What you will be able to do
- Describe what happens in the Bronze, Silver and Gold layers and who each layer is for
- Place 3NF and Data Vault models in the Silver layer and star schemas and data marts in the Gold layer
- Explain why Gold models are denormalized and read-optimized, while Silver models are built for integration and change
Key concept
Medallion architecture — A design pattern that splits lakehouse data into Bronze (raw), Silver (validated) and Gold (enriched) layers. Data gets better at each hop, and each layer suits a different kind of data model.
1.Bronze, Silver, Gold: layers of increasing data quality
Before you can say where a star schema or a Data Vault belongs, you need the structure it fits into. The medallion architecture is a data design pattern that organizes lakehouse data into three layers. The layer names describe data quality: bronze is raw, silver is validated and gold is enriched. Data flows Bronze ⇒ Silver ⇒ Gold, and each hop improves its structure and quality. You may also see it called a multi-hop architecture. Databricks recommends the pattern, but it is a best practice and not a requirement.
| Layer | What happens in this layer | Intended users |
|---|---|---|
| Bronze | Raw data ingestion | Data engineers, data operations, compliance and audit teams |
| Silver | Data cleaning and validation | Data engineers, data analysts who need detailed records, data scientists |
| Gold | Dimensional modeling and aggregation | Business analysts and BI developers, data scientists and ML engineers, executives, operational teams |
Bronze applies no analytical model at all. It keeps the raw state of each source in its original format and grows by incremental appends. It keeps all history so the data can be reprocessed and audited. The Databricks blog describes Bronze table structures as matching the source system's tables as-is, plus metadata columns such as load date/time or process ID. Bronze is not meant for analysts. It feeds the workloads that build Silver tables. The modeling decisions this subdomain covers therefore start in Silver and finish in Gold.
Checkpoint 1 of 5· Check yourself
A new analyst asks for read access to the Bronze tables to build a revenue dashboard. Based on Databricks guidance, what is the Bronze layer intended for?
Bronze keeps raw, unvalidated data in its source format. Its consumers are the pipelines that build Silver, not analysts.
“Is intended for consumption by workloads that enrich data for silver tables, not for access by analysts and data scientists.”Source: docs.databricks.com
Checkpoint 2 of 5· Exam question
A data analytics team is building the reporting layer for an AI/BI dashboard on top of already-cleansed Silver tables. Analysts need dashboard queries to hit a single join hop per dimension for fast interactive filtering. Which modeling approach best matches the Gold layer of the Medallion Architecture for this goal, and why?
Correct answer: A — A Kimball-style star schema, because it denormalizes each dimension into one flat table joined directly to the fact table, matching the Gold layer's read-optimized, business-ready design.
- A. A Kimball star schema keeps dimension tables denormalized and directly connected to the fact table, so a dashboard query only needs one join per dimension. This directly matches the Gold layer's purpose of serving de-normalized, read-optimized, business-ready aggregates.
- B. Normalizing dimensions into subdimensions is the snowflake schema approach, which trades query simplicity for storage savings and requires more joins per report. That extra join overhead works against the Gold layer's goal of fast, low-join dashboard reads.
- C. Hub, link, and satellite tables describe data vault modeling, which is built for auditability and tracking every historical source change, not for serving fast single-hop dashboard joins. Data vault structures are typically transformed further before reaching a reporting layer.
- D. Fully normalized third-normal-form modeling maximizes referential integrity but multiplies the number of joins a dashboard query needs, which is the opposite of the low-join, read-optimized design the Gold layer is meant to provide.
Sources1
2.Silver: where the enterprise data warehouse is modeled
Silver is where data cleansing, deduplication and normalization happen. Typical Silver operations are schema enforcement, handling of nulls, deduplication, resolving late-arriving data, type casting and joins. The Databricks data warehousing page gives this layer a specific modeling job: the data warehouse is modeled in Silver, and it feeds specialized data marts in Gold. Here you integrate data from separate sources in line with your business processes. That warehouse is described as schema-on-write, atomic and optimized for change, so it can be adjusted quickly when business processes change. Defining primary and foreign key constraints helps end users see how tables relate in Unity Catalog.
The Databricks medallion blog says the same thing in different words. Silver gives an "Enterprise view" of key business entities, such as master customers, stores, non-duplicated transactions and cross-reference tables. It has more 3NF-like models, and Data Vault-like models fit here because they are write-performant. That suits how Silver is loaded. Lakehouses usually follow ELT, so only "just-enough" transformation happens on the way into Silver, where speed and agility come first. The heavy, project-specific business rules are applied later, between Silver and Gold.
Checkpoint 3 of 5· Check yourself
Why do Databricks sources describe Data Vault-like models as a good fit for the Silver layer?
Silver loading puts speed and agility first and applies only minimal transformation, so write-performant models fit. Minimizing joins describes Gold, and keeping data as-is describes Bronze.
“Data Vault-like, write-performant data models can be used in this layer.”Source: www.databricks.com
3.Gold: dimensional models, data marts and aggregates
The Gold row of the layer table says "Dimensional modeling and aggregation", and both halves count. The Databricks warehousing page calls Gold the presentation layer, which holds one or more data marts. Those data marts are often dimensional models: sets of related tables that each capture one business perspective. The medallion documentation says Gold is where you model data for reporting and analytics with a dimensional model, by establishing relationships and defining measures. The blog names the industry-standard styles directly: Kimball-style star schemas and Inmon-style data marts fit in Gold. Gold models are de-normalized and read-optimized, with fewer joins. That is the opposite of Silver's priorities.
Because Gold models a business domain, it is organized around how the business consumes data. The blog describes consumption-ready, project-specific databases such as Customer Analytics or Inventory Analytics. Some customers create multiple Gold layers for different needs, such as HR, finance and IT. Gold holds fewer datasets than Silver and Bronze. Large volumes of detailed history usually stay in Silver instead of being materialized in Gold. Aggregation is the other Gold job: frequently used measures are pre-computed so analysts don't have to rebuild them.
CREATE OR REPLACE MATERIALIZED VIEW main.example_output.weekly_bookings AS
SELECT date_trunc('week', check_in) AS week,
property_id,
status,
count(*) AS total_bookings,
sum(total_amount) AS total_revenue
FROM samples.wanderbricks.bookings
GROUP BY week, property_id, statusCheckpoint 4 of 5· Match them up
Match each data model to the medallion layer where Databricks sources place it
Tap a term, then the definition that fits it.
Bronze copies the sources, Silver holds the integrated, write-friendly warehouse, and Gold holds the read-optimized dimensional models and data marts that analysts query.
“We see a lot of Kimball style star schema-based data models or Inmon style Data marts fit in this Gold Layer of the lakehouse.”Source: www.databricks.com
Checkpoint 5 of 5· Exam question
An analytics engineer is deciding where in the Medallion Architecture to apply deduplication, standardized column types, and enterprise-wide conformance rules to data that already exists in the Bronze layer. Which layer should own this work, and what data modeling characteristic should it use?
Correct answer: A — The Silver layer should own this work, using a normalized model closer to third normal form so multiple downstream teams share one consistent, conformed version of each entity.
- A. Deduplication and enterprise-wide conformance are Silver layer responsibilities, and Silver typically uses a normalized model similar to third normal form so many teams can build on one shared, cleansed version of each entity before it is denormalized further downstream.
- B. Bronze exists to capture source data as-is with minimal transformation for auditability and rapid ingestion; applying conformance rules and a star schema at Bronze would defeat its purpose as an untouched historical record.
- C. Hub, link, and satellite tables describe data vault modeling, and conformance cleanup is not the Gold layer's role; Gold consumes already-conformed Silver data to produce business-level aggregates, not raw audit trails.
- D. Silver and Gold serve different purposes and typically use different structures: Silver favors a normalized, conformed model while Gold favors a denormalized, aggregate-ready one, so treating them as interchangeable misrepresents both layers.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Every Databricks lakehouse must implement Bronze, Silver and Gold layers.Why is that wrong?
The medallion architecture is a recommended pattern, not a mandatory one.
Covered in Bronze, Silver, Gold: layers of increasing data quality
2.Data Vault and 3NF models belong in the Gold layer because Gold is the most refined.Why is that wrong?
The integrated warehouse, often 3NF or Data Vault, is modeled in Silver and feeds the dimensional data marts in Gold.
Covered in Silver: where the enterprise data warehouse is modeled
3.Star schemas are built in Silver, because Silver is where data is cleaned and joined.Why is that wrong?
Silver does cleaning and integration. Dimensional models such as star schemas are built in Gold for reporting and analytics.
Covered in Gold: dimensional models, data marts and aggregates
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Following the medallion architecture is a recommended best practice but not a requirement.”
↩︎ Bronze, Silver, Gold: layers of increasing data quality“Is intended for consumption by workloads that enrich data for silver tables, not for access by analysts and data scientists.”
↩︎ Bronze, Silver, Gold: layers of increasing data quality“Is where you perform data cleansing, deduplication, and normalization.”
↩︎ Silver: where the enterprise data warehouse is modeled“some customers create multiple gold layers to meet different business needs, such as HR, finance, and IT.”
↩︎ Gold: dimensional models, data marts and aggregates“The medallion architecture describes a series of data layers that denote the quality of data stored in the lakehouse.”
↩︎ Key concept“Following the medallion architecture is a recommended best practice but not a requirement.”
↩︎ Exam trap 1“The gold layer is where you'll model your data for reporting and analytics using a dimensional model by establishing relationships and defining measures.”
↩︎ Exam trap 3 - 2.
“The data warehouse is modeled in the silver layer and feeds specialized data marts in the gold layer.”
↩︎ Silver: where the enterprise data warehouse is modeled“Frequently, data marts are dimensional models in the form of a set of related tables that capture a specific business perspective.”
↩︎ Gold: dimensional models, data marts and aggregates“The data warehouse is modeled in the silver layer and feeds specialized data marts in the gold layer.”
↩︎ Exam trap 2“Often, this data follows a Third Normal Form (3NF) or Data Vault model.”
↩︎ Prediction - 3.https://www.databricks.com/blog/what-is-medallion-architectureSecondary source
“Data Vault-like, write-performant data models can be used in this layer.”
↩︎ Silver: where the enterprise data warehouse is modeled“The Gold layer is for reporting and uses more de-normalized and read-optimized data models with fewer joins.”
↩︎ Gold: dimensional models, data marts and aggregates“We see a lot of Kimball style star schema-based data models or Inmon style Data marts fit in this Gold Layer of the lakehouse.”
↩︎ Checkpoint