Metadata and Master Data Management
What this domain tests
Can you make data findable and produce one trusted version of core entities? Expect questions on metadata types, lineage, glossary versus dictionary, MDM styles, matching and survivorship.
Core rules
- Metadata types: business (glossary, definitions, owners), technical (schemas, types, mappings, lineage), operational (run times, row counts, errors) and administrative (policies, retention, access).
- A business glossary defines terms in business language; a data dictionary documents structures. They complement each other and are not substitutes.
- Column-level lineage enables impact analysis before a change and root-cause tracing after a defect. Table-level lineage is insufficient for regulatory or break-fix work.
- Active metadata is consumed by tools for catalog search and policy enforcement; passive metadata sits in documents nobody reads.
- Master data domains: customer, product, supplier, employee, location. Reference data is the smaller set of allowed values such as country or status codes; do not confuse the two.
- MDM styles: registry (index only, data stays in sources), consolidation (copy for reporting), coexistence (central plus bidirectional sync), central/transactional (creates and owns the record).
- Entity resolution: standardize, block, match with deterministic and probabilistic rules, review exceptions, assign a global ID, apply survivorship rules for the golden record, and keep cross-references to source keys.
Worked example
CRM and ERP hold the same customer with different names and addresses. Reasoning: (1) Standardize and parse before matching. (2) Match deterministically on tax ID; fall back to fuzzy name plus postal code plus phone with a threshold. (3) Route borderline matches to a steward queue; never auto-merge on weak evidence. (4) Assign a party ID and store source IDs in a cross-reference table. (5) Survivorship: legal name from ERP, preferred name and email from CRM, most recently verified address. (6) Publish the golden record with lineage back to both sources.
Traps that cost marks
- Treating a business glossary as an IT data dictionary.
- Presenting table-level lineage as column-level impact analysis.
- Buying an MDM tool without matching rules, stewardship and survivorship decisions.
- Assuming the golden record requires one physical source of truth; it is often virtual.
- Merging on name alone; false merges are harder to reverse than duplicates.