Labelwatch

Bluesky doesn’t have a single moderation team deciding what you see. Instead, it has labelers — independent services that tag posts and accounts with metadata. Your app then decides what to do with those tags: show them, warn you, or hide them. It’s moderation as a protocol, not a policy.

That’s a genuinely new architecture. It also means nobody’s watching the watchers. There are currently 490 labelers on the network. Some do safety moderation. Some tag K-pop stans. Some are test instances someone spun up and forgot about. A few show behavioral patterns — rapid label-then-unlabel cycles, burst targeting of specific accounts, unexplained gaps in coverage — that are worth paying attention to whether they’re bugs or policy.

Labelwatch observes this layer. It doesn’t moderate anything. It ingests label events from every declared labeler on the network, runs anomaly detection, and surfaces behavioral patterns so you can decide what they mean. Think of it as an audit log for the governance layer itself.

Behavioral tags at a glance