“Keeping the final dataset is enough for an audit trail.”
Not quite.An audit trail also needs the inputs, transformations, versions, actors, and execution context that produced it.
Architecture & Systems
Data Lineage and Audit Trail
Data lineage traces each reported number back through runs, snapshots, transformations, and raw sources for full auditability.
Mental model
A number is trustworthy when its path from source data through transformations and runs can be reconstructed.
An audit trail connects reported risk or P&L figures to the underlying runs, including snapshot IDs, model IDs, and library versions.
Snapshots and models themselves link back to transformation configurations and raw data snapshots from external vendors.
Explore the mechanism
Can every published number be traced to the source data, transformation, actor, and run that produced it?
Data lineage and audit trail
A reported number should link back through runs, snapshots, transforms, and raw sources. Click nodes to inspect metadata.
Lineage depth
Depth
3
Completeness
100% of expected metadata present (toy scoring).
Interpretation
- Lineage is not “nice to have”. It is what makes numbers auditable.
- A run without snapshot/model/version is not reproducible.
- “Challengeability” comes from being able to replay each step.
Lineage tree
click a node
Metadata
Reported number
key
INFO
metric: VaR(99%)
key
INFO
date: 2026-02-01
key
INFO
desk: Rates
In production you want immutable IDs + checksums + config hashes so every step is replayable.
CONSEQUENCE PROBE
Can the published number be traced to its producing state?
raw quote→transform hash→model + config→published result
ReconstructableEach transition identifies both its input and the operation applied.
UNDERSTAND
What to notice
Lineage enables auditors and model risk teams to reconstruct how a number was produced and challenge each step if necessary.
A robust lineage system supports governance, reduces dispute resolution time, and builds trust in a central analytics framework.
Where it appears
- regulatory and model-risk evidence
- debugging data-quality failures
- reproducing reports and valuations
Where the intuition breaks
What this view leaves out
- Recorded lineage can still be incomplete or misleading if identifiers are unstable, transformations happen outside the tracked system, or metadata is mutable.
Common mistakes
Where this concept sits
Follow the strongest prerequisites, applications, extensions, and contrasts.
Connections for Data Lineage and Audit Trail
Data Lineage and Audit TrailYOU ARE HERE
Supported byRaw vs Curated Data and TransformationsLineage records how raw inputs were transformed into curated data used downstream.
Supported byRun ID and ProvenanceRun-level provenance links computational outputs to the broader lineage of inputs, transformations and code.
Supported bySnapshot Versioning and CorrectionsVersioned corrections preserve the history needed to explain which data was known or changed at each point in time.
Supported byStructured Logging and Run IDsStructured events carrying run IDs provide evidence that links inputs, transformations and outputs in an audit trail.
SupportsReplay of Valuation RunsLineage identifies the exact inputs and transformations that a replay must reconstruct rather than silently substituting current data.
Supported bySchema Migrations and Backward CompatibilityTraceable schema evolution is necessary to explain how historical records and downstream transformations should be interpreted.
Supported by · Raw vs Curated Data and Transformations
Lineage records how raw inputs were transformed into curated data used downstream.
Explore concept →