Skip to content

Full CICS Estate Mapping & Byte-for-Byte Java Translation

Choose a tag to compare

@squid-protocol squid-protocol released this 28 Sep 16:21
· 2401 commits to main since this release
Immutable release. Only release title and notes can be modified.
6991970

We can now ingest and scan entire IBM CICS estates—mapping the transactional boundaries, UI (BMS) screens, and datasets—and pass them to our cobol-to-java generator.

• Ground Truth Validation: We have successfully locked in our Mainframe Ground Truth milestones. Our Java generated outputs are continually validated byte-for-byte against the original COBOL outputs via our Golden Crucible test suite.
• The Translation Reality Check (n = 2): While the byte-for-byte engine is proven (for happy path only - not exhaustive path options). Both required manual intervention to correct edge cases. We have not yet achieved a "one-shot" translation out of the box for a never seen before repo, but the feedback loop is rapidly closing.

🕸️ Network Maps: The 50% Recall Pattern

As part of our benchmarking, we started heavily measuring our network mapping metrics (Call Graphs and Import Maps) against industry-standard compilers and Tree-sitter. A very clear, encouraging pattern has emerged:

• Precision is near 100%. If GitGalaxy draws an edge or connection, it is mathematically correct. We do not hallucinate dependencies.
• Recall is currently ~50% for call graphs only, near 100% for import graphs. We are currently capturing about half of the connections that a full compiler sees.
• The Outlook: This is incredibly good news. Achieving 100% precision is the hardest problem in static analysis. Fixing the 50% recall gap is highly achievable—it simply requires expanding our regex dictionaries and structural signatures to catch the remaining dialect variants, which we are actively doing.

🗺️ What’s Next

• The "One-Shot" Goal: We will continue iterating on the cobol-to-java generator until a complex legacy CICS transaction translates, compiles, and deploys without a single human keystroke.
• Closing the Recall Gap: We will be rapidly deploying new structural signatures to push our Call Graph recall toward 100%.
• Language Documentation: We are currently expanding our dictionaries into newly scoped architectures (AS/400 RPG, Software AG Natural, Mark IV). Documentation for these new language integrations is pending and will follow in upcoming minor releases.