A green test suite can still approve the wrong behavior. Semantic verification asks whether a system does what the business intended, not merely whether it passes mechanical checks such as types, tests, contracts, and schemas. Mechanical checks become cheaper when code generation becomes cheaper; assessing business meaning still requires judgment. section 1.5 section 1.8
Generating both code and tests with the same model can create a closed loop: shared blind spots may miss the same edge case twice. A model reviewing another model’s changes can widen the review surface, but it is not an independent second opinion. section 1.5 section 5.3.2
Feature: Refund policy
Rule: Time-bounded, single-order refunds
Example: Within 30 days is allowed
Given an order "O1" purchased at 2025-08-01T10:00:00Z
When a refund of 25.00 USD is requested at 2025-08-28T09:59:00Z
Then the refund is approved and remaining refundable equals 75.00 USD
Example: After 30 days is rejected
Given an order "O2" purchased at 2025-08-01T10:00:00Z
When a refund is requested at 2025-08-31T10:00:01Z
Then the request is rejected with reason "expired"
Example: Partials must not exceed the original
Given prior refunds of 90.00 USD on a 100.00 USD order
When a refund of 20.00 USD is requested
Then the request is rejected with reason "exceeds-remaining"The refund-policy examples turn business meaning into concrete checks: a time limit, an expired request, and a partial refund that exceeds the remaining amount. Their value depends on whether those rules actually express the intended policy. section 1.5
Put human attention where it scales: author and curate specifications, design adversarial and mutation-resistant test plans, and maintain shared, versioned assets such as policy packs and spec templates. Reading every generated line is not a reliable control at agent scale. Automated reviewers can help flag concerns, while deterministic checks serve as gates. section 1.5 section 5.3.2
