it's pretty simple: 

publishing:

get the relay lists of people tagged in an event you are going to publish for a client, and publish them to them

fetching:

using the same relay lists, when searching for posts from a user, query the relays in this list

it's not hella scalable but while we are not yet in a stage where high accessibility of paid relays is a thing, and relays are relatively underdeveloped because relay devs get all the airtime and relay devs don't really have so much to easily demonstrate, relays should eventually have "peering" arrangements with other relays so they can fetch this stuff, and further, to propagate queries in order to build up maps of the location of content

yes there is a lot behind all that... me personally, i just gotta get my relay project to production level (in a few weeks) then i can start thinking about how to enable such features, and have relays do more of the legwork so clients aren't so much the weakest link in this chain 

 I'm pretty sure that I'm not the person needed to understand all this... 😅 

Thank you.  

 relays are like people's message servers

best way to be sure they see them is to send them to theirs

best way to get the messages they publish is to use theirs

and paid relays could make exceptions for requests ONLY for paying users notes with some limitations on how much of this they allow (to stop spam and viral post meltdowns) 

 That makes sense. 

One of these decades, it would be good to set up my own relay.  

 https://relay.tools is good, interface is getting pretty nice now, lets you set up filters for all kinds of things... i have one at https://mleku.nostr1.com that allows posts from a few of my favourite people, of whom many of them now have it on their list, makes it more sure that i will not miss their stuff 

 I think that's what I did with Stella's relay. Cool.