However, it could be better: first we have been converging on a single Rust crypto impl (at last) - see the crypto section of https://matrix.org/blog/2023/12/25/the-matrix-holiday-update.... Then we’re shifting to TOFU (so unsigned devices never get encrypted for). Then, finally, we’re shifting to client-constrained group membership (unless MIMI gets there first, and then we may switch horses to their MLS-based solution).
edit: the other issues raised by the nebuchadnezzar paper were impl bugs in the 1st gen crypto impls, which were fixed at the time. also, the premise that you can’t add encryption posthoc to a protocol is flawed, imo. Matrix may not have nailed it, but as long as you consider the crypto as if it’s a new protocol and have the agility to evolve to it, everyone wins.
edit 2: there is no centralised db of who talks to who in Matrix. individual servers can see who shares a conversation which their server participates in, but this is very hard to avoid given you could always figure it out via traffic analysis (unless you go down a Nym style mixnet solution).