Makes sense, and agreed. Still… it's a difficult ecosystem, with clients and users doing things in very different ways.
For example, some clients encrypt all tags in the user mute list; some people may argue that even publicly muting someone shouldn't influence the social graph; and kind 1984 is what should really count as negative signal.
But then again, NIP-56 specifically asks relays not to perform automatic moderation based on reports... But (but after but after but), one could argue that aggregated results based on WoT analysis (where one trusted npub report = one vote) are a bit harder to game.
Lots to think about, and no easy answers here.
My current take is that people use all these tools in such unique ways that it may be better to keep a private ownertrust list for pubkeys in the relay itself…sort of like how PGP does it.
I.e., I'm leaning toward using the followers list to build the trust graph, ignoring all other kinds, and allowing users to privately set how much they trust a given pubkey.
For example:
1. I follow a certain npub but don’t trust it at all because I know it follows bots
2. I marginally trust an npub, so I’ll use its follow list for the WoT score only if three other users do as well (default case)
3. I fully trust a key, so I’ll allow all of its followers to write to my relay
My dilemma is that a PGP-like WoT is too complex, and I don’t want to build yet another model that only security folks can use. But then again, I’m struggling quite a bit to simplify it. It feels technically correct… just hard to use.
Makes sense, and agreed. Still… it's a difficult ecosystem, with clients and users doing things in very different ways.
For example, some clients encrypt all tags in the user mute list; some people may argue that even publicly muting someone shouldn't influence the social graph; and kind 1984 is what should really count as negative signal.
But then again, NIP-56 specifically asks relays not to perform automatic moderation based on reports... But (but after but after but), one could argue that aggregated results based on WoT analysis (where one trusted npub report = one vote) are a bit harder to game.
Lots to think about, and no easy answers here.
My current take is that people use all these tools in such unique ways that it may be better to keep a private ownertrust list for pubkeys in the relay itself…sort of like how PGP does it.
I.e., I'm leaning toward using the followers list to build the trust graph, ignoring all other kinds, and allowing users to privately set how much they trust a given npub.
For example:
1. I follow a certain npub but don’t trust it at all because I know it follows bots, ignore it when computing WoT scores.
2. I marginally trust an npub, so I’ll use its follow list for the WoT score. If 3 users follow a npub, I'll trust it as well (default case - What we do in HAVEN right now)
3. I fully trust a key, so I’ll allow all of its followers to write to my relay
My dilemma is that a PGP-like WoT is too complex, and I don’t want to build yet another model that only security folks can use. But then again, I’m struggling quite a bit to simplify it. It feels technically correct… just… hard to use.
Makes sense, and agreed. Still… it's a difficult ecosystem, with clients and users doing things in very different ways.
For example, some clients encrypt all tags in the user mute list; some people may argue that even publicly muting someone shouldn't influence the social graph; and kind 1984 is what should really count as negative signal.
But then again, NIP-56 specifically asks relays not to perform automatic moderation based on reports... But (but after but after but), one could argue that aggregated results based on WoT analysis (where one trusted npub report = one vote) are a bit harder to game.
Lots to think about, and no easy answers here.
My current take is that people use all these tools in such unique ways that it may be better to keep a private ownertrust list for keys in the relay itself…sort of like how PGP does it.
I.e., I'm leaning toward using the followers list to build the trust graph, ignoring all other kinds, and allowing users to privately set how much they trust a given npub.
For example:
1. I follow a certain npub but don’t trust it at all because I know it follows bots, ignore it when computing WoT scores.
2. I marginally trust an npub, so I’ll use its follow list for the WoT score. If 3 users follow a npub, I'll trust it as well (default case - What we do in HAVEN right now)
3. I fully trust a key, so I’ll allow all of its followers to write to my relay
My dilemma is that a PGP-like WoT is too complex, and I don’t want to build yet another model that only security folks can use. But then again, I’m struggling quite a bit to simplify it. It feels technically correct… just… hard to use.
Showing page 1 of
1 pages
