Model-First Legacy Migration

Legacy migration is usually framed as translation from one implementation to another. That framing leaves the most consequential uncertainty untouched: no one knows which observed behaviors are intended obligations, which are relied-upon accidents, which are defects, and which depend on an environment that the replacement does not reproduce. This paper develops a model-first alternative. A bounded behavioral model is extracted from exact source custody, static structure, observations, operational knowledge, policy, and downstream dependencies; it is then adjudicated before it may become normative. Candidate realizations are compared against both the model and the original system within declared envelopes, equivalence relations, observers, and claim ceilings. The proposal does not solve general program equivalence or infer intention from behavior alone. It replaces an unaudited rewrite with a governed chain of explicit models, gaps, assumptions, tests, promotion decisions, rollback paths, and revision records. Its strongest capability is not automatic modernization. It is the conversion of hidden dependencies into inspectable and contestable migration obligations while preserving a disciplined refusal to claim beyond observed and justified evidence.