◆ Autonomous Worker
Traces every column back to its source and forward to every report and model that consumes it, keeping the map current as pipelines change. Column-level lineage resolves a 'where did this number come from' query in a single graph walk and flags downstream regulatory reports when an upstream transform changes.
Memory
Working The pipeline change being parsed and its lineage delta.
Semantic The current lineage graph and catalogue glossary.
Store Knowledge graph (column-level lineage)
Orchestration
MCP
Harness · Managed Agents: scheduled + change-triggered; code-exec for pipeline parsing.
Tools
⌕ Pipeline code / DAG repos Retrieval { } Lineage graph store API { } Catalogue platform API ›_ Lineage parsing sandbox Code exec
Evals & guardrails
- Lineage completeness measured against known critical-data-element coverage.
- Catalogue descriptions sampled by an agent-as-judge for accuracy.
- Versioned graph; lineage changes are diff-reviewable.
Frontier edge
- ▲Causal / counterfactual reasoning: answers 'if this source column changes, which reports and models break' by walking the lineage graph forward, not just recording history.
- ▲Multimodal parsing: reads pipeline code, SQL, notebooks and even dashboard screenshots natively to recover lineage the DAG metadata misses.
- ▲Proactive impact alerts: flags downstream regulatory reports at risk the moment an upstream transformation changes, before the report runs.
In numbers
96%
Critical-data-element lineage coverage
< 1 hour
Lineage staleness
Handoffs
Fed by ← Data Quality Agent
Across ⇢ Compliance → Regulatory Reporting for BCBS 239 attestation