How do you validate data quality?

Direct answer

What a strong answer covers

How it works

Quality checks run where data crosses a boundary: at ingestion to catch upstream changes, after transformation to catch logic errors, and before publication to protect consumers. Each check asserts one property, schema and types, required fields present, keys unique, values within valid ranges or sets, foreign keys resolvable, data fresh within the expected delay, row counts within the usual band, and distributions close to recent history. A failing check has a severity that decides whether the pipeline stops or only alerts, and an owner who receives it. Results are tracked over time so silent decay becomes visible.

Example

An orders model runs dbt tests for unique order_id, not-null customer_id, accepted status values, and a freshness check of two hours; a custom test compares today’s row count with the trailing 7-day median and alerts beyond 30 percent deviation. When a source system started sending amounts in cents instead of dollars, the range check caught it before the finance dashboard did.

Common mistakes

• Checking only schema and letting semantically wrong values through.
• Alerting on everything with the same severity until people ignore the channel.
• Running checks with no owner so failures sit unresolved for weeks.

A useful answer structure

Answer from your own work.

Do not memorise a script: use only examples you can explain honestly in follow-up questions.