dbt-architecture

Medallion architecture organizes dbt models across three layers: Bronze (raw data staging), Silver (cleaned and validated data), and Gold (business-ready analytics). Structure your project with separate folders for each layer, apply naming conventions (stg_ for staging, int_ for intermediate, dim_ for dimensions, fct_ for facts), and enforce one-to-one staging models with no direct joins to sources.

Standard dbt naming conventions use prefixes to clarify model purpose: stg_ for staging models that transform raw data, int_ for intermediate business logic, dim_ for dimension tables, and fct_ for fact tables. Apply these consistently across your project, pair them with layer-based folder organization, and configure materialization policies per layer to maintain clarity and enforce dependency rules.

Layer separation in dbt uses Bronze/Silver/Gold folders with folder-level configurations and standardized materialization policies. Apply ref() over source() within and across layers, enforce one-to-one staging, and use dbt_project.yml to configure materialization and naming conventions per folder, creating a scalable production-grade project structure.

Staging models (prefixed stg_) are the single point of entry for raw source data in medallion architecture. Use one-to-one staging models to apply light transformations and validation before passing data to Silver layer models. Never join directly to sources; always reference staging models via ref(), which enforces dependency tracking and project modularity.

Define materialization policies in dbt_project.yml at the folder level for each layer: Bronze typically uses views or ephemeral models, Silver uses tables for persistence, and Gold uses tables or views depending on query patterns. Layer-level configuration ensures consistency, reduces errors, and makes materialization decisions explicit and maintainable.