GUIDE, read 10 Nov 2026
Audit lineage works when classification labels follow the column through every hop, down to the report
An auditor asks for a path, not a diagram
The audit question is usually specific. Where does this customer email go after it lands in the warehouse? Which reports show it? Who can open them? The answer must be a list that can be checked: starting column, each hop, each endpoint, each access rule. A screenshot of a lineage graph does not meet that bar unless the graph is complete and dated.
Databricks says lineage helps track sensitive data flow for compliance audits. That is the common framing, and it is accurate as far as it goes. Lineage is one input. The other inputs are classification, which says which columns are regulated, and access control, which says who can see them. The useful tools join these three. This guide covers the first two, and notes where the third shows up.
Three questions the evidence must answer
First, completeness: does the path end at the real consumers, including BI fields and exports, or at the last table the tool could parse? Second, labeling: does the regulated label travel with the column, so that a downstream column derived from an email column is also labeled? Third, currency: can you show what the graph said on the audit date?
Each question maps to a limit in the evidence. Completeness is the BI-reach problem covered in the guide on why lineage stops at BI. Labeling is the propagation table below. Currency is Snowflake's one-year retention, which means a graph older than that is not recoverable from the tool.
Which tools move classification labels along lineage
Propagation is the feature that turns a lineage graph into an audit trail. Without it, someone must label each downstream column by hand, and the labels drift. The table gives what each vendor documents.
| Tool | What the vendor documents | Automatic? |
|---|---|---|
| Atlan | Metadata Propagator pushes owners, certificates, tags and custom metadata from upstream assets to downstream dependents along lineage | App-driven (a propagator app) |
| DataHub | Glossary Term Propagation pushes glossary terms (classification labels) to downstream lineage columns and assets | Public Beta in DataHub Cloud |
| Snowflake | Lineage tab detects missing or mismatched tags on upstream and downstream columns and helps apply them; tags inherit down the object hierarchy and can be set to propagate | Detection plus assisted apply |
| Select Star | Says lineage powers automations such as propagating PII classification tags | Vendor claim |
| Euno | Docs say Cyera classifications are carried through column-level lineage into BI fields, dashboards and reports | Vendor claim |
| Manta (IBM) | Lineage is enriched with external metadata such as profiling information, quality scores and PII labels | Enrichment, propagation not stated |
| Alation | Automatic propagation to downstream lineage nodes: none found on 10 Nov 2026 | None found |
| Collibra | Lineage page names no mechanism that propagates tags, classifications or ownership; none found | None found |
| Foundational | Automatic downstream propagation of tags: none found on 10 Nov 2026 | None found |
| OpenLineage | Facets describe metadata per entity; propagation is left to consuming catalogs | None found |
Tools with the strongest documented propagation
Atlan's approach is explicit. Its Metadata Propagator app pushes owners, certificates, tags and custom metadata downstream, and it can restrict propagation to single-upstream targets, to avoid conflicting metadata when several sources feed one table. Check that it is switched on for the tags your auditor cares about before you rely on it.
DataHub's glossary term propagation pushes classification labels to downstream columns and assets. DataHub's page marks it as Public Beta in DataHub Cloud, available through a representative. Check which edition you run before you count on it.
Snowflake works at the column. Its lineage tab detects missing or mismatched tags on upstream and downstream columns and helps apply them. Snowflake also says tags are inherited down the object hierarchy and can be set to propagate automatically. Snowflake's lineage details shown to a viewer depend on that viewer's access control privileges, so an auditor's account needs enough privilege to see the whole path.
Euno says its Cyera integration carries classifications through column-level lineage into BI fields, dashboards and reports. That is a vendor claim about reaching the report layer, which is where auditors often look. Select Star makes a similar claim about PII tag propagation.
Where propagation is missing or manual
We found no Alation page documenting automatic propagation of tags, classifications or glossary terms to downstream lineage nodes (none found on 10 Nov 2026). Foundational's MCP examples query columns tagged pii and their downstream consumers, which gives you the audit list on request, but we found no documentation of automatic downstream propagation.
Collibra's lineage page markets lineage with business context and audit trails, and names no mechanism that propagates tags or classifications along lineage. OpenLineage and Marquez leave propagation to the catalog consuming the events.
Manual does not mean wrong. A team can label every downstream column by script and keep a ledger. The risk is drift, and an audit is where drift is found.
The tool's own evidence has limits too
Check three things about the lineage data itself. How far back can you query it? Snowflake retains lineage for one year, and lineage captured before November 2024 for data movement is not available. Who can see it? Snowflake limits the details shown to the viewer's privileges. What was excluded? Snowflake lineage is not available for objects in shared databases or the SNOWFLAKE database.
Then check the tool as a vendor, since the lineage store now holds a map of your sensitive data. Each tool page on this site records the security statements found on the vendor's public pages, and an auditor may ask for them.
A worked example with one email column
Take a customer email column in a raw table. In the warehouse it passes through a staging model, a customer dimension and a marketing mart. In the BI tool it feeds one dashboard field that shows the domain part of the address. The audit path has five hops, and the regulated label must be present on all five.
Start from the label, not the graph. Ask the tool for every column carrying the email classification today. Then ask for every column downstream of the raw email column. The two lists should match, apart from columns where masking or hashing removed the sensitivity. Each difference is a finding: either a downstream column is missing its label, or a column carries a label it should not. Snowflake's lineage tab is built for the first case, since it detects missing or mismatched tags along lineage.
Then check the last hop. If the tool stops at the dashboard, the field that shows the domain is not in the path, and you must add that hop from the BI tool's own metadata. Tools that document BI field lineage, such as Euno, DataHub, Foundational and Select Star in our matrix, can include it. The others need a manual step.
A pack to give the auditor
Produce one document per regulated field: its classification, the lineage path with each hop dated, the list of reports and fields at the end, and the access rules on each. Run the same query again a month later and compare. A difference is either a change you made or a gap in the tool, and both are worth knowing. The glossary defines propagation and classification as this site uses them.
Sources
- Databricks: docs.databricks.com/aws/en/data-governance/unity-catalog/data-lineage, read 10 Nov 2026
- Snowflake: docs.snowflake.com/en/user-guide/ui-snowsight-lineage.md, read 10 Nov 2026
- Snowflake: docs.snowflake.com/en/user-guide/object-tagging/introduction.md, read 10 Nov 2026
- Atlan: docs.atlan.com/product/capabilities/lineage/how-tos/propagate-metadata-through-l, read 10 Nov 2026
- DataHub: docs.datahub.com/docs/automations/glossary-term-propagation, read 10 Nov 2026
- Select Star: www.selectstar.com/product/data-lineage, read 10 Nov 2026
- Euno: docs.euno.ai/sources/data-security/cyera-integration/cyera-integration-use-cases, read 10 Nov 2026
- Manta (IBM): www.ibm.com/products/watsonx-data-intelligence/data-lineage, read 10 Nov 2026
- Alation: www.alation.com/product/data-governance, read 10 Nov 2026
- Collibra: www.collibra.com/products/data-lineage, read 10 Nov 2026
- Foundational: docs.foundational.io/en/articles/11194236-use-foundational-mcp-with-your-ai-tool, read 10 Nov 2026
- OpenLineage: openlineage.io/docs, read 10 Nov 2026
Last reviewed 10 Nov 2026.