Medallion Architecture
A data lake design pattern that organises data into three progressive quality tiers — Bronze (raw), Silver (cleansed and enriched), and Gold (business-ready aggregates) — with quality gates between each layer.
What is Medallion Architecture?
Medallion Architecture (also called the Bronze/Silver/Gold pattern) is a data engineering design pattern that organises a data lake or lakehouse into three progressive quality tiers. Bronze holds raw data exactly as ingested — full fidelity, no transformations, append-only. Silver holds cleansed, deduplicated, and schema-validated records that have passed quality gates. Gold holds business-ready aggregations and domain models that power analytics dashboards, machine learning features, and operational APIs.
The value of medallion tiering is explicit quality semantics: consumers of Silver data know it has passed quality checks; consumers of Gold data know it has been curated for a specific business use case. This eliminates the 'I'm not sure if I should trust this' uncertainty that pervades ungoverned data lakes.
Medallion Architecture maps naturally to data contracts: Bronze-to-Silver promotion is gated by a contract that specifies required fields, acceptable null rates, and value range constraints. Silver-to-Gold promotion requires additional business logic validation. AiRAT implements medallion pipelines using Apache Spark on Databricks or EMR, with Delta Lake or Iceberg table formats providing ACID guarantees, schema enforcement, and time-travel for audit and rollback.
Examples in production
Financial transaction pipeline
Raw payment events land in Bronze with full schema. Silver validates account numbers, removes test transactions, and enriches with merchant category codes. Gold aggregates daily settlement totals and running exposure by counterparty — used by both treasury reporting and fraud monitoring.
Schema evolution across tiers
A new field added to upstream payment events appears in Bronze immediately. Silver promotion adds the field only after a data contract update is approved and quality checks pass. Gold tables are only updated when the finance team's business logic validation is complete.
Questions about Medallion Architecture
Does every data lake need medallion architecture?
Not necessarily — but any data lake serving multiple consumer teams benefits significantly from explicit quality tiers. If you have a single consumer and a simple ingestion pattern, medallion adds overhead without value. As soon as you have multiple teams consuming data with different quality expectations, tiering prevents the trust breakdown that turns lakes into swamps.
What is the difference between Bronze, Silver, and Gold?
Bronze: raw data, exactly as ingested, no transformations, append-only, full fidelity. Silver: cleansed, deduplicated, schema-validated data that has passed automated quality gates — trusted for analysis. Gold: business-curated aggregates and domain models optimised for specific consumers — dashboards, models, or operational APIs. Each tier has a data contract that defines what 'passing' means.
Running a system that uses Medallion Architecture?
We build production-grade data engineering systems for regulated enterprises.