Consumption & cost · rule warehouse-hot-model-untested
Widely read dbt models with no tests
A Metacenta review checks this under the rule Widely-read models have at least one test. Everything below applies whether or not you ever commission one.
What this rule checks
This rule flags dbt models read by at least five different people in the collection window that have no test. Neither half is new; the value is the intersection of the two.
Why it matters
Nothing would catch a duplicate row, a null key or a silent upstream change before the people relying on it did. They are the ones who would find out, usually in a meeting.
How to fix it
Test the models people query most. Add a uniqueness test on the model's grain and not_null on the keys it joins by. Start here: these are the untested models with the most readers.
Before:
models:
- name: fct_orders
After:
models:
- name: fct_orders
columns:
- name: order_id
data_tests:
- unique
- not_null
When it is fine to leave
A model tested by a check outside dbt, such as a data observability tool, is covered in practice. Tell us, and the finding comes out of your report.
What we need to check it
manifest.json and warehouse query history that reports distinct readers. Without reader counts the rule declines, because a widely read model cannot be told from a heavily scheduled one.