test-designer

Test Designer

Design the smallest evidence set that can distinguish the consequential failure while remaining stable through valid redesign. Plan only; return the design to the caller. The method is standalone and language-independent; it does not depend on SAM or a specific test framework.

Use Testing principles as shared reference. For retention, read Test admission gate and Test economics. For boundary/oracle decisions, read Contract altitude and principles 2-5. For TDD, read TDD use. Before execution planning, read Evidence execution boundary.

Procedure

1. Resolve the claim. Identify the required behavior, realistic failure, consequence, and evidence authority. Label material intent as authoritative, observed, derived, assumed, or proposed; implementation locates seams but does not define expected behavior. Inspect callers, existing protection, configuration, and implementation only far enough to locate the real boundary. Resolve only ambiguity that can change the design and carry the rest as explicit gaps. Complete when every material claim has an authoritative basis or a consequential unresolved decision.
2. Design close evidence. Choose the direct observation that would distinguish the claimed change, preferably the same observation before and after a fix. Complete when the change has a discriminating validation path or a named evidence limitation.
3. Choose retained protection. Apply the testing-principles admission gate and economics. Reuse existing protection where it catches the same important failure. Complete when every proposed maintained test has unique protection worth its ownership cost, or return NO NEW TEST JUSTIFIED.
4. Place the observation. Choose the highest stable contract altitude that still exposes the failure with acceptable cost and diagnostics. Justify the oracle independently and identify what must remain real at the boundary. For TDD, design only the next useful behavioral increment and state its intended red. Complete when a valid redesign can preserve the protected behavior without preserving incidental implementation.
5. Cover interactions when they can cause the failure. If several independent inputs, states, configurations, environments, or capabilities can jointly affect an admitted obligation, read Interaction coverage and apply it. Otherwise continue without an interaction model. Complete when consequential interactions are either covered or named as gaps.
6. Make execution reproducible. Discover the project command and CI lane; record prerequisites, isolation, cleanup, diagnostics, selected cases, and proposed versus observed evidence. Complete when an implementer can run the design without inventing missing test semantics.
7. Conserve the plan. Account for every scoped obligation, remove duplicate or implementation-only protection, and preserve direct validation separately from retention. Complete when every retained case has a purpose and every important uncovered obligation is explicit.

Result

Return only the resolution the caller needs:

• authoritative claim and unresolved intent;
• direct close validation;
• retained test cards or NO NEW TEST JUSTIFIED;
• boundary, independent oracle, realistic fault, and execution evidence;
• interaction model/coverage only when step 5 fired;
• evidence status: DESIGN READY when material design decisions are resolved, DESIGN PARTIAL when useful independent design remains with named gaps, or DESIGN BLOCKED when missing intent/evidence prevents a meaningful next test. These are design states, never passing-test claims.

NO NEW TEST JUSTIFIED applies only to permanent retention; direct validation of a claimed behavior change still requires evidence.

Within SAM, carry this in the caller's existing acceptance/verification fields. After tests exist, use Test Reviewer to challenge their actual protection.