Skip to content

Test Pyramid vs. Trophy

When to Use

Use the Pyramid when most bugs live in isolated logic. Use the Trophy when most bugs appear in component interactions. Neither is universally correct.

Decision

If your stack looks like... Lean toward... Rationale
Business logic-heavy backend (rules, financial, permissions) Pyramid — many unit tests Logic is verifiable in isolation; unit tests are cheap and comprehensive
Modern JS/TS frontend, React, Next.js Trophy — mostly integration Components are only meaningful in context; RTL + Vitest make integration tests fast
Drupal (PHP modules, services, forms) Pyramid with Kernel tests in the middle PHPUnit Unit → Kernel → Functional matches the Drupal test type hierarchy
Distributed microservices / APIs Pyramid + contract tests at boundaries Service boundaries need contract coverage; unit tests handle internal logic
UI-heavy app (CMS, admin tools) Trophy + selective VR Interactions matter more than units; visual regression catches regressions unit tests miss

Pattern

Test Pyramid (Mike Cohn, ~2009): 

         [E2E]         ← few, slow, high confidence
        [INTEG.]       ← some, medium speed
    [UNIT / UNIT / UNIT] ← many, fast, low individual confidence
      [STATIC ANALYSIS]  ← free — linters, type checkers

Testing Trophy (Kent C. Dodds, 2018): 

      [E2E]            ← some (more than pyramid)
    [INTEGRATION]      ← MOST — highest ROI tier
  [UNIT UNIT UNIT]     ← some (not majority)
  [STATIC ANALYSIS]    ← free foundation

The right metric: what level of test catches the bugs that actually occur in your stack?

Common Mistakes

  • Wrong: Copying the Pyramid shape blindly for all projects → Right: Evaluate where bugs actually occur in your stack
  • Wrong: 100% unit, 0% integration → Right: High unit confidence + zero confidence in whether they work together is a false positive
  • Wrong: One giant E2E suite instead of a balanced shape → Right: CI takes 30+ minutes; developers skip it
  • Wrong: "More tests = better" → Right: Shape matters as much as count

See Also