Differential verification for COBOL-to-Java migration.
Strust v1 runs a COBOL reference and Java candidate under an explicit test contract, compares their observable behavior, and retains the evidence needed to reproduce or challenge the verdict.
The question
Did the candidate preserve the behavior that matters?
Compilation success and source similarity cannot answer that question. A migration can still change a rounding boundary, fixed-width field, collation order, error path, output artifact, or persistent side effect.
Strust fixes the workload, source revisions, runtime settings, fixtures, adapters, normalization policy, comparator, resource limits, and observation scope before executing either system. It then returns MATCH, DIVERGE, or INCONCLUSIVE instead of forcing every run into a misleading pass/fail result.
Published evaluation
The result includes the failures.
Five cases were designed to agree, four to diverge, and one to agree after EBCDIC normalization. The demonstrator classified eight correctly. The paper reports the two misses because a trust system should expose its own blind spots.
Accuracy
8 / 10
Eight controlled scenarios classified correctly
Mismatch precision
0.75
Three true divergences and one false positive
Mismatch recall
0.75
Three divergences found and one missed
Mismatch F1
0.75
Balanced diagnostic result for the mismatch class
Known limits
What v1 does not establish.
1.Ten manually prepared snippets are a feasibility study, not a production benchmark or scalability result.
2.The v1 demonstrator does not yet implement the proposed dependency graph, slice planner, or production mainframe adapters.
3.GnuCOBOL does not establish compatibility with every Enterprise COBOL dialect or z/OS subsystem.
4.Finite differential testing establishes agreement only for the tested workload and declared contract; it cannot prove universal equivalence.
Put the method against a real workflow
Bring Strust one migration worth verifying.
We are looking for engineering teams, modernization firms, and coding-agent builders willing to map one important software change into a design partnership.