Flash Cards · Retrieval

Flash Card: Freshness and Access Are Metadata

July 20, 2026 · 3 min read

Exam-style

One Knowledge Base serves both HR and Finance. Each team must only ever see its own documents, and superseded policy versions must never be cited even though they stay in the corpus for the record. What is the cheapest way to enforce both?

Reveal the answer

A. Attach department and effective-date metadata to each document and apply a metadata filter on every retrieval request

Access scope and freshness are attributes of the document, not properties of its meaning, so they belong in metadata and get enforced with a filter rather than being coaxed out of the embedding. Filtering is orthogonal to how retrieval works, so it stacks with hybrid search and reranking without interfering. Duplicating the corpus per department multiplies the ingestion bill and the drift, and both the embedding trick and the rerank filter are approximations that leak the first time a Finance document happens to mention HR.

Generative AI Development · part of The Exam Room

Q. Retrieval must respect access control and document freshness. What lever, cheaply?

A. Metadata filtering: restrict retrieval by attributes such as department, date, or document type, before or after the vector match. It is orthogonal to the retrieval method and stacks with hybrid search and reranking.

Why? Freshness and access questions are often metadata problems, not embedding problems.

These posts are LLM-aided. Backbone, original writing, and structure by Craig. Research and editing by Craig + LLM. Proof-reading by Craig.