Medallion architecture organizes lake/lakehouse data into quality layers:
- Bronze: raw ingested data, little change, history preserved
- Silver: cleaned, conformed, joined, deduplicated
- Gold: business-level marts and aggregates for consumers
sources -> bronze (raw) -> silver (clean) -> gold (marts / metrics)
Tiny example
- Bronze:
bronze.stripe_paymentsJSON as received - Silver: typed
paymentswith standardized currency and ids - Gold:
finance.daily_revenueby region
Why it helps
You can reprocess silver/gold from bronze when logic changes. Consumers only touch gold. Quality expectations get stricter by layer.
Interview tip: Map medallion to your stack (for example bronze files, silver dbt staging, gold dbt marts). Mention grain and ownership per layer.