A housing policy assistant must find exact program names while still supporting semantic questions. Which retrieval approach fits?

Hybrid search combining semantic vectors and keyword filters on program names is correct because housing policy assistants must handle both exact identifiers and paraphrased resident questions. Semantic vector search captures meaning when residents describe benefits in everyday language; keyword (or lexical) filters and exact matching catch official program names, acronyms, and code strings that embeddings may blur. Hybrid retrieval gives municipal housing answers precision on names without abandoning semantic flexibility.

Hybrid search combining semantic vectors and keyword filters on FAQ answers misplaces lexical precision because filtering on FAQ answer text does not enforce identity on official program names, so an exact title typed by a resident or caseworker can be missed while unrelated FAQs pass the filter. Hybrid search combining semantic vectors and keyword filters on document chunks applies lexical matching too broadly across chunk text rather than constraining it to the program-name field, so it cannot guarantee precision on titles, acronyms, and code strings that embeddings may blur. Hybrid search combining semantic vectors and keyword scoring on program names targets the right field but with the wrong control, because scoring only boosts name matches while still letting a strong semantic match outrank the exact title, whereas filters are needed to guarantee the exact program name surfaces.

Exam caveat: hybrid search = semantic + keyword/lexical. Stems that need exact program names and semantic questions point to hybrid with keyword filters on program names, not filters on FAQ answers or broad document chunks or scoring without filtering.

Operational check: implement hybrid retrieval with keyword boosts or filters on program-name fields plus vector search on policy text, evaluate on a mixed set of exact-name and paraphrased queries, tune fusion weights, and verify housing program lookup accuracy before production promotion.