pagination is broken for now and let's try to optimize it for a corpus with about 10K documents (and more after)