Testing Pyramid Reference
Target Agents
manager-develop - applies patterns during test creation, coverage analysis, and RED-GREEN-REFACTOR cycles
Test Pyramid Ratios
/ E2E \ 10% — Critical user journeys only
/----------\
/ Integration \ 20% — API endpoints, DB queries, service boundaries
/----------------\
/ Unit Tests \ 70% — Functions, hooks, utilities, pure logic
/--------------------\
| Level | Speed | Reliability | Maintenance | Coverage Target |
|---|
| Unit | Fast (<100ms) | High | Low | 70% of tests |
| Integration | Medium (1-5s) | Medium | Medium | 20% of tests |
| E2E | Slow (10-60s) | Lower | High | 10% of tests |
Coverage Targets by Context
| Context | Target | Rationale |
|---|
| Critical business logic | 95%+ | Revenue/security impact |
| API endpoints | 90%+ | Contract compliance |
| Utility functions | 85%+ | Reuse reliability |
| UI components | 80%+ | Rendering correctness |
| Configuration/glue code | 60%+ | Low complexity |
| Generated code | 0% | Don't test generated code |
Test Pattern: AAA (Arrange-Act-Assert)
// Arrange: Set up test data and preconditions
input := CreateTestUser("test@example.com")
// Act: Execute the function under test
result, err := service.CreateUser(ctx, input)
// Assert: Verify the outcome
assert.NoError(t, err)
assert.Equal(t, "test@example.com", result.Email)
Unit Test Patterns
| Pattern | When | Example |
|---|
| Table-Driven | Multiple input/output combinations | Go: tests := []struct{...} |
| Mock/Stub | External dependencies (DB, API) | Interface injection, mock frameworks |
| Snapshot | Complex output comparison | Jest snapshots, golden files |
| Property-Based | Mathematical properties | quickcheck, hypothesis |
| Boundary Value | Edge cases | 0, -1, MAX_INT, empty string, nil |
Integration Test Patterns
| Pattern | When | Example |
|---|
| Testcontainers | Real DB needed | Docker-based PostgreSQL for tests |
| HTTP Test Server | API endpoint testing | httptest.NewServer (Go), supertest (Node) |
| In-Memory DB | Fast DB tests | SQLite for development |
| Fixture Loading | Consistent test data | Factory functions, seed files |
What to Test vs What NOT to Test
ALWAYS Test
- Business logic and calculations
- Input validation and error handling
- Authentication and authorization flows
- Data transformations and mappings
- Edge cases and boundary conditions
- Race conditions (with -race flag in Go)
NEVER Test
- Framework internals (React rendering, Express routing)
- Third-party library behavior
- Simple getters/setters with no logic
- Private methods directly (test via public API)
- Generated code (protobuf, swagger)
- CSS styling and layout (use visual regression tools instead)
Test Quality Metrics
| Metric | Target | Tool |
|---|
| Line Coverage | 85%+ | go test -cover, istanbul, coverage.py |
| Branch Coverage | 75%+ | go test -covermode=count |
| Mutation Score | 70%+ | go-mutesting, Stryker |
| Test Execution Time | <2 min (unit), <10 min (all) | CI timer |
| Flaky Test Rate | <1% | CI history analysis |
Test File Conventions
| Language | Test File | Location |
|---|
| Go | *_test.go | Same package |
| TypeScript | *.test.ts / *.spec.ts | __tests__/ or co-located |
| Python | test_*.py | tests/ directory |
| Java | *Test.java | src/test/ mirror |
| Rust | #[cfg(test)] mod tests | Same file or tests/ |
TDD RED-GREEN-REFACTOR Quick Reference
RED: Write a failing test that defines expected behavior
GREEN: Write minimal code to make the test pass
REFACTOR: Clean up while keeping tests green
Rules:
- Never write production code without a failing test
- Write the smallest test that fails
- Write the simplest code that passes
- Refactor only when all tests are green
- One assertion per test (when practical)
Common Rationalizations
| Rationalization | Reality |
|---|
| "E2E tests cover everything, unit tests are redundant" | E2E tests are slow and flaky. Unit tests provide fast, precise feedback. The pyramid exists because each level serves a different purpose. |
| "Integration tests are more realistic than unit tests" | Realism comes at the cost of speed and isolation. A balanced pyramid gives both fast feedback and realistic validation. |
| "100% code coverage means the code is well tested" | Coverage measures execution, not correctness. A test that executes code without meaningful assertions provides zero value. |
| "Mocking is bad, I prefer real dependencies" | Real dependencies make tests slow and non-deterministic. Mock at boundaries, test business logic in isolation. |
| "This test is flaky, but it catches real bugs sometimes" | Flaky tests erode trust in the entire suite. Fix the flakiness or quarantine the test with a tracking issue. |
DAMP over DRY: Test code should be descriptive and self-contained. A reader should understand the test without reading shared fixtures or helper methods.
Red Flags
- Test pyramid inverted: more E2E tests than unit tests
- Unit tests depend on external services (databases, APIs, file systems)
- Test assertions check implementation details instead of behavior
- No integration tests between unit and E2E layers
- Flaky test present without a quarantine label or tracking issue
Verification
