You actually described my expectation in that last paragraph. I fully expected to have to invent a curated list of endpoints. What I instead found was that even given an endpoint (like beehaw.org) there is no way to discover what is on that instance. Which is a big issue. I've discussed this a bit (albeit extremely slowly) with the folks in the activitypub irc, and pretty much got it confirmed. If you want to develop a service that is intent on delivering any form of public content at all, supporting activitypub, then you have no protocol based way of communicating this content to anyone. So going by that the lemmy implementation is actually "correct", in as much as the protocol simply fails to provide guidance for this critical step.