For teams migrating legacy COBOL batch systems

Private reference implementation, unpublished

No Public Proof Yet That Rust Matches COBOL Arithmetic

I am building a Rust reference in my own private repo, unpublished, for COBOL PICTURE-clause arithmetic, REDEFINES overlaps and PERFORM THRU control flow, with no client system behind it.

Hire me - remote W2Resume PDF

No number is claimed, and the cost is that behavioral equivalence stays unproven until a public run produces one; The repo is private, so no CI status is public.

Limit: behavioral equivalence stays unproven until a public run produces a result.

The repo is private, so no CI status is public. No throughput number, test count, or CVE count is claimed until the numbers come straight from a public run.

This is a private, unpublished reference implementation for a specific class of legacy-modernization problem: COBOL batch-processing systems built on PICTURE-clause fixed-point arithmetic, implicit REDEFINES memory overlaps, and PERFORM THRU control flow that no single document ever fully describes. The private repo holds the migration pattern, with no client story.

The Problem Class

Three COBOL constructs resist a naive line-for-line rewrite, and each needs hand-ported arithmetic, verified by replay:

• PICTURE clause arithmetic. Every PIC V9 declaration has to be re-derived in Rust using checked arithmetic with explicit overflow guards, so no silent truncation path can exist in the rewrite.
• REDEFINES memory overlap. COBOL REDEFINES clauses share memory regions across incompatible types. Each overlap has to be mapped and replaced with Rust enum variants that carry explicit discriminants.
• PERFORM THRU control flow. Implicit fall-through PERFORM chains translate to explicit Rust state machines, with every execution path enumerated.

The Replay Design

The proof mechanism the repo is built around is deterministic replay: feed the COBOL and Rust implementations the same input record, and diff the output byte-for-byte. Behavioral equivalence is meant to be binary. That mechanism is designed into the repo. No passing run is claimed here, only the plan, which is not public until the repo is published.

Status

The repo is private, so no CI status is public. No throughput, latency, or test-count number is published for this repo, and none will be until a public run produces one.