Oddbean

Awesome! I think it’s going to be fun to think about what to do with scores in ranges like that. Your goal is to eliminate spam from your relays, which means prohibiting anyone with a score below some cutoff. But what is the cutoff that will best serve your purposes? Do you find out empirically? Pick a cutoff from first principles? Or use some other method? And what other data besides follows and mutes would you like to incorporate into the score? That’s the beauty of this method: it can synthesize data from multiple sources, and you can weight each source in whatever manner best suits your purpose. For example, I haven’t included zaps yet but might do that next.

it's interesting to me for the relays yes, but also as a nostr power user, I'm also interested in seeing new/different content from this context to switch up my feed.. I am going to load some more lists up and browse them with amethyst or gossip (both have easy list switching and outbox model). i did that with the top 1000, it was very similar to my follows feed :) I also will start publishing using a different relay list and maybe increase limits to 40k-50k and see how the clients perform when loading em.

Cool, makes sense. I def like the idea of using these scores for new/different content. Have you checked out the Content Discovery app on brainstorm? The system is set up to make recommendations in any context. I just need to add a button to export the lists. These lists should be much shorter but much more focused. Right now for example there are 4 users on the list for the “philosophy” context, generated bc people wrote nostrapedia articles on the topic of philosophy and then other trusted users liked those articles.

i'll check it out. i experimented around, and currently melting gossip and amethyst with a list that's 3500 (.298mb). It's chewing through it though, very interesting. This size was about as large as I could get before hitting a problem somewhere with ndk.Publish, or the relay itself (although I set the relay to allow very large lists, so it should have worked to go up to 10k events and ~.9mb.