Medallion Architecture

Medallion architecture organizes data through bronze, silver, and gold layers. Bronze retains source-oriented input, silver makes records usable through validation and cleaning, and gold shapes results for business consumption. These names communicate refinement responsibilities rather than a mandatory number of physical systems.

For orders, bronze can preserve received deliveries, silver can parse amounts and distinguish redeliveries from updates, and gold can expose daily totals. State the row grain and correction policy at each boundary. A table called silver is not automatically valid, and gold does not mean every dataset must be aggregated.

A four-layer design can separate cleaned records from shared business models before serving. Its boundaries do not map uniquely to three medallion labels. Choose boundaries for consumers, ownership, and operating requirements, and account for extra copies, latency, and rebuild work.

Layer names do not make a whole pipeline atomic. Transaction scope depends on the storage, engine, and publication procedure. If a later build fails, define what consumers can still read and how corrected input will be replayed.

Reference: Microsoft: medallion architecture. See Designing End-to-End Data Architecture for the worked example.


Discover more from Insightful Data Lab

Subscribe to get the latest posts sent to your email.

Similar Posts

Questions, corrections, or additional insights?

This site uses Akismet to reduce spam. Learn how your comment data is processed.