For each nanopublication type there is a repository holding only nanopublications of that type, and for each publisher's key one holding only theirs. A query about, say, presentations runs against the presentations repository and never touches the rest, which is what makes views fast. A type's repository comes into existence when the first nanopublication of that type is published — until then, a query naming it finds nothing — so a new kind of nanopublication is introduced with an example first.