Data Platform Control Plane

Quality gates + GCP-native reliability

Two separate controls: Great Expectations validates data contracts before release; Cloud Scheduler and Cloud Monitoring verify the live BigQuery platform after deployment.

48/48GE expectations passed

55,500Source rows contract-tested

50,000Warehouse facts verified live

6-hourGCP reliability schedule

✓ Quality gate passed — release allowed

Great Expectations: release-boundary contracts

GE runs automatically in CI before release: once over the full source delivery, then again over the AI-facing enrichment. Domain-specific PII, identity, and lineage checks run as a separate custom checkpoint.

	Contract group	Coverage	Observed result
✓	`source_release_contract`	55,500 rows · 27 expectations	27/27 passed
✓	`enriched_ai_contract`	497 rows · 21 expectations	21/21 passed
✓	`exception visibility`	108 negative billing values observed	Within explicit 0.5% tolerance
✓	`release behavior`	Critical contract failure	Blocks publish

✓ Reconciliation passed — no silent row loss

Source-to-warehouse reconciliation

The grain chain accounts for every source row and verifies that entity resolution and the warehouse patient dimension agree.

55,500 raw→5,500 exact duplicates removed→50,000 encounters→40,235 canonical patients

✓ Live GCP probe — final verification true

Platform reliability: detect, retry, verify, escalate

A GCP-native probe runs every six hours against the live BigQuery warehouse. Each execution writes a verified receipt to healthcare_analytics.pipeline_run_history; HTTP failures trigger bounded Scheduler retries and a Cloud Monitoring alert policy.

	Reliability control	Native GCP mechanism	Latest proof
✓	Detect	Cloud Run probe → BigQuery	50,000 fact rows queryable
✓	Retry	Cloud Scheduler exponential backoff	Enabled
✓	Verify	BigQuery run receipt	`final_verification=true`
✓	Escalate	Cloud Monitoring alert policy	Enabled on HTTP 5xx