Many of the functions provided by a Mastodon service is split into separate services in the AT Protocol. This means there are instances that just handle an end users data, instances that just handle indexing and streaming out the amalgamated end user data being streamed to the "relay", there are instances that are just filtering the stream from the indexing relay. so basically the various backend parts are modular with the AT Protocol rather than monolithic as is assumed by the ActivityPub protocol where separation is assumed to be only between the frontend and backend of the service.