dbt Testing in Practice: A Data Quality Baseline You Can Trust

dbt tests can dramatically reduce breakages — but only if they are chosen and operated carefully. Too few tests and you miss real issues; too many expensive tests and teams stop running them.

Baseline tests for every model

• unique + not_null on primary keys
• not_null on critical foreign keys
• accepted_values for enums that drive logic

Add pipeline-level checks

Row-count anomalies and freshness checks catch upstream outages. Even simple thresholds outperform ‘green job == good data’.

Keep tests fast

• Prefer column-level tests that scan fewer bytes.
• Partition big tables so tests can run on recent partitions.
• Move expensive reconciliations to scheduled jobs.

yaml
version: 2
models:
  - name: mart_orders
    columns:
      - name: order_id
        tests: [unique, not_null]
      - name: status
        tests:
          - accepted_values:
              values: ['created','paid','shipped','refunded']

Operate tests like production

When a test fails, the workflow should be clear: who triages, how to rerun, and how to fix upstream. Add alert routing and dashboards for failing tests over time.