Data dictionary

A data dictionary is a structured reference that lists every table, column, or field in a database or dataset, along with its data type, allowed values, meaning, and any constraints, such as whether it can be null or must be unique. It is typically generated from the schema itself, then annotated with human-written descriptions.

A data dictionary documents structure: what exists and what shape it takes. It differs from a business glossary, which defines business concepts like "active customer" independent of any one table, and from a full data catalog, which adds search, ownership, lineage, and usage statistics on top of dictionary-level detail. Many catalogs generate and host a data dictionary automatically as one of their views, as part of broader metadata management.

Data dictionaries matter most at the point someone writes a query against unfamiliar data: they answer whether a status column uses active/inactive or 1/0, and whether a date is stored in UTC. Without one, analysts guess, and different teams encode the same fact differently. The main pitfall is staleness: a dictionary generated once during a data modeling project and never refreshed after schema evolution becomes actively wrong, which is worse than no documentation at all because it is trusted by default.