A certified or golden dataset is an officially approved source of truth for a specific business domain, verified to meet strict standards for data quality, documentation, ownership, and freshness SLAs. It gives analysts and data scientists a trusted foundation for reporting, eliminating confusion caused by conflicting ad-hoc tables created across the organization.
Defining certified sources of truth
To earn certification status, a dataset must satisfy rigorous platform criteria:
- Verified quality checks: The underlying pipeline must run automated, blocking assertions on every execution, validating primary key uniqueness, referential integrity, null thresholds, and source-to-target reconciliations.
- Comprehensive documentation: Every column, metric formula, calculation logic, and business assumption must be thoroughly documented in the data catalog.
- Explicit ownership and support: The dataset must have a designated engineering owner with an active on-call rotation responsible for addressing failures and maintaining schemas.
- Guaranteed SLAs: Backed by formal commitments regarding delivery schedules, freshness windows, and query availability.
Dataset Status Quality Gates Passed Documented SLA Guarantee Catalog Badge Golden / Cert Yes (Blocking) Yes Yes Official Green Badge Draft / Sand Partial Minimal No Untrusted / Sandbox
Surfacing certified datasets changes how organizations interact with data:
- Catalog and BI badges: Certified tables display visual verification badges in data catalogs, like Unity Catalog or DataHub, and BI platforms, like Looker or Tableau, immediately signaling trustworthiness to users.
- Prioritized search rankings: Data catalogs rank certified datasets at the top of search results, helping users select canonical models over private scratch tables.
- Eliminating rogue tables and metric drift: When certified datasets are readily discoverable, analysts stop creating unmonitored summary tables, reducing duplicate compute costs and preventing conflicting executive metrics.
Maintaining certification integrity
Certification is not a permanent label awarded once and forgotten. If a dataset repeatedly breaches its freshness SLAs or experiences recurring test failures, governance policies temporarily revoke its badge until reliability is restored.