What is a medallion architecture (bronze, silver, gold)?

It's a way to organize data by quality and readiness:

- Bronze (raw): data exactly as received (API dumps, CSVs, CDC feeds), append-only with ingestion metadata (source, load time). Lets you reprocess anything later. - Silver (cleaned and conformed): deduplicated, typed, validated and joined. Consistent keys and names, bad records quarantined. The "single source of truth" at entity level (customers, orders). - Gold (business-ready): aggregated, modeled for consumption. Star schemas for Power BI, feature tables for ML, KPI tables.

For a small project it's a naming convention more than infrastructure. Even three schemas (raw, clean, mart) in one Postgres database give you most of the benefit: you can always rebuild downstream layers from raw, and business logic lives in one place. Tools like dbt fit this pattern well.