Data Quality Test Planner

When to use this

Use when setting up or auditing a data pipeline's test coverage, before relying on a dataset for decisions.

Prompt template

You are a data quality engineer. Dataset/pipeline: {{dataset_description}}
Schema (key columns): {{schema}}
How it's used downstream: {{downstream_use}}

Propose data quality checks covering:
1. Completeness (nulls where they shouldn't be, missing expected rows).
2. Uniqueness (keys that should be unique).
3. Validity (value ranges, allowed categories, referential integrity).
4. Freshness/timeliness (how stale is too stale, given the downstream use).

For each check, state the rule, the severity if it fails (block pipeline vs. warn), and why that severity fits the downstream use.

Customize

Preview

You are a data quality engineer. Dataset/pipeline: {{dataset_description}} missing
Schema (key columns): {{schema}} missing
How it's used downstream: {{downstream_use}} missing

Propose data quality checks covering:
1. Completeness (nulls where they shouldn't be, missing expected rows).
2. Uniqueness (keys that should be unique).
3. Validity (value ranges, allowed categories, referential integrity).
4. Freshness/timeliness (how stale is too stale, given the downstream use).

For each check, state the rule, the severity if it fails (block pipeline vs. warn), and why that severity fits the downstream use.

Usage instructions

Describe the real downstream use so severity levels make sense -- a dashboard refreshed hourly needs different freshness rules than a monthly batch report.

Example output

Checks like 'order_id must be unique and non-null -- block pipeline on failure', 'amount must be >= 0 -- block', 'created_at must have rows for yesterday by 8am -- block, since the 9am dashboard depends on it'.

Limitations

Severity recommendations are a starting point; actual blocking thresholds should be agreed with the teams who'd be paged.

Rate this prompt