Lineage is mandatory for the audit and partial in practice.

Every regulator wants the same thing: trace the number in the report back to where it began. Most lineage diagrams answer a narrower question, which is where the number was last seen.

What happened is that lineage matured as a standard faster than it matured as coverage. OpenLineage is an open framework for collecting lineage metadata across pipelines, with Marquez as its reference implementation under the Linux Foundation's LF AI & Data. The tooling to record provenance is real, governed, and adopted. The question is how far up the chain the recording actually reaches.

It matters because the audit need is specific. Column-level lineage tracks how an individual field was computed and transformed, not merely that one table fed another. Table-level lineage tells an auditor that customer data reached a report; column-level tells them which field, through which transformation, produced the figure under review. The gap between those two sentences is the entire point of the exercise.

What it reveals about the field is a familiar boundary problem. The Basel Committee's risk-data principles, in force since 2013, require banks to aggregate risk data fully and accurately, and a 2023 progress report found additional work required at all assessed banks to attain or sustain full compliance. Lineage that begins where data lands in the warehouse quietly excludes the spreadsheets, hand-edits, and source systems where the number was actually shaped.

What to watch is where the graph terminates. A lineage program that stops at the warehouse door is documenting custody, not origin. The reveal: lineage explains everything except the one thing the audit asked, which is where the number came from and who is accountable for the transformation that made it.