How to Build a dbt Staging-to-Marts Project

The staging-marts pattern

A mature dbt project has a two-layer model hierarchy:

• Staging (models/staging/) - one model per raw source table. It selects, casts types, renames columns, and filters nothing. This is the single clean copy of the raw data that every downstream model uses.
• Marts (models/marts/) - business-question models that join and aggregate staging models into the tables analysts actually query.

The split matters because it decouples "what the source looks like" from "what the business wants to know."

Project scaffolding

A minimal dbt_project.yml sets the project name, profile, and materializations per layer:

name: shop
version: 1.0.0
config-version: 2
profile: shop

model-paths: ["models"]
target-path: target

models:
  shop:
    staging:
      +materialized: view    # staging models are cheap views
    marts:
      +materialized: table   # mart models are materialized tables

Declare sources

models/sources.yml tells dbt where your raw data lives. Using {{ source() }} instead of a hard-coded schema name means you can swap schemas (dev vs prod) without touching the model SQL:

version: 2
sources:
  - name: raw
    schema: public
    tables:
      - name: orders
      - name: customers

The staging model

models/staging/stg_orders.sql selects from the source reference and normalizes column names and types - nothing else:

SELECT
    id           AS order_id,
    customer_id,
    total_amount AS amount,
    created_at
FROM {{ source('raw', 'orders') }}

The mart model

models/marts/customer_orders.sql uses {{ ref('stg_orders') }} to reference the staging model. dbt's DAG resolver guarantees staging runs first:

WITH orders AS (
    SELECT * FROM {{ ref('stg_orders') }}
),
customers AS (
    SELECT id AS customer_id, name AS customer_name
    FROM {{ source('raw', 'customers') }}
)
SELECT
    c.customer_id,
    c.customer_name,
    COUNT(o.order_id)          AS total_orders,
    COALESCE(SUM(o.amount), 0) AS total_spend
FROM customers c
LEFT JOIN orders o ON o.customer_id = c.customer_id
GROUP BY c.customer_id, c.customer_name

Adding tests

A schema.yml next to the mart file attaches column-level tests. dbt build runs models AND tests in a single command:

version: 2
models:
  - name: customer_orders
    columns:
      - name: customer_id
        tests:
          - not_null
          - unique

Run the full project with:

cd dbt_project
dbt build --profiles-dir ..

dbt build compiles the Jinja SQL, materializes staging views, materializes the mart table, and runs every schema test - one command for the full lineage.