@dmonad I think I found the place where client IDs gets reassigned in Yjs, so forget about my previous question 
I’d like to understand though how much overhead (at least on the wire) there would be if client IDs were arbitrary strings for example. I was trying to find more details about the document update binary encoding layout, but I couldn’t find any. So I was trying to figure it out from code reading thought the v2 format. Maybe asking you would be faster .
The application I’m working on doesn’t need real-time collaboration (at least for now), but we really need to make sure there’re no collisions of client IDs because published document updates would be cryptographically signed and immutable.
Now that I’m thinking about it, it’s probably close to what y-dat would do.