1.6Element 4: Coverage Goals

Without explicit coverage goals, the LLM defaults to happy-path tests. This is perhaps the most critical element to get right, because it directly determines whether your test suite catches bugs or merely occupies disk space.

Explicit coverage requirements:

**Coverage requirements:**
- Every acceptance criterion must have at least one test
- Include at least 2 negative/error cases per endpoint
- Include 1 boundary value test for every numeric field
- Test auth: valid token, expired token, missing token, wrong role
- Test idempotency: calling the endpoint twice with the same cart_id
- Test concurrent access: two users modifying the same resource
- Test pagination: first page, last page, empty page, page beyond total

Coverage categories to consider:

Category	Examples	Why AI Misses It
Happy path	Valid input, successful response	AI covers this well
Validation errors	Missing fields, wrong types, out-of-range	AI covers partially
Auth failures	No token, expired token, wrong role	AI often omits role-based tests
Resource conflicts	Duplicate key, stock conflict, race condition	AI rarely considers
Boundary values	Min, max, min-1, max+1 for every numeric field	AI covers if prompted
State transitions	Valid and invalid state changes	AI almost never considers
Pagination	Edge cases around page boundaries	AI rarely generates
Idempotency	Repeat requests, duplicate submissions	AI almost never considers
Concurrency	Parallel modifications, optimistic locking	AI cannot generate well

The negative-to-positive ratio

A good test suite has approximately a 2:1 ratio of negative (error) tests to positive (success) tests. This is because there are more ways for things to go wrong than right. If the AI generates 10 tests and 8 are happy-path, your prompt needs stronger coverage goals.

**Coverage ratio guidance:**
For each endpoint:
- 2-3 happy path tests (valid variations)
- 4-6 validation error tests (missing fields, wrong types, boundaries)
- 2-3 auth error tests (no token, wrong role, expired)
- 1-2 conflict/state tests (duplicate, out-of-stock)
- 1-2 edge case tests (empty arrays, maximum sizes, special characters)