Ideally speaking, it shouldn't matter much which instance you pick, but that's one of the biggest miscommunications about how all this stuff works, speaking ideally rather than realistically.

Realistically instance choice matters both regarding technical stuff like how well it handles traffic and social stuff like whether folks are discussing anything that interests you to begin with and whether the instance's moderation style appeals to you. When all of this pans out, the tech should fade into the background, but as we've seen, it's early days yet in that regard.

It's like email. When you sign up for Gmail, you get all the Gmail features, use the Gmail website to access your email, and can send email to any other email, like Proton or Hotmail or whatever. But if Gmail goes down you can't read or send any email.

With Lemmy, you can see communities from any other server, like lemmy.ml or tchncs.de. But some servers might have a different interface, slightly different features, and if your instance goes down you won't be able to access log in unless you have an account somewhere else.

The extra cool thing is this extends beyond Lemmy. Some other social medias like Mastodon and Pixelfed communicate the same way that Lemmy does, they just look different. You can see Lemmy communities on Mastodon, and see Mastodon toots on Pixelfed.

A complete lack of documentation has made the whole process of converting to Lemmy a massive pain in the ass.

Another main problem is that it's not working how it's designed.

If an instance gets bogged down, or an instance is misconfigured, then data doesn't always replicate. Comments go missing from certain instances, etc.

The most basic explanation I can give is that: yes, instances can communicate with each other, but the don't share data automatically. A user from instance A must go interact with data from instance B by directly browsing to it via the correct URL string (instanceA.com/c/community@instanceB.com), and then interact with content in that community. Data from that specific community will then show up.

That's a large part of the reason that smaller instances have partial data from larger ones. Their users haven't interacted with enough communities outside their own instance.