Data Quality Validation Checklist
A validation checklist covering accuracy, completeness, consistency, and timeliness of data after a migration or pipeline change. It moves beyond row counts to checksums, referential integrity, conversion checks, and aggregate reconciliation. Use it to make data quality a measured, owned gate.
When to Use This Checklist
Use this checklist whenever data moves or changes shape: after a database migration, an ETL/ELT pipeline change, a schema evolution, or a new data source onboarding. Data quality is the degree to which data is fit for its intended use. Without explicit validation, errors stay hidden until a report is wrong or a model misbehaves.
How to Use This Checklist
Start by defining the quality dimensions that matter for each dataset and the thresholds that count as pass or fail. Then run the structural checks: counts, checksums, nulls, referential integrity, and uniqueness. Follow with semantic checks like aggregate reconciliation and distribution profiling. The required items catch the failures that corrupt analytics; the optional items add automation and statistical depth. Always route results to a data owner who signs off rather than relying on a silent green light.
What Good Looks Like
Good validation has explicit, documented thresholds per dimension, not a vague sense that the data looks fine. Counts and checksums confirm parity, referential integrity holds, and business aggregates reconcile to the source. Conversions for dates, timezones, precision, and encoding are verified, not assumed. The checks are automated so they protect every future run, and failures have a named owner who must approve before data is trusted. Quality becomes a measured, repeatable gate rather than a one-off manual review.
Common Pitfalls
Checking only row counts is the most common shortcut; counts can match while values are wrong. Teams frequently miss timezone and numeric-precision drift, which produces small but compounding errors. Silent default substitution, where nulls become zeros or empty strings, distorts aggregates. Skipping uniqueness checks lets duplicate keys slip in during merges. Finally, manual one-time validation gives no protection against future regressions, so the same errors reappear later.
Related Resources
Ground this work in data-quality-management and codify expectations as data-contracts. Use a data-governance-framework to assign ownership and data-lineage to trace failures back to their source.