What is staging in dbt?

Staging is the first dbt layer above raw sources. Staging models usually have a 1:1 relationship with source tables and do light, reversible cleanup.

Typical staging work

• Rename columns to clear analytics names
• Cast types (string → timestamp, cents → dollars)
• Trim/standardize simple fields
• Soft delete filters if clearly marked
• Pass-through of source keys

What staging usually does NOT do

• Heavy business definitions ("active customer")
• Wide multi-source joins into final facts
• Complex windowed metrics

source(raw.orders) → stg_orders → int_* → fct_orders / dim_*

Example

-- models/staging/stg_stripe__payments.sql
with src as (
    select * from {{ source('stripe', 'payments') }}
)
select
    id as payment_id,
    order_id,
    cast(amount as numeric) / 100.0 as amount,
    cast(created as timestamp) as created_at
from src

Conventions

Many teams name files stg_<source>__<table>.sql and materialize staging as views.

Interview tip: "Staging is the anti-corruption layer: stabilize names and types once, so marts do not all re-cast raw junk."