- UserA starts a feed at dat://UserA, and publishes stuff in Activity Streams format
- UserB follows dat://UserA
- UserB comments on UserA feed: it creates dat://CommentsFromUserBonUserA, and publishes a comment in Activity Streams format there
- UserB pings UserA telling them about dat://CommentsFromUserBonUserA: UserA is naturally interested, so they just subscribe to this feed, and republishes stuff into dat://UserA
- UserC subscribes to dat://UserA. They get UserA's posts, along with UserB's comments, without every subscribing to dat://UserB. All activity about UserA is in one place, completely under UserA control, with no dependency on a centralized service
- Same can be done for mentions of UserA
This obviously assumes pinging is easy, which it isn't. Perhaps some polling of a DHT ? It also doesn't solve the issue of hashtags... but it makes dependency on centralized service less important
Web Crawling is a reliable federated system that doesn't impede user freedoms. The crawlers are so fungible that anything centralized about them doesn't really concern me.