For Likes and Replies, the way the Nostr protocol is designed now you have to pull down every record created for every note being displayed -- there's no aggregates stored. That's untenable once Nostr has millions of active users. Can you imagine the client having to retrieve tens of thousands of records every time it loaded a page of notes just to show the like/reply counts? Even if the protocol does develop a solution, querying a single relay will only give you the counts from that relay. I'm aggregating all the top relay data (27 atm) so the numbers are pretty accurate.