Case Study 9.1: GlobalBank's Copybook Migration

Background

In 2015, GlobalBank's core banking system contained 347 programs that processed account records. The ACCT-MASTER record layout was defined inline in every program — 347 copies of the same 01-level definition, each potentially different due to years of independent maintenance.

Maria Chen, newly promoted to senior developer, discovered the scope of the problem during a routine audit. She found 12 distinct versions of the account record definition across the 347 programs. Seven programs defined ACCT-CURR-BALANCE as PIC S9(9)V99 instead of PIC S9(11)V99. Three programs were missing the ACCT-LAST-TXN-DATE field entirely. One program had a misspelled field name (ACCT-BALENCE) that had survived for eight years because it happened to work — the program only used MOVE to copy the entire record and never referenced individual fields.

The Problem

A regulatory change required adding a new field: ACCT-COUNTRY-CODE (PIC X(2)). Under the current approach, this meant:

1. Modifying 347 programs
2. Testing each program individually
3. Coordinating deployment across all 347 programs in a single release window

Estimated effort: 3 developers, 6 weeks. Estimated risk: High (past experience showed that manual record layout changes produced a 2-3% error rate per program).

Maria's Solution

Maria proposed a three-phase migration to centralized copybooks:

Phase 1: Create the Canonical Copybook (1 week)

Maria created ACCT-REC in GLOBALBANK.PROD.COPYLIB, defining the single authoritative version of the account record layout. She included: - A comprehensive header comment with version number and modification history - 88-level condition names for every coded field - 30 bytes of FILLER for future expansion - The new ACCT-COUNTRY-CODE field (carved from what would have been FILLER)

Phase 2: Incremental Migration (8 weeks)

Rather than converting all 347 programs at once, Maria prioritized: - Tier 1 (Week 1-2): The 5 most critical programs (ACCT-MAINT, TXN-PROC, BAL-CALC, RPT-DAILY, ACCT-CLOSE) — replaced inline definitions with COPY ACCT-REC - Tier 2 (Week 3-4): The 20 programs that had known discrepancies from the canonical layout - Tier 3 (Week 5-8): Remaining 322 programs, converted 40-50 per week by junior developers following Maria's written procedure

Phase 3: Prevention (Ongoing)

Maria established a coding standard: no new program could be approved in code review if it defined the account record inline. All programs must use COPY ACCT-REC.

Results

Lessons Learned

1. Incremental migration works. Converting 347 programs all at once would have been too risky. Tiered migration allowed the team to build confidence and catch issues early.

2. Standards require enforcement. Without the code review gate, developers would have slowly drifted back to inline definitions under deadline pressure.

3. The FILLER strategy paid off immediately. The ACCT-COUNTRY-CODE field was the first of five fields added over the next three years, all carved from FILLER without changing the record length.

4. Compiler listings are essential. During migration, the expanded source listing was the primary tool for verifying that the copybook insertion produced the expected result.

Discussion Questions

1. What risks did Maria accept by migrating Tier 1 programs first? What risks would she have faced by starting with Tier 3?

2. How would you handle a program that references a field name from the old inline definition that does not exist in the new copybook?

3. If GlobalBank used source control (e.g., Git or Endevor) during this migration, how would you structure the branches or packages to manage the transition?

4. Estimate the total cost savings over five years, assuming the "before" approach required 6 weeks of effort per field addition and the "after" approach requires 0.5 days. Assume 2 field additions per year and an average developer cost of $80/hour.