Architecture

Medallion Architecture

A layered data quality pattern — Bronze (raw), Silver (cleansed), Gold (business-ready) — with each layer stored on object storage.

8 connections 3 resources

Summary

What it is

A layered data quality pattern — Bronze (raw), Silver (cleansed), Gold (business-ready) — with each layer stored on object storage.

Where it fits

Medallion is the most widely adopted data quality pattern within lakehouses. It organizes S3 data into progressive quality tiers, giving each tier a clear contract and making it safe for different consumers to read at different quality levels.

Misconceptions / Traps
  • Three layers is a convention, not a rule. Some organizations use two layers; others add more. The pattern is about progressive refinement, not a fixed number of tiers.
  • Medallion does not solve the small files problem — it can worsen it. Each layer transformation may produce many small output files, especially with streaming Silver→Gold pipelines.
Key Connections
  • is_a Lakehouse Architecture — a specialization of the lakehouse pattern
  • constrained_by Legacy Ingestion Bottlenecks, Small Files Problem
  • AWS S3 used_by Medallion Architecture — each layer resides on S3
  • Apache Spark, Apache Flink used_by Medallion Architecture — compute engines for tier transformations
  • scoped_to Lakehouse, Data Lake

Definition

What it is

A layered data quality pattern that organizes data into three tiers — Bronze (raw), Silver (cleansed/conformed), Gold (aggregated/business-ready) — with each layer stored on object storage.

Why it exists

Raw data arriving in S3 is messy, inconsistent, and not query-ready. The Medallion pattern provides a structured progression from raw ingestion to business-quality data, with clear contracts at each tier.

Primary use cases

Data lake quality management, incremental data refinement, separation of raw ingestion from analytics-ready data.

Recent developments

Latest signals
  • Three-layer canonical structure: Bronze (raw) → Silver (validated) → Gold (enriched). Bronze contains raw unvalidated data in original source formats; Silver contains validated/cleansed/deduplicated/normalized data with quality enhancements; Gold contains business-level aggregates + features ready for analytics and ML. Per Databricks — What is Medallion Architecture.
  • Official Microsoft Fabric reference implementation. Microsoft documents an end-to-end medallion implementation on Fabric / OneLake — the canonical 2026 deployment pattern for Azure-side lakehouses. Per Microsoft Learn — Medallion in Fabric.
  • ACID guarantees across layers via Iceberg/Delta/Hudi. The architecture guarantees atomicity, consistency, isolation, and durability as data passes through validation + transformation layers — implemented via the table-format transactional layer (Iceberg, Delta Lake, or Hudi) at each medallion tier. Per Tacnode — Medallion Architecture.
  • 2026 lakehouse-design default. Per the 2026 medallion guide round-up, the bronze/silver/gold pattern is now the default starting point for almost all greenfield lakehouse deployments — incremental refinement + clear team-handoff boundaries + composable downstream consumers. Per Dataforest — Medallion Architecture 2026 Guide.
  • Reuse + governance benefits at scale. Different teams get clear entry points into shared data; security/governance policies can be applied per-tier (broad read on Silver, restricted on Bronze raw PII); pipeline reuse increases. Per Azure Databricks — Medallion architecture.

Connections 8

Outbound 5
Inbound 3

Resources 3