If the scale of the application is small enough, you might not need CRDT. One GenServer per document can do the job of serializing concurrent editing. You can keep the client side dumb and rely on the full loop of client → LiveView → GenServer for the doc → LiveView → client for every mutation.
