Im 100% for going back to "having more than 5 websites" and orgs hosting their own forums and git servers and whatnot, but I also absolutely cant handle checking 50 sites and I refuse to handle a million notifications coming to my email.
The next leg of my work after we get this arbitrary-binary-with-metadata format thing working will be to start building the p2p client, and im going to start designing that as a personal read/write webcrawler to pull in notifs and b able to crosspost and whatnot. Remember trillian and pidgin where you could log into AIM and MSN and IRC and etc. In the same client? That was a good idea, lets do more of that.
@gwil i am still nailing down versioning, but i am thinking basically a typical version merkle treee orthogonal to a discretized merkle DAG.... so at creation, subgraph ID is the root node of a hashed DAG (the discretization part being the construction of a subgraph with an artificially designated root node that is the parent of all nodes in the graph) similar to a git repository, but rather than the tree just being the filesystem, you have an arbitrary RDF-like (gulp, bear with me) triple graph. The important thing being acknowledging that binary data has both a treelike representation to do bittorrent v2-like things, and also a graphlike metadata representation that can contain eg. filewide metadata like EXIF or id3, but also byte-ranged subfile metadata like eg. for an hdf5 file.
POINT BEING that what we want to get to is something where
a) with no additional knowledge about the binary string, it would be shared very similarly to bittorrent (at the protocol level of a hashtree of pieces being exchanged between peres)
b) with basic file extraction metadata linked to a schema one could do something like have a music collection indexable like @ jonny:music:albums:albumname etc. think sparql-able attributes except not being afraid of relative identifiers and path traversal
c) with our own schema that we make over time through conversation and use, we might also have some term a la an RDF controlled vocabulary term for something we talk about all the time (except without the pretense of universality, like how #ThisHashtag might be @ jonny:ThisHashtag identified as "my version of that thing") that allows us to relate a song and some sentence in an essay, to be able to refer to things in a common space despite differing representations, levels of description and detail, etc.
speaking telegraphically rn bc need to go to bed but yes. want to be able to refer to things, want to be able to represent things without needing to be "correct" about that representation except as far as it is useful to me, etc.
They're separate things. Federation is how posts travel between servers. Algorithms (in this context) is how they're displayed when they get there. 'Chronological ordering of posts by people I follow' is an algorithm. 'Everything on my server containing #ThisHashtag ' is another one. What BlueSky proposes is to make sorting and filtering algorithms a third-party service users can choose between at their end, and toggle on and off.