Governance in Production

• Run a governed catalog: owners, definitions, classification, retention as code
• Operationalize access, audit and lifecycle policy
• Map compliance obligations (GDPR, SOC2) to concrete catalog controls

• Regulated (GDPR/HIPAA/SOC2) obligations become concrete, testable controls instead of promises.
• A trustworthy catalog stops the 'define it yourself' chaos that produces duplicate, contradictory tables.
• Retention and erasure automated by policy keep legal risk and cost in check.
• Lineage through governance means impact analysis and audits use the same map.

Codes (policy) are dead text without the inspector checking every build step (catalog-driven controls, CI gates, audit). Governance is the inspector who checks doors, exits and wiring as work happens — not a binder of rules on a shelf.

• Catalog as registry: every dataset declares owner, steward, class, PII flags, retention window at creation.
• Policy as code: access rules, masking and lifecycle rules versioned in the repo and applied at deploy.
• Access management: role-based grants via groups; requests flow through the catalog, not ad-hoc SQL.
• Retention/lifecycle: policy-driven deletion/archival jobs with approval and audit for exceptions.
• Erasure: GDPR right-to-erasure workflows query the classification inventory and execute per-subject deletes.
• Audit + attestation: periodic access reviews and compliance checks run from the catalog, not from memory.

• A bank's new-request catalog requires class and retention for every table; a CI check refuses deploy until they exist.
• A retailer runs GDPR erasure as a lifecycle job: subjects in a queue are deleted from every table whose policy says PII, with confirmation logs.
• A cloud company does quarterly access reviews generated from the catalog: stale grants are auto-revoked and exceptions are escalated to the steward.

• Make the catalog authoritative and code-checked at creation — fixing governance later is ten times the cost.
• Automate retention/erasure; manual deletion is where legal risk hides.
• Review access on a cadence and auto-revoke stale grants with audit.
• Map compliance controls to exact catalog checks so 'are we compliant' is queryable.

• Creating a governance committee that owns policy docs nobody executes.
• Retention policies that exist but kill jobs that read beyond the window (policy vs reality divergence).
• Audit as an afterthought — no immutable log of who granted what.
• Governance applied only in production, so dev copies grow unclassified PII.

• 'Denied by policy' → check class + masking + retention together; contradictory layered policies are the top cause.
• Erasure gaps → the classification scan missed a derived table; rerun discovery over lineage expansion.
• Retention job deletes something needed → sign-off/approval workflow was bypassed; add a two-eye check to exceptions.
• Reviews drift → schedule attestation from the catalog query so people review a real grant list.