been thinking about this a lot and i think that an architecture detail that would really help guide client dev is this:

instead of duplicating effort and writing a custom relay cache

use an internal relay, which does this exact kind of broadcasting for you, including NIP-46 bunker auth for outbound relaying

then you centralise the codebase for relaying in one place and we stop wasting time writing two versions of it

i have run into a really nasty, sneaky bug where i didn't know that my relay wasn't returning events correctly, because clients always cache them

with an internal relay in clients, and a simple switch to disable it and use your NIP-65 list, relay development becomes a lot more reasonable

i'm going to be advising front end devs on forking one of y'all's client codebases or three and modifying them to run an internal relay from the one i'm building is going to be key

i know how to build go code to run as internal servers for mobile platforms already (keybase has been doing this for yeras) and baking in NIP-46 is probably essential

with relays working in this way, the only thing that we need is some guidance on how to set up reverse proxies on subdomains with wireguard tunnels then your chat friends can literally send their DMs straight to your inbox if you are online, ie, in an interactive session