So far, we create a dataset given a single ranker r and metric m.
We want to find a dataset that covers a set of rankers and metrics such that given a query and its expanded version, we can say that the expanded version is agnostic to the underlying rankers and metrics