Open Data Infrastructure
Foundation for AI Data Lineage SLAs
Why AI programs need lineage service levels for freshness, owner coverage, transformation detail, and incident review.
Lineage that arrives after the AI incident is documentation, not infrastructure.
Lineage needs service levels
AI systems use data faster than most governance processes can explain it. That creates a simple problem. If lineage is missing, stale, or too shallow, the AI system may still run, but the team cannot explain the path from source to answer.
A foundation for AI needs lineage SLAs. Those SLAs should cover freshness of lineage, owner coverage, transformation detail, policy context, critical data products, and incident-review access.
Coverage is not the same as usefulness
A graph can show many nodes and still miss the details an AI incident requires. Which source changed? Which transformation created the field? Which owner approved the product? Which policy applied when the answer was produced?
The useful SLA defines what lineage must answer, how quickly it must be available, which systems are in scope, and which products are too critical to tolerate unknown lineage.
Core idea: AI lineage SLAs define how quickly the organization can explain what the system knew.
The ODI pattern turns lineage into runtime context
Open Data Infrastructure treats lineage as operational context. It should support discovery, impact analysis, policy review, quality investigation, and agent behavior.
For adjacent context, read metadata SLAs for AI, catalog coverage gaps, and data lineage for AI-ready infrastructure.
What breaks first
- Lineage exists for warehouse models but not for retrieval indexes or tool outputs.
- Ownership metadata is present for tables but absent for derived context.
- Critical products have lineage gaps hidden behind manual processes.
- Incident review depends on screenshots instead of queryable lineage evidence.
Questions to ask
Ask which lineage questions must be answered in minutes, which data products require full coverage, and how lineage freshness is measured. Ask whether agents can retrieve lineage context without bypassing governance.
Sources to start with
These primary sources anchor the technical claims in this guide.
- OpenLineage object model documentation
- DataHub lineage documentation
- OpenMetadata lineage documentation
- NIST AI Risk Management Framework
AI lineage is ready when it can explain the answer while the decision still matters.