Medallion Architecture (2026) | Bronze, Silver & Gold Layers for the Data Lakehouse, Delta Lake, Iceberg, dbt & Fabric

What Is the Medallion Architecture?

7 entries

The medallion architecture is a data design pattern used to organize data logically in a lakehouse. Its goal is to incrementally and progressively improve the structure and quality of data as it flows through successive layers: bronze (raw), silver (validated) and gold (enriched). It is sometimes called a multi-hop architecture. Each hop adds validation, cleaning, modeling or aggregation so data becomes more reliable and better suited to business intelligence and machine learning. Databricks describes it as a recommended best practice rather than a requirement.

01What is the medallion lakehouse architecture? (Databricks docs)Reference description of the three layers, with an example and cost guidance. 02What is Medallion Architecture? (Databricks)Databricks overview of the pattern. 03What is the medallion lakehouse architecture? (Azure Databricks)Microsoft Learn version for Azure Databricks. 04Medallion lakehouse architecture in Microsoft FabricHow the pattern maps to OneLake lakehouses and warehouses in Fabric. 05What is a Data Lakehouse? (Databricks)Background on the lakehouse concept the pattern organizes. 06Lakehouse: A New Generation of Open Platforms (CIDR 2021)The research paper that set out the lakehouse architecture. 07What is a Data Lakehouse? (AWS)Vendor-neutral style explanation from AWS.

The Three Layers

4 entries

What each layer holds, who uses it and what happens there.

08Bronze: raw ingestionRaw, unvalidated data in its original form, appended incrementally from files, message buses and federated sources, mainly for data engineering, audit and reprocessing; keep validation minimal and store fields loosely typed to survive schema changes. 09Silver: cleaned and validatedCleansing, deduplication, schema enforcement, handling nulls and late data, and joins, producing at least one validated, non-aggregated representation of each record; modeling often starts here. 10Gold: curated and aggregatedBusiness-ready, dimensional and aggregated datasets for reporting, dashboards, ML and applications, often one gold layer per domain such as finance or HR. 11Delta LakeACID table storage underpinning the layers on Databricks.

Open Table Formats & Storage

8 entries

ACID tables on object storage.

12Delta LakeOpen table format with ACID transactions, time travel and schema enforcement. 13Delta Lake on GitHubSource code and releases. 14Apache IcebergOpen table format for huge analytic tables, supported by many engines. 15Apache HudiOpen table format with strong upsert and incremental processing. 16Microsoft Fabric OneLakeUnified lake for Fabric lakehouses and warehouses. 17Lakehouses in Microsoft FabricOverview of Fabric lakehouse items. 18Apache Iceberg on AWSAWS prescriptive guidance for Iceberg-based lakehouses. 19Apache Iceberg tables in BigQueryGoogle Cloud support for Iceberg tables.

Building the Pipelines

9 entries

Ingest, transform and orchestrate between layers.

20Spark Declarative Pipelines (Databricks)Declarative streaming and batch pipelines (formerly Delta Live Tables) that fit bronze-to-gold flows. 21Auto LoaderIncremental file ingestion into bronze tables. 22Structured Streaming (Apache Spark)Streaming engine used to read bronze and write silver incrementally. 23Apache Spark on GitHubSource code. 24Snowflake Dynamic TablesDeclarative incremental transformations between layers in Snowflake. 25dbt: how we structure projectsthe dbt staging, intermediate and marts layering, which maps naturally to silver and gold. 26Apache AirflowWorkflow orchestration. 27DagsterAsset-based data orchestration. 28PrefectWorkflow orchestration for data pipelines.

Data Quality, Contracts & Governance

10 entries

Trust each hop and control access.

29Unity CatalogGovernance for data and AI on Databricks; catalogs and schemas often separate layers. 30Unity Catalog (open source)Open catalog for data and AI assets. 31Data profiling and quality monitoring (Databricks)Monitor table quality over time. 32dbt data testsTest models between layers. 33Great Expectations (GX)Open-source data quality validation. 34SodaData quality checks and monitoring. 35Data Contract CLIOpen-source tool for data contracts. 36Data contractsSpecification and guidance for data contracts. 37OpenLineageOpen standard for lineage across pipelines. 38Apache PolarisOpen catalog for Iceberg tables.

Modeling & Serving the Gold Layer

8 entries

Dimensional models and fast queries.

39Dimensional Modeling Techniques (Kimball Group)Star schemas, facts and dimensions for gold tables. 40Data warehousing modeling techniques on the Databricks LakehouseDimensional and Data Vault modeling on a lakehouse. 41Data Mesh Principles (Martin Fowler)Domain-owned data products, an alternative or complement to central layers. 42TrinoDistributed SQL engine for querying lakehouse tables. 43DuckDBIn-process SQL engine for local analytics on Parquet and Iceberg. 44DremioLakehouse query platform built on Iceberg. 45StarburstTrino-based analytics platform. 46OnehouseManaged lakehouse built on Apache Hudi.

Books & Further Reading

2 entries

Go deeper.

47The Big Book of Data Engineering (Databricks)Free ebook with lakehouse pipeline patterns. 48Databricks Data Lakehouse platformVendor overview of the lakehouse platform.

10 entries

Neighbouring pages on this site.

49Data EngineeringData engineering hub. 50Data Lakehouse PlatformsLakehouse platforms. 51Data ArchitecturesData architecture patterns. 52Data Quality FrameworksFrameworks for data quality. 53Data Observability ToolsObserving data pipelines. 54Data Governance PlatformsGovernance platforms. 55Data Modeling ToolsTools for modeling data. 56DataOpsDataOps practices. 57Data Integration ToolsMoving data between systems. 58Cloud Data WarehousesCloud data warehouses.

Frequently Asked Questions

What is medallion architecture?

Medallion architecture is a data design pattern that organizes data in a lakehouse into bronze, silver and gold layers, progressively improving structure and quality as data moves from raw to validated to business-ready. It is also called a multi-hop architecture.

What is stored in the bronze, silver and gold layers?

Bronze holds raw, unvalidated data in its original form for audit and reprocessing. Silver holds cleaned, deduplicated and validated data, with at least one non-aggregated record-level representation. Gold holds business-ready dimensional models and aggregates for reporting, BI and machine learning.

Is the medallion architecture mandatory or only for Databricks?

No. It is a recommended best practice, not a requirement, and the layer names and counts vary by organization. Although popularized by Databricks, it applies on other platforms such as Microsoft Fabric, Snowflake, dbt-based stacks and open table formats like Iceberg and Hudi.

What are the trade-offs and criticisms?

More layers mean more storage, pipelines and latency, and the labels can hide unclear ownership. Teams must decide where business logic lives, avoid copying data needlessly and keep quality checks at each hop. Some prefer data mesh or Data Vault approaches, which can be combined with medallion layers.

How current is this page?

Links were checked in October 2026. The pattern descriptions follow Databricks and Microsoft documentation and are general guidance, not a standard; layer names and conventions vary by organization. Sites that block automated checks are kept and may need a manual click.

Medallion Architecture: Bronze, Silver & Gold Data Layers