Golden Tests

Prove the Implementation Matches the Specification

Generating software from a formal specification creates a critical requirement:

How do we know the resulting system actually reflects the domain?

Golden Tests provide a reference point.

The Specification as the Expected Result

A golden test defines what the implementation should represent.

The expected entities.

The expected rules.

The expected operations.

The expected transitions.

The expected policies.

The expected events.

The expected behavior.

The implementation can then be evaluated against that reference.

From Requirements to Verification

Traditional tests often validate isolated pieces of implementation.

Golden tests can validate the relationship between the domain specification and the resulting software.

This makes them particularly valuable in AI-generated and specification-driven development.

Protecting the Domain During Evolution

As specifications evolve, their corresponding golden tests can evolve with them.

This creates a continuous verification loop:

Specification → Generation → Implementation → Verification

If the implementation diverges from the specification, the difference becomes visible.

Trust Through Traceability

The objective is not simply to have more tests.

It is to establish a traceable relationship between:

What the domain says should exist

and

What the software actually implements.

That relationship becomes increasingly important as AI takes a larger role in software generation.

The specification defines what should exist. Golden Tests verify that it does.