For Data Stewards
What Changes for You
Section titled “What Changes for You”The roche-data platform automates the infrastructure work that surrounds your expertise — so you can focus on the governance decisions that actually require human judgement.
Before: Manual Assembly
Section titled “Before: Manual Assembly”Today, bringing a new data product to life requires you to:
- Manually define data contracts in documents or spreadsheets
- Coordinate with data engineers to create Snowflake schemas
- Work with analysts to build dbt models for each layer
- Write quality check definitions and hand them to specialists
- Curate governance metadata in Collibra by hand
- Repeat for every entity in every domain
Typical timeline: 6-8 weeks per data product, involving 4-6 specialists.
After: Automated Generation
Section titled “After: Automated Generation”With roche-data, your work becomes:
- Define the model in RTiS — the entity, its attributes, its relationships
- Review generated artifacts — data contracts, quality rules, governance metadata
- Approve through pull request — peer-reviewed, version-controlled, auditable
Everything else is generated: Snowflake schemas, dbt models, quality gate predicates, DQ rules, API specifications, semantic definitions, and documentation.
Typical timeline: Minutes from model definition to deployed data product.
What You Still Own
Section titled “What You Still Own”Automation does not replace stewardship — it amplifies it.
| Your responsibility | How the platform supports it |
|---|---|
| Data definitions | You define entities and relationships in RTiS. The platform compiles them into every downstream artifact. |
| Quality rules | You bind shared CEL rules to entity properties in rules.yaml. The platform compiles them into dbt tests and Snowflake UDFs. See Writing DQ Rules. |
| Ownership & SLAs | Governance metadata from Collibra is pulled automatically and embedded in data contracts. |
| Access policies | Access control is managed through Snowflake RBAC and Collibra governance metadata. |
| Review & approval | Every generated artifact passes through a pull request. You review and approve before deployment. |
Quality Gates — Your Safety Net
Section titled “Quality Gates — Your Safety Net”Every data product passes through four progressive quality gates. You define the rules; the platform enforces them.
| Gate | What it checks | Your role |
|---|---|---|
| G1 Completeness | Schema compliance, required fields, data types | Define the schema in RTiS |
| G2 Validity | Master data lookups, referential integrity | Specify which reference data to validate against |
| G3 Business Rules | Range checks, cross-field consistency, freshness | Bind rules from shared library in rules.yaml — see QA Approach |
| G4 Consistency | Trend analysis, anomaly detection | Review flagged deviations |
Data that passes all four gates earns Certified status and becomes available to AI tools (Cortex Analyst) and external APIs.
The Ratification UI
Section titled “The Ratification UI”For taxonomy and tag management, the platform provides a dedicated Streamlit application:
- Review and approve proposed changes to classifications, definitions, and tags
- Track change history with full audit trail
- Collaborate with domain experts through a structured approval workflow
See Ratification UI for details.