File-native search over the canonical corpus files. Exact and proximity searches are supported first; public regex is deliberately avoided.