What problem does it solve? Data platforms that grow organically end up with dashboards reading raw tables, multiple conflicting definitions of the same entity, and no way to reprocess data when logic bugs are found. This Skill provides a structured workflow for designing medallion (bronze/silver/gold) architectures where each layer has one job, promotion between layers is gated by executable tests, and reprocessing from raw data is a drilled capability rather than a hope. ## Core Features & Use Cases - Layer Contract Definition: Assigns each layer a single responsibility — bronze preserves raw data immutably, silver conforms and deduplicates entities, gold serves consumption-shaped marts — with written schema and quality contracts per layer. - Executable Promotion Gates: Encodes bronze-to-silver and silver-to-gold rules as blocking dbt tests, Delta constraints, or expectations suites, including quarantine tables with per-batch reconciliation (bronze = silver + quarantine). - Read-Path Enforcement and Reprocessing: Restricts layer access through warehouse grants rather than policy documents, and requires a staging rebuild drill to prove silver/gold can be reconstructed from bronze before it is needed. - Use Case: A scale-up with 400 dbt models and three conflicting definitions of orders uses this workflow to consolidate into 9 conformed silver entities, apply SCD2 only where analytically justified, and later replays 90 days of bronze in an afternoon when a currency bug is found in silver logic. ## Quick Start Use the medallion architecture skill to design the bronze, silver, and gold layers for our Databricks lakehouse, including promotion gates and ownership for each layer.
