Skip to content

Fedi v2

Referenced by: Social Media, The IndieWeb/Signature Blocks, Weird

A placeholder name for a theoretical new federated network that is client-centric, in contrast to the server-centric Fediverse

There are further discussions about actually implementing all this within the Weird community

Inspiration

  • A Plan for Social Media - Rethinking Federation
    • This article doesn't address many implementation details:
      • If the server is a relay, can content not be viewed anonymously?
      • How to handle storing large amounts of data on every client?
      • Don't you still need to associate with a server for people to direct their messages to?
  • Single-user Mastodon Instance is a Bad Idea
    • Focuses on the non-feasibility of self hosting, contributing to Federated Social Media not actually having all the upsides it should theoretically have by virtue of being Decentralized
  • The Commune community
  • Existing protocols:
    • Nostr
      • Currently suffers a culture problem by being associated with alt right and crypto users, making broad adoption more difficult
    • ATProto
      • Focused on a few large instances, to be run by large corporations. Still requires associating your identity with a server you don't own
  • A lot of these ideas are learned lessons from the usenet days

Identity

  • Federated Identity
  • Private and public keys anyone can create and store how they want
    • Fully free to create and store with no server dependencies
  • Profile information
    • Sent as a signed message through all the relays
    • How would you trust a username?
      • Petnames could be used to display human readable names via contacts or decentralized "naming hubs"
      • In most conversations online, you can trust their display name and add them as a contact as that display name
        • You only need to verify they are the same person you interacted with previously
        • You only need to trust people you want to send money to or otherwise "important identities"
        • For important identities, you can trust your contacts forming a chain of trust, or a authoritative naming hub
          • E.g. a white house ran naming hub that verifies the identities of the president and people of Congress
          • People typically wouldn't reach out to a naming hub, as it's not typically necessary
        • Contacts supercede naming hubs, so if a naming hub is breached, anyone I've previously added as a contact is still the source of truth
          • This only fails if the private key itself was breached
        • I'm just thepaperpilot, my display name. For most online communication, this is sufficient
          • My website can have a nameserver saying this publickey is the same as the site owner
          • If I write a paper at a scientific journal, they can say the author of x paper is my publickey
  • How to handle losing your private key
    • If you do have a naming hub you can verify with, they can say the identity has a new publickey
    • Contacts can "vouch" for a identity having a new publickey
    • Clients can decide to trust the new publickey based on contacts and naming hubs saying to
    • Also applies to stolen or compromised keys

Servers

  • Act as relays, merely storing messages and sending them to any clients or servers that have subscribed
  • May decide to publicly display messages its received
    • These servers are how discovery would work
    • Different servers may offer unique displays, filters, etc.
  • Users can send their content to any server - no authentication or account required, as the identity suffices
    • Even replies can work this way - no need to know from where a given message originated
  • Private servers could require some password when sending messages or subscribing to things
    • Useful for a school or other entity that wants an internal social network
  • Different ways to subscribe to a server's messages
    • All messages the relay hears about (new relays essentially subscribe like this to some existing relay)
    • All messages from a specific poster ID
    • Any replies to a message created with a specific poster ID
    • Shallow subscriptions, to lighten the load when subscribing to communities

Content

  • Protocol should dictate how to convey text, image, audio, video, and binary content
  • Protocol should include reacting to content with arbitrary text, including a URL
    • Upvotes and downvotes are implemented with this system
  • Each message contains fields for the poster's ID (public key) and a signature that verifies the content was made by that poster
    • That signature serves as an ID for the message itself
      • Anything can be replied to using the ID as the "parent" property in a new post
  • Edits are handled as replies with some flag to indicate it's updating the parent messages' content
    • Naturally, this reply would only be respected if it matches the same creator ID
    • Servers should replace the original message entirely with this one and indicate its an edited message
      • Some servers will inevitably keep a full history though
  • Groups/communities are just specially flagged messages
    • Posting to a community is just replying to that message
    • Subscribing to a community is just subscribing to that message
    • The original message creator effectively owns the group
  • IndieWeb pages could publish these messages as well, effectively serving as clients within the network
    • Perhaps use a bit to actually send those messages to other relays within the network

Moderation

  • In general, edits and delete requests are made by replying with a specially flagged message
  • Edit and deletion messages are ignored unless they have the correct public key and signature
    • Parent messages form a hierarchy of permission - if someone replies to your message, you can send a delete request for that message
    • Relay owners cannot fully delete messages, but can choose to stop relaying replies etc. of messages as the server owner wishes
  • Posts can be publicly reported with a specially flagged reply
    • How to make anonymous reports?
  • Users can send deletion or edit messages even without a matching public key, and clients (or relays) can choose to respect those messages if that public key is whitelisted as a moderator
    • Messages (and by extension, groups) can have replies granting or removing permission to other public IDs at that hierarchy level
    • People can setup accounts with their desired heuristic for sending delete messages, such as looking at public reports or analyzing the content with AI
      • This way clients can effectively customize their preferred moderation
  • Clients can also choose to add additional rules for hiding content, such as any reports by followed users
  • Perhaps delete messages pull double duty as public reports in and of themselves?

Problems to solve

  • No anonymity
    • All upvotes, downvotes, etc. are linked to your public key
    • Perhaps a client could generate new keypairs for every action for anonymity, but then it'd be hard to determine if such an account and action was a genuine user or a bot
  • Servers could probably determine the identity of clients sending their messages to them
    • A client that only ever sends messages with a specific public key is unlikely to be a server
    • A client that doesn't subscribe to all messages is unlikely to be a server
    • Perhaps clients and servers can be identified as such, and subscribing to new messages is something you only do to servers, not clients
  • Illegal material will likely be placed on the hard drive at least temporarily
    • Messages will be downloaded and, even if you follow a moderator bot that looks for illegal material, there's likely to be a delay between receiving the initial message and receiving the bots delete message
  • You have to download all spam messages
    • For redundancy, you'd likely subscribe to multiple relay servers
    • You cannot trust several relay servers to have identical rules on not relaying messages that don't pass whatever moderation heuristic
    • Therefore, the filtering out of spam has to be done by the client, after downloading it all

Last updated: