Article · 3 min read

Mainframe data lineage: what COBOL, JCL and DB2 code says, and what it doesn't

Stéphane Goethals
Stéphane Goethals
COO & co-founder of Vokse
LinkedIn
Mainframe data lineage: what COBOL, JCL and DB2 code says, and what it doesn't

The ECB’s May 2024 RDARR guide, which sets out supervisory expectations under BCBS 239, asks for lineage “from capture”. For a bank still running on a mainframe, this expectation has a precise meaning: data lineage must cross the DB2 tables, the JCL steps and the COBOL programs, not only the risk data warehouse and the regulatory reporting engine.

This is where most tools stop. A tool that reads database schemas and warehouse processing covers loading, calculation and reporting. The rules that change the figure sit further upstream, in extraction, transformation and reconciliation, written twenty or thirty years ago by teams that have often left the bank.

What reading the code recovers

Take the exposure value of a corporate loan portfolio, reported in COREP template C 07.00. By reading the code as it executes, you recover:

  • the original DB2 table and the columns actually read (amount, currency code, counterparty, collateral);
  • the file produced by the JCL step, including when its name comes from a symbolic parameter overridden in some jobs;
  • the two readings of the same area when the copybook uses a REDEFINES depending on the product type;
  • the currency conversions, and the rate table they depend on;
  • the records dropped at reconciliation, with the condition that drops them.

Each link points to the line of code that justifies it. That reference is what makes the difference in front of an inspector: lineage no longer rests on documentation rebuilt for the occasion, it can be checked in the source.

What the code doesn’t say

Reading the code is not enough to establish everything, and it is better to know that before an inspection.

The code says what a program does, not why. A rule that drops loans without a currency code may be intentional or an oversight. Only someone from the business can decide, and that decision has to be written down.

The code also says nothing about what happens outside it. A manual fix in a spreadsheet between two runs, a file reloaded by hand after an incident, a parameter changed by operations on a reporting-date evening: none of this appears in the source. These workarounds are precisely the ones that ECB inspections flagged between 2022 and 2024.

Finally, technical lineage only becomes useful once it is linked to the business. Knowing that an amount field in the copybook feeds the exposures table is of no use until you know that it carries the “exposure amount” attribute of the glossary, and that it is part of the critical data of the indicator.

What this changes for the teams

For a risk IT team, the benefit is concrete. The question “where does this figure come from” is no longer settled by asking the two people who still know the chain. When the code changes, the lineage follows without re-entry. The time spent rebuilding documentation before each supervisory review shifts to what the code cannot say: business decisions and manual workarounds.

One condition remains non-negotiable in a bank: the tool is installed on the bank’s premises and only reads the code. Nothing leaves, nothing is modified.

See how Vokse links a COREP cell to its source code