Right now, resolvers for various "count" queries use type int32, which means their max value is a little more than 2 billion. A single user or subject might not have 2 billion studysets, but the total counts for a bunch of different cusor-paginated queries also use int32, so in the future that might become a problem for other use cases.