Enterprise search is often envisioned as challenges involving a corpus of business documents. This may be fundamentally misleading. I suspect it’s better to model it as multiple overlapping corpora all mixed together. Almost everything in IR is about sets anyway.