3. BRING OVER SOME CONTENT WHEN FIRST FEDERATING COMMUNITIES
When an instance first federates a community, it should bring across at very least the last (let's say) 100 threads or the last (let's say) 7 day's worth of threads from that community (plus associated comments), whichever is greater.
This will prevent the following scenario: A user finds a community that's hosted on another instance, joins the community, but then finds no evidence of activity on their instance, because when an instance federates a community, it only starts pulling across posts from that moment in time. It makes it look like the community is dead, even if it isn't. While there may be a "Browse this community on the original instance" message, but that may well confuse people, and it doesn't mitigate the initial impression that the community has not posts.
Related to this - any pinned posts from a community should also be brought across by default, as these posts often contain information that a new user will find useful or that the moderators want all users to be aware of.
4. IDEALLY BRING OVER (OR ALLOW TO BE SEARCHED/SORTED) ALL CONTENT WHEN FEDERATING COMMUNITIES
The shelf life of posts in most communities is pretty short. If you subscribe to /news it probably doesn't matter if you can't see the top-ranked post from three years ago. But other communities curate content that has a much longer shelf life. A community like /askhistorians for instance, or /buyitforlife, where a user might want to search the archives for a great overview on the events leading up to the building of the Berlin Wall, or recommendations for the best compression socks. Allowing new users to search the complete history of a community, or sort posts by something like "most upvoted by all time", makes the community more useful.
So ideally if you subscribe to a community hosted on another instance on your home instance you should be able to browse/search/sort that community's entire archive.
I know you can click a link to browse a community on the original instance, but that can be confusing because suddenly you are now browsing on a site where you do not have an account.
Copying over the entire database for a community has storage/bandwidth implications (although I would argue data consumption issues are inherent to the fediverse model, which could lead to another discussion around the fediverse's scalability limitations). But perhaps there is a way for searches and sorts to interrogate the host instance of a community (which presumably has the most complete database) rather than the local instance.