Reference — 27 terms

Data Architecture Glossary

Short, precise definitions of the vocabulary of data architecture — written to be quotable, not exhaustive. Most terms link to a full essay that goes deeper.

Change Data Capture (CDC)

Change data capture streams every insert, update, and delete out of a database by reading its transaction log, instead of repeatedly querying for changes.

Conformed Dimension

A conformed dimension is a dimension shared with identical meaning across multiple fact tables, letting metrics from different processes be compared correctly.

Data Catalog

A data catalog is a searchable inventory of an organisation's data assets — tables, owners, definitions, lineage — the layer that makes data findable.

Data Contract

A data contract is an explicit, enforced agreement between a data producer and its consumers covering schema, semantics, quality, and change management.

Data Lake

A data lake is a large store of raw files in cheap object storage — any shape, structured later on read, with no transactions or enforced schema of its own.

Data Lakehouse

A lakehouse is a data lake with an open table format layered on top — one copy of data on object storage, with warehouse-grade guarantees, queryable by any engine.

Data Lineage

Data lineage is the traced path of data through systems — where each field came from, what transformed it, and what depends on it downstream.

Data Vault

Data vault is a warehouse modeling method splitting entities into hubs (keys), links (relationships), and satellites (history) to absorb change and support audit.

Data Warehouse

A data warehouse is a centralized, structured analytical database — modelled, governed, and optimized for reliable BI queries across an organisation.

Dimension Table

A dimension table stores descriptive context — the attributes you filter and group by, like customer, product, or date — surrounding the facts in a dimensional model.

ELT (Extract, Load, Transform)

ELT loads raw data into the warehouse first and transforms it there with SQL — the modern default, keeping raw history and putting logic where analysts work.

ETL (Extract, Transform, Load)

ETL extracts data from sources, transforms it before loading, and loads the finished result into the warehouse — transformation happens outside the destination.

Fact Table

A fact table stores the measurements of a business process — one row per event at a declared grain, carrying numeric measures and foreign keys to dimensions.

Factless Fact Table

A factless fact table records that an event or condition occurred — with no numeric measures — such as attendance, eligibility, or a promotion being active.

Grain

The grain is the business definition of what one row in a fact table represents — declared first, before dimensions and measures, in dimensional modeling.

Idempotent Pipeline

An idempotent pipeline produces the same correct result whether run once or five times — the property that makes retries, backfills, and replays safe.

Medallion Architecture

The medallion architecture organises a lakehouse into bronze (raw), silver (cleaned), and gold (consumable) layers of increasing refinement and trust.

Natural Key

A natural key is an identifier that comes from the business or source system itself — like an email, SKU, or account number — as opposed to a generated surrogate key.

OLAP (Online Analytical Processing)

OLAP is the analytical workload — large, read-heavy queries aggregating millions of rows — served by warehouses and columnar engines rather than operational databases.

OLTP (Online Transaction Processing)

OLTP systems run the business in real time — many small, concurrent reads and writes of individual records, on normalized schemas optimized for integrity.

Open Table Format

An open table format is a published metadata spec — Iceberg, Delta Lake, Hudi — that turns files in object storage into tables with ACID transactions and time travel.

Semantic Layer

A semantic layer is a shared definition layer between the warehouse and its consumers, where metrics like revenue are defined once and queried consistently everywhere.

Slowly Changing Dimension (SCD)

A slowly changing dimension is a dimension whose attributes change over time; SCD types define whether to overwrite history (Type 1) or preserve it (Type 2).

Snowflake Schema

A snowflake schema is a dimensional model whose dimensions are normalized into multiple related sub-tables instead of kept flat as in a star schema.

Star Schema

A star schema is a dimensional model with a central fact table joined directly to flat, denormalized dimension tables — the default shape for analytics.

Surrogate Key

A surrogate key is a meaningless, system-generated identifier used as a dimension table's primary key instead of the source system's natural key.

Vector Database

A vector database stores embeddings — numeric representations of meaning — and retrieves items by similarity, powering semantic search and RAG in AI systems.