Data quality checks often begin with the table: missing values, duplicates, outliers, and broken joins. Those checks matter, but they can all pass while the analysis is still wrong.
A definition check asks what each measure actually includes. Is revenue gross or net? Is a customer a person, account, household, or transaction? Does the date represent an order, shipment, or payment?
It also checks the population and time period. A clean comparison can become misleading if one side excludes closed stores, partial weeks, or customers without a recorded attribute.
Clarifying those meanings early is fast, and it often exposes the largest risk before any code is written.
Related reading
- A fuller checklist for deciding whether a dataset is safe to use.
- A projection error caught by checking whether the totals still made sense.
This note forms part of the data quality and analytical judgement guide.