means they can also essentially function as user badges. Over time we'll continue to extend the system to support richer metadata and more behaviors, which can be used to do things like community notes or label-driven reply gates.
It's broadly a system for publishing metadata on posts called "Labels". Application clients specify which labeling services they want to use in request headers. Those labels get attached to the responses, where they can then be interpreted by client.
This is an open system. Clients can choose which labelers they use, and while the Bluesky client hardcodes the Bluesky moderation another client can choose a different primary Labeler. Users can then add their community labelers, which I describe below. We aim to do the majority of our moderation at that layer. There are also "infrastructure takedowns" for illegal content and network abuse, which we execute at the services layer (ie the relay).
Within the app this looks like special accounts you can subscribe to in order to get additional filters. The labels can be neutral or negative, which means they can also essentially function as user badges. Over time we'll continue to extend the system to support richer metadata and more behaviors, which can be used to do things like community notes or label-driven reply gates.