Banking · Tier 2 US commercial bank

A Tier 2 bank mapped 2.5 million lines of COBOL and CICS in 18 days instead of 8 months.

The bank's deposit, lending and ACH cores ran on IBM z/OS. A planned eight-month manual discovery was replaced by a PYD ENTERPRISES Enterprise Portfolio scan that finished the full AST parse, DMN extraction and OpenAPI synthesis in 18 days.

ENGAGEMENT PROFILEBanking

Customer: Tier 2 US commercial bank
Code: 2,500,000 LOC
Stack: IBM Enterprise COBOL 6.3CICS TS (BMS maps)JCL & PROCsVSAM KSDSDB2 for z/OSEndevor

from first PDS export to signed-off rule inventory: 18 days; from first PDS export to signed-off rule inventory
discovery budget not spent ($1,150,000 planned minus $85,000): $1,065,000; discovery budget not spent ($1,150,000 planned minus $85,000)
of program runway returned to the build phase: 7 months; of program runway returned to the build phase
business rules extracted, each linked to file and line range: 6,214; business rules extracted, each linked to file and line range

The challenge

What they were up against.

The bank had committed to moving deposits and commercial lending off the mainframe within three years. The original discovery plan staffed 12 consultants for 8 months, roughly $1,150,000 in billing at the blended rate in the statement of work, to read 2,500,000 lines of COBOL, 1,900 CICS transactions and 4,100 JCL jobs and write specifications by hand. The lending subsystem alone held about 1.5 million lines, including interest accrual written in 1989 and amended through dozens of rate-change projects. Nobody on staff could say which of the 14 accrual paragraphs were still reachable. The chief enterprise architect also had a hard constraint: no source code could leave the bank's data center, so any hosted analysis tool was out before the evaluation started.

The approach

How the work ran.

Week 1: install inside the perimeter and parse everything

PYD ENTERPRISES was installed from signed images on a three-node Kubernetes cluster in the bank's private data center with egress denied at the network policy layer. Source came from Endevor as PDS exports: COBOL members, copybook libraries, BMS mapsets, JCL and PROCs, plus DB2 DDL for the lending and deposit databases.

The first full parse of 2.5 million lines finished in under three hours. Copybook expansion resolved 1,408 copybooks with a maximum nesting depth of five. The parser flagged 31 programs using ALTER and 212 using GO TO DEPENDING ON, and mapped 3,880 REDEFINES overlays to byte-accurate layouts so packed-decimal balances read through either name resolved to the same storage.

Weeks 2 to 3: extract, review, sign off

The rule extractor separated business logic from plumbing: file status checks, abend handlers, audit logging and CICS HANDLE CONDITION blocks were classified as infrastructure, which left 6,214 business rules. Rules were grouped into 402 DMN 1.5 decision tables covering interest accrual (actual/360 and actual/365 bases), tiered fee schedules, overdraft limits and ACH return-reason handling.

Bank analysts and two retired mainframe developers brought back on contract reviewed every table in the sign-off view, with each DMN row shown beside the COBOL lines it came from. They rejected or renamed 7% of rule names and confirmed the logic itself. 118 paths were flagged for human review, mostly dynamic CALL targets resolved through a DB2 lookup table, and the bank resolved them by supplying the table extract.

Output handed to the build teams

402 DMN 1.5 decision tables, imported into the bank's Camunda environment for the new decision services
214 OpenAPI 3.1 operations synthesized from CICS COMMAREAs and BMS maps
A data lineage graph from every VSAM and DB2 field to the programs that write it
2,960 behavioral equivalence fixtures for the accrual and fee rules
Mermaid ER diagrams for the lending and deposit schemas

Outcome

The planned 8-month discovery phase closed in 18 days at the Enterprise Portfolio price of $85,000, against $1,150,000 budgeted: $1,065,000 not spent and 7 months returned to the build phase. More useful to the architecture board was the dead code finding: 9.6% of procedure division code in lending was unreachable, which removed three planned microservices from scope before anyone wrote them.

“We extracted thirty-five years of undocumented lending calculation rules from 1.5 million lines of COBOL, and not one byte left our data center.”

Chief Enterprise Architect, Tier 2 US commercial bank

Start with one system.

Request a Discovery Scan for one system up to 500,000 LOC, or try PYD ENTERPRISES on 100,000 lines for 14 days.

Request a codebase scan Start 14-day trial