dbt semantic model keys with no unique test

For each dbt semantic model, this rule takes every primary or unique entity that names a plain column. It passes when a uniqueness test on the model covers that column alone. A natural entity is not judged, because it repeats by design.

MetricFlow assumes one row per key when it joins through the entity. A duplicate key multiplies every metric joined through it, and nothing fails. The number is simply wrong.

Test every semantic model key for uniqueness. Add a unique test on each flagged key column, in the model the semantic model reads. A where clause on the test stops it counting, so test every row.

Before:

semantic_models:
  - name: customers
    model: ref('dim_customers')
    entities:
      - name: customer
        type: primary
        expr: customer_id

After:

models:
  - name: dim_customers
    columns:
      - name: customer_id
        data_tests:
          - unique
          - not_null

If the model really can hold two rows per value, the entity is not a primary key. Declare it as natural or foreign instead. A composite grain test does not make any one of its columns unique, so it does not pass here.

manifest.json alone. unique, the dbt_expectations uniqueness tests and a one-column unique_combination_of_columns count. A test limited by where does not. An entity built from an expression cannot carry a column test, so it is not judged.