Sprint 1 Complete

COBOL Migration
Accelerator

From Zero to Knowledge Graph in One Sprint

AWS CardDemo · 200+ Mainframe Artifacts · March 2026

12
Artifact Types Covered
352
Tests / Zero Failures
96%
Code Coverage
The Challenge

The Business Problem

Mainframe migrations fail because teams cannot understand what the system does before they change it.

Operationally Opaque

COBOL programs encode decades of business logic in formats no modern tool can read. Knowledge lives in tribal memory, not documentation.

No Dependency Map

Programs CALL other programs, COPY copybooks, EXEC CICS, and access VSAM files. No single artifact reveals the full dependency web.

No Interface Catalog

COMMAREA and LINKAGE SECTION structures define inter-program contracts but are buried in source code with no API registry equivalent.

Weeks of Manual Discovery

Traditional discovery requires mainframe SMEs to manually trace call chains, map data flows, and document business rules. Slow and error-prone.

Architecture Context

CardDemo
Mainframe Source
KMFlow
Evidence Parsers
Knowledge Graph
Neo4j
Migration Outputs
Schemas, APIs, UI
Delivery

What Was Built (Business View)

🔍

Complete Automated Discovery

Every COBOL artifact type parsed and cataloged: source, copybooks, JCL, BMS maps, DB2 schemas, CICS resources, IMS definitions, and scheduler jobs.

📚

Financial Services Vocabulary

252 pre-loaded domain terms covering banking, payments, accounts, and compliance. The knowledge graph speaks the business language from day one.

🚀

Migration Planning Outputs

Auto-generated PostgreSQL schemas from VSAM/DB2, service boundary recommendations, UI scaffolding from BMS maps, and OpenAPI contracts from COMMAREA.

⚡

Platform Ready

COB project live in Blaze with Jira integration (8 epics), GitHub repository, and full onboarding wizard. Customer can start immediately.

End-to-End Flow

Mainframe Source

COBOL, JCL, BMS, DDL, CSD, DBD, PSB, CA-7

Evidence Parsers

7 new parsers, 16 existing

Knowledge Graph

Ontology + relationships

Migration Engine

Schema, arch, codegen

Target Architecture

PostgreSQL, FastAPI, React

Results

Outcomes & Evidence

12
Artifact Types (was 0)
352
Tests Passing
96%
Code Coverage
8
Epics Structured

Evidence-First Methodology

CDD Compliance

Evidence collected at every phase of the SDLC. Every parser, every test, every design decision has a traceable audit trail from requirement to deployment.

Real-World Validation

All 352 tests run against real AWS CardDemo files, not synthetic test data. Parser outputs verified against known mainframe program structures and data layouts.

Complete Audit Trail

From Jira epic to GitHub PR to deployed parser, every artifact is linked. Compliance managers can trace any output back to its originating requirement.

Roadmap

What This Enables (Next 90 Days)

Month 1 — Complete Discovery

Full CardDemo Ingestion

Ingest all 200+ artifacts from AWS CardDemo. Populate knowledge graph with complete program dependency map, data flow catalog, and business rule inventory.

  • Run all 7 parsers against full CardDemo corpus
  • Resolve cross-program CALL chains
  • Map VSAM file access patterns
  • Catalog all CICS transaction IDs
Month 2 — Architecture Blueprint

Auto-Generate Boundaries

Use knowledge graph to identify natural service boundaries. Generate BPMN process models from JCL batch flows and CICS transaction sequences.

  • Service boundary recommendations from call graph analysis
  • BPMN generation from JCL procedures
  • Data ownership mapping per service
  • API contract drafts from COMMAREA structures
Month 3 — Migration Kickoff

Code Generation Begins

PostgreSQL schemas from VSAM/DB2 layouts. FastAPI stubs from service boundaries. React component scaffolding from BMS screen maps.

  • PostgreSQL DDL from VSAM DEFINE CLUSTER
  • Alembic migrations with type-mapped columns
  • OpenAPI specs from interface contracts
  • React forms from BMS map definitions
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
W11
W12
Discovery
Blueprint
Migration
Strategy

Strategic Value

Platform Asset

7 parsers + 252 domain terms are reusable for every COBOL customer. Built once, deployed to every future mainframe engagement. The investment compounds.

Amortized cost per engagement drops with each new customer onboarded.

Competitive Moat

The knowledge graph is permanent institutional memory. Competitors deliver a migration. Blaze delivers a migration plus an auditable knowledge base that persists beyond the project.

Knowledge graph becomes a living asset the customer keeps forever.

Accelerated Economics

Next engagement starts with 80% of discovery pre-built. Only customer-specific artifacts need parsing. Time-to-value drops from months to weeks.

Sprint 1 for customer N+1 delivers what took 3 sprints for customer 1.

Part Two

Technical Deep Dive

Parser architecture, knowledge graph ontology, data flow pipelines, and migration module internals.

Architecture

Parser Architecture

KMFlow's parser factory was extended with 7 new COBOL-specific parsers that plug into the existing 16-parser framework.

KMFlow Parser Factory

Parser Factory
register() / create() / parse()
↓

Existing Parsers (16)

TypeScript
Python
JavaScript
Java
Go
Rust
C#
SQL
YAML
JSON
Markdown
HTML
CSS
Shell
Docker
Terraform

New COBOL Parsers (7)

COBOL Source
DB2 Schema
CICS Resource
JCL
BMS Map
IMS Definition
Scheduler

Parser Detail Cards

CobolSourceParser

.cbl.cob.cpy.asm
PROCESSENTITYTEXTRELTABLE

94% coverage

Db2SchemaParser

.ddl.dcl.ctl
ENTITYRELTABLE

98% coverage

CicsResourceParser

.csd
PROCESSENTITYREL

95% coverage

JclParser

.jcl.prc
PROCESSENTITYRELTEXT

97% coverage

BmsMapParser

.bms
PROCESSTABLEENTITYTEXT

96% coverage

ImsDefinitionParser

.dbd.psb
ENTITYREL

97% coverage

SchedulerParser

.ca7.controlm
PROCESSRELTABLE

97% coverage

Full Coverage Table

ParserExtensionsFragment TypesCoverage
CobolSourceParser .cbl .cob .cpy .asm PROCESS, ENTITY, TEXT, REL, TABLE 94%
Db2SchemaParser .ddl .dcl .ctl ENTITY, REL, TABLE 98%
CicsResourceParser .csd PROCESS, ENTITY, REL 95%
JclParser .jcl .prc PROCESS, ENTITY, REL, TEXT 97%
BmsMapParser .bms PROCESS, TABLE, ENTITY, TEXT 96%
ImsDefinitionParser .dbd .psb ENTITY, REL 97%
SchedulerParser .ca7 .controlm PROCESS, REL, TABLE 97%
Ontology

Knowledge Graph & Ontology

The knowledge graph captures all mainframe artifacts and their relationships in a queryable Neo4j graph. Five node types form the ontology core.

Process (Programs, Jobs, Transactions)
DataObject (VSAM, DB2, IMS)
InterfaceContract (COMMAREA)
Activity (CICS, SQL, DL/I)
BusinessRule (EVALUATE, 88-level)
Scheduler (CA-7, Control-M)

Process

COBOL programs, batch jobs, CICS transactions. Every executable unit in the mainframe ecosystem.

  • COBOL programs (.cbl, .cob)
  • JCL job steps
  • CICS transaction IDs

DataObject

VSAM files, DB2 tables, IMS segments. Every persistent data store the mainframe reads or writes.

  • VSAM KSDS/RRDS/ESDS files
  • DB2 tables and views
  • IMS database segments

InterfaceContract NEW

COMMAREA and LINKAGE SECTION structures that define inter-program contracts. The missing API catalog.

  • COMMAREA field layouts
  • LINKAGE SECTION definitions
  • Parameter passing conventions

Activity

CICS commands, SQL operations, IMS DL/I calls. Every action a program performs against an external resource.

  • EXEC CICS SEND/RECEIVE/READ/WRITE
  • EXEC SQL SELECT/INSERT/UPDATE
  • CBLTDLI calls (IMS)

BusinessRule

EVALUATE blocks, 88-level conditions, decision tables. The encoded business logic that governs program behavior.

  • EVALUATE WHEN decision trees
  • 88-level condition names
  • PERFORM UNTIL iteration rules

4-Phase Ingestion Pipeline

Phase 1: Foundations

Copybooks, DDL, DBD, BMS

Phase 2: Programs

COBOL, DCLGEN, PSB

Phase 3: Orchestration

JCL, CSD, ASM

Phase 4: Scheduling

CA-7, Control-M

Pipeline

Data Flow Pipeline

Tracing a single COBOL program through the entire pipeline, from source to migration output.

COBIL00C.cbl (COBOL Source)
AWS CardDemo — Bill Payment Transaction
↓
CobolSourceParser
Parse IDENTIFICATION, DATA, and PROCEDURE divisions
↓
Fragment Extraction
PROCESS_ELEMENT: Program COBIL00C
ENTITY: WS-VARIABLES (01-level group items)
ENTITY: ERR-FLG-ON (88-level business rule)
RELATIONSHIP: COPY COCOM01Y (INCLUDES)
RELATIONSHIP: EXEC CICS SEND (EXECUTES)
RELATIONSHIP: CALL 'CSUTLDTC' (CALLS)
TABLE: EVALUATE decision table
↓
Knowledge Graph (Neo4j)
Process: COBIL00C
CALLS → CSUTLDTC
INCLUDES → COCOM01Y, COBIL00, CVACT01Y
EXECUTES → CICS:SEND, CICS:RECEIVE, CICS:READ
HAS_RULE → ERR-FLG decision table
↓
Migration Module
SchemaDeriver → PostgreSQL DDL
ArchitectureRecommender → Service boundaries
CodegenOrchestrator → FastAPI stubs + React components
Migration

Migration Module

Schema Derivation

Type mapping from mainframe to PostgreSQL with full fidelity.

COMP-3 (packed)→NUMERIC(p,s)
PIC X(n)→VARCHAR(n)
PIC 9(n)→INTEGER
PIC S9(n)V9(m)→NUMERIC(n+m,m)
COMP / BINARY→BIGINT
VSAM KSDS→TABLE + PK index

Architecture Recommender

Pattern mapping from mainframe constructs to cloud-native equivalents.

CICS Transaction→REST Endpoint
JCL Batch Job→Async Worker
BMS Screen→React Component
VSAM File→PostgreSQL Table
IMS Database→PostgreSQL + JSONB
COMMAREA→OpenAPI Contract

Codegen Orchestrator

Multi-output code generation from knowledge graph queries.

  • OpenAPI 3.1 specs from interface contracts
  • Alembic migration scripts from schema derivation
  • FastAPI route stubs from CICS transaction mapping
  • React form components from BMS map definitions
  • pytest-bdd feature files from business rules
  • AsyncIO workers from JCL batch flows
Roadmap

Epic Roadmap

8 epics spanning 9 sprints. COB-1 and COB-2 completed in Sprint 1. COB-3 is next.

S1
S2
S3
S4
S5
S6
S7
S8
S9
DONE COB-1 Parsers
DONE COB-2 Ontology
NEXT COB-3 Ingestion
PLAN COB-4 Migration
PLAN COB-5 Codegen
PLAN COB-6 BPMN
PLAN COB-7 Testing
PLAN COB-8 Pilot

Epic Dependencies

COB-1
Parsers
COB-2
Ontology
COB-3
Ingestion
COB-4
Migration
COB-5
Codegen
COB-6
BPMN
COB-7
Testing
COB-8
Pilot
Platform

Platform Onboarding

The COBOL Migration customer (COB project) was onboarded through the standard Blaze provisioning wizard. Five steps from zero to a fully configured workspace.

Step 1: Industry and Compliance selection Step 2: Subscription tier selection Step 3: Integration configuration Step 4: Review and confirm Step 5: Provisioning started

Scroll horizontally to see all onboarding steps. Each screenshot shows a stage of the multi-tenant provisioning wizard.

Deliverables

Deliverables & Next Steps

Files Delivered

CategoryFilesLines
Parser implementations7~3,200
Unit test suites7~4,100
Test fixtures (CardDemo)35+~2,800
Ontology module4~1,500
Migration module5~1,800
Domain vocabulary1252 terms
Configuration & setup6~400

Total: ~65 files, ~14,000+ lines of code and tests

Next Steps

ActionOwnerTarget
Ingest full CardDemo corpus into Neo4jEngineeringSprint 2
Validate knowledge graph completenessEngineeringSprint 2
Generate architecture blueprintEngineeringSprint 3
PostgreSQL schema generation from VSAMEngineeringSprint 4
OpenAPI contracts from COMMAREAEngineeringSprint 4
Customer demo with populated graphDeliverySprint 3
Production pilot planningManagementSprint 6