Skip to content

Test Pyramid, Trophy and Core Principles

Test Pyramid

        ▲
       /E2E\       ← few, slow, expensive
      /------\
     /  API   \    ← moderate, medium speed
    /----------\
   /   Unit     \  ← many, fast, cheap
  /______________\
Layer Count Speed Confidence
Unit 70–80 % < 1 ms Logic correctness
API / Integration 15–20 % ms–s Wiring correctness
E2E 5–10 % s–min Business flow

Trade-offs: Speed vs Coverage

Approach Speed Coverage
All unit Very fast Low integration coverage
All E2E Very slow High but brittle
Pyramid Balanced Comprehensive

Test Trophy (Kent C. Dodds)

      [E2E]
   [Integration]   ← largest slice
  [Unit] [Static]

Trophy prioritises integration tests over unit tests. Useful when: - Business logic lives in orchestration rather than pure functions - UI + API interaction is the product's core value

Core Test Design Principles

DRY — Don't Repeat Yourself

Problem Solution
Duplicated setup blocks Shared fixtures and helpers
Repeated request builders Reusable factory functions
Copy-pasted assertions Custom assertion helpers

Rule: if test setup appears more than twice, extract it.

KISS — Keep It Simple, Stupid

  • Test one behaviour per test case
  • Avoid conditional logic inside tests
  • Prefer explicit over clever

Bad: if condition: assert A else: assert B Good: two separate test cases with clear names.

Isolation

Rule Reason
Tests must not share state Order-dependent failures
Each test owns its data Parallel execution safety
External side-effects cleaned up No pollution between runs

Isolation strategies: per-test DB transaction rollback, in-memory fakes, test-scoped DI containers.

Determinism

  • Same input → same output on every run
  • No time.now() calls inside test logic (inject clock)
  • No random seeds without explicit fixture
  • No sleep() — use event-driven waits or polling with timeout

Readability

Structure every test with Arrange / Act / Assert (AAA):

# Arrange
user = build_user(role="admin")
client = AuthenticatedClient(user)

# Act
response = client.get("/api/resource/1")

# Assert
assert response.status_code == 200
assert response.json()["id"] == 1

Test names: test_<subject>_<scenario>_<expected_outcome>

Maintainability

Risk Mitigation
Tests break on every refactor Test behaviour, not implementation
Fragile locators in UI tests Stable data-testid attributes
Tests tightly coupled to schema Schema-validated responses, not field-by-field
Long setup blocks Extract to fixture factories