The medallion architecture (Bronze, Silver, Gold) is a data layering convention for progressive refinement in lakehouses — not a Databricks-exclusive feature. All three tiers map cleanly onto open table formats: Apache Iceberg, Delta Lake, and Apache Hudi. Bronze serves as an append-only raw landing zone for replay and audit; Silver handles deduplication, upserts, and SCD logic via MERGE INTO; Gold delivers business-ready aggregates optimized for reads. Key trade-offs include copy-on-write vs. merge-on-read for update-heavy Silver tables, and mandatory table maintenance (compaction, snapshot expiry) at Bronze and Silver. The post also critically examines when to skip Bronze entirely — it's only worth keeping if you genuinely need an immutable replay point — and warns against reflexively applying three layers to every dataset.
