New Tech

Microsoft Fabric medallion architecture: Bronze, Silver, Gold without the buzzword bingo

By Adrien Yvin ·

fabriclakehousemedalliononelakearchitecture

Medallion architecture is a filing system for data lake tables. In Fabric, OneLake makes the folders real — if your team agrees what each layer is for.

Layer responsibilities

Layer Holds Do not put here
Bronze Raw ingest, schema-on-read, full history Business-friendly column names
Silver Cleaned, conformed, deduped entities Report-specific KPIs
Gold Curated marts: star schemas, grain locked Raw JSON blobs

Rule: Power BI consumers read Gold (Direct Lake or Import from Gold shortcuts). Analysts explore Silver. Engineers own Bronze.

Fabric workspace layout

Typical pattern per domain (Sales, Finance):

DomainLakehouse/
  Files/Bronze/source_system_x/
  Tables/Silver/customer/
  Tables/Gold/fact_sales_daily/

Use shortcuts to a central Bronze lakehouse if multiple domains ingest the same source. Do not copy raw files per workspace without retention policy.

Naming that scales

  • Tables: {layer}_{entity} or schema-per-layer folders
  • Pipelines: pl_{domain}_{source}_to_{layer}
  • Semantic models: sm_{domain}_{subject} linked to Gold only

Document grain in the Gold table description field. Future you (and Copilot) will need it.

Power BI attachment

Prefer Direct Lake on Gold parquet when capacity supports it. Import from Gold only when DAX complexity or offline latency requirements demand it. Never bind executive dashboards directly to Bronze.

Migration in slices

Pick one subject area. Land Bronze for two weeks, promote one Silver entity, ship one Gold mart and one report. Compare refresh time and query cost before replatforming the whole ERP extract.

Designing medallion layout on OneLake for your tenant? Book a 30-minute call.