Context layer · schema + semantics + metrics

Data landscape

Lookup registers meaning, not data. Every table here carries its structure (columns, types, keys, grain), its semantics (what each field means, synonyms, units, value vocabulary), canonical metrics, and the foreign-key graph. Fetchers pull rows from the real sources on demand. This is the one-time context that lets the agent map a question to the right tables and stitch them correctly.