Oddbean new post about login | logout The majority of clients now (although not the ones with most users) are using the "outbox model" (also misleadingly known as "gossip model") which mean they will connect to any rogue relays they think they can find notes from -- so if I'm publishing to wss://pyramid.fiatjaf.com/ and you follow me then your client will connect to that even if that is not in your "relay list" (even the idea of a static "relay list" is an antipattern in my view). This means anyone can actually run a relay and that relay will be useful and used. How to find out which relays each person is publishing to is the hardest part and not straightforward, but in practice using NIP-65, NIP-05 relay lists and relay hints in events works well enough, and there are many client libraries that do it automatically. You can reverse this pattern for other use cases, like mentioning people and replying to their posts in a way that they will see your stuff even if they don't follow you (NIP-65 specifies this) and do other variations for other use cases.
Incidentally that also solves the spam problem: for following people you don't have to worry as you'll only fetch stuff from people you follow. For spam on "global" timelines and in reply to other people's posts, you can easily limit your read to relays of a certain kind that provide some form of anti-spam control, and each relay can implement different techniques, some may use Lightning, others may use closed whitelists, others AI-filtering and so on.
Nostr clients that utilize indexers for content display - like Primal - are even better at surfacing content from “long tail” relays. Anyone can run a low-powered obscure relay. As long as the relay is included in the NIP-65 or NIP-05 relay list of a single nostr user, that relay will be indexed. You can totally run a relay on a raspberry pi and have your content accessible by millions of users. Now, an argument can be made that this model introduces centralization at the indexer level, which is true. The good news is that anyone can run an indexer, and we are already seeing many indexers being stood up. I envision hundreds of indexing services with all kinds of configurations in terms of relay lists, event types and date-based pruning. The most encouraging part is that these different decentralization techniques are sprouting organically by people acting in their own self interest and without any central coordination. The core nostr protocol is solid and makes it easy to build these types of solutions. Therefore nostr decentralization is much better than it looks at a first glance. @pkt @niftynei