Metadata
Metadata
Metadata is the domain most often under-estimated by candidates: it looks descriptive but is the mechanism by which data becomes findable, trusted, and change-safe. The exam tests metadata types, repositories and catalogs, standards, lineage and impact analysis, and the operating processes that keep metadata current.
Core rules and concepts
- Metadata is data about data: structure, meaning, ownership, quality, usage, and movement. DAMA separates business metadata (definitions, glossary terms, business rules, ownership), technical metadata (schemas, data types, keys, ETL mappings, models), and operational metadata (row counts, load times, job status, error logs); process, audit, and provenance metadata are often added.
- A business glossary holds agreed business terms and definitions; a data dictionary holds technical structure. They must be linked, or the glossary remains aspirational.
- ISO/IEC 11179 provides principles for metadata registries: data elements with names, definitions, value domains, and classification. Other standards include Dublin Core for documents, OMG CWM for warehouse and metadata exchange, and W3C vocabularies.
- Lineage records where data came from and how it changed. Impact analysis uses lineage in reverse: if a source column changes, which reports, models, or APIs break?
- Metadata management activities: define strategy, understand requirements, define architecture, create and update, integrate, distribute, query and report, maintain, govern, and measure. Metadata itself needs owners, quality rules, and a lifecycle.
- Metadata architecture patterns: centralized repository, distributed with a catalog indexing sources, and hybrid; automated harvesting via connectors and APIs is preferred to manual entry.
Worked example
A warehouse column CUST_STAT_CD is being retired. Business metadata tells you it means customer lifecycle status and that the glossary definition was approved by Sales Operations. Technical metadata shows VARCHAR(2), not null, sourced from CRM.customer_status. Operational metadata shows it loads nightly at 02:00 and 3.1 million rows were processed last run. Lineage shows it feeds two churn models and a regulatory report. Impact analysis therefore flags the churn models and the report as affected. Without linked business and technical metadata, a steward would see only a column name and could approve a change that silently breaks a regulatory output.
Traps that cost marks
- Defining metadata as technical only; operational and business metadata carry equal exam weight.
- Treating a data catalog as a one-off project rather than a governed, continuously harvested asset.
- Confusing lineage with ETL job documentation; lineage traces data values through transformations.
- Assuming a glossary term is a definition of a physical column; the link must be explicit.
- Ignoring metadata quality: stale definitions and broken lineage cause wrong impact decisions.