An Anti-Corruption Layer separates two systems that must communicate but do not share the same model. It accepts concepts from one side, translates them, and calls the other side in its own terms. This keeps the new component from spreading legacy field names, protocols, and business assumptions through its design.
What the layer translates
Translation can cover more than data formats. The systems may divide responsibilities differently, use different identifiers, or disagree about the sequence of a workflow. The layer makes those differences explicit at one boundary.
| Difference | Legacy side | Layer responsibility | New side |
|---|---|---|---|
| Data shape | Shared or irregular records | Map and validate fields | Domain-specific model |
| Protocol | File, queue, or old API | Adapt transport and errors | Current interface |
| Vocabulary | Historical names and codes | Translate meanings | Current terminology |
| Workflow | Multi-step legacy operation | Coordinate required calls | One defined capability |
The layer should translate between models. If it becomes the permanent home of unrelated business rules, ownership is harder to trace and the boundary becomes another legacy component.
How it supports staged replacement
The Strangler Fig Pattern often requires the old and new systems to coexist. A moved capability may still need data or behaviour that remains in the legacy application. The Anti-Corruption Layer provides the connection without making the new component call legacy internals directly.
The direction of translation matters. A new component can protect its model when it calls the legacy system. The legacy application may need a separate adapter when it calls a new service. Treating both directions as one undifferentiated interface can hide ownership and failure behaviour.
What an audit checks
A legacy codebase audit traces the contracts that the layer must preserve and the assumptions it should contain. That includes:
- Semantic differences. Fields or operations with similar names but different meanings.
- State ownership. Which system is authoritative for each record during coexistence.
- Failure mapping. How timeouts, missing data, and partial work appear on each side.
- Security boundaries. Which identities, permissions, and validation rules apply to the call.
- Retirement intent. Whether the layer remains after migration or disappears with the legacy system.
These checks prevent a translation boundary from being treated as a simple format converter when it also carries operational and domain meaning.
Operational consequences
The layer adds a component to the request or event path. Its logs and traces need to connect activity across both systems. Retries must account for operations that are not safe to repeat. Data mapping errors need a visible handling path rather than silent substitution.
Branch by Abstraction can introduce a related boundary inside one codebase, while an Anti-Corruption Layer commonly mediates between models or subsystems. The two can work together during migration, but they address different dependencies. One lets implementations share a contract. The other prevents one system's concepts from controlling the design of another. That distinction helps a migration plan state which boundary is temporary, which is part of the target architecture, and what evidence permits the old connection to be removed.
