Medallion architecture on Microsoft Fabric, in practice

Bronze, silver and gold are easy to draw and easy to get wrong. What each layer should and shouldn't do when you build a data platform on Microsoft Fabric.

The medallion pattern organizes a data platform into three layers. On Microsoft Fabric it maps naturally onto Lakehouses, Warehouses and OneLake. The diagrams all look the same. The difference between a platform people trust and one they work around is in the rules for each layer.

Bronze: land it as it came

Bronze holds raw data exactly as the source sent it, stored as Delta tables in a Lakehouse.

  • Don't fix, filter or rename anything here.
  • Add load metadata: source system, load time and batch ID.
  • Keep history, so you can always rebuild the layers above.

If someone asks "what did the ERP actually send last Tuesday?", bronze answers it.

Silver: clean and conform

Silver is where data becomes consistent across systems.

  • Standardize types, units, codes and time zones.
  • Remove duplicates and apply data-quality rules, and record what failed rather than dropping it silently.
  • Conform shared entities such as customers, products and locations so every system's version maps to one ID.

Silver is for engineers and analysts, not for executive reports.

Gold: shaped for the business

Gold holds business-ready data products: facts and dimensions modeled for how the business asks questions. This is often where a Fabric Warehouse fits best, with SQL access for analysts.

  • Each gold table has a named owner and a clear definition.
  • Business logic, such as how revenue or available stock is calculated, lives here once and is reused everywhere.

The semantic model is part of the platform

Power BI semantic models on top of gold, using Direct Lake where it fits, give every report the same measures and definitions. Without a governed semantic model, each report author rebuilds the logic and the numbers drift apart again.

Governance from the first table

Ownership, quality checks between layers, lineage and role-based access aren't a phase-two project. Adding them later means retrofitting every pipeline. Start with a short list of rules that every new table must follow, and enforce them in review.

Dealing with this in your own system? We do this work for a living. The first call is free.

Book a call