Testing

Testing

Core Philosophy

AI-generated code is untrusted until it passes a human-reviewed test suite. Tests are not optional — they are the safety net that makes refactoring and agentic code generation safe. Treat test code with the same quality standards as production code.

The Verify Output Principle: Every piece of AI-generated code should pass through an AI-generated but human-reviewed test suite before being considered complete.

Test Pyramid

        /\
       /E2E\          ← Few, slow, high confidence
      /------\
     /Integr. \       ← Some, medium speed
    /------------\
   /   Unit Tests  \  ← Many, fast, isolated
  /------------------\

• Unit tests — 70% of your suite. Fast, isolated, no external dependencies.
• Integration tests — 20%. Test how components work together.
• E2E tests — 10%. Test real user flows from browser to DB.

Unit Testing

Principles

• Test behavior, not implementation details
• One assertion concept per test
• Use the AAA pattern: Arrange → Act → Assert
• Mock all external dependencies (APIs, databases, file system)
• Test edge cases: null, empty, boundary values, error paths

Structure

describe('ComponentOrFunction', () => {
  describe('methodName', () => {
    it('should [expected behavior] when [condition]', () => {
      // Arrange
      const input = ...

      // Act
      const result = functionUnderTest(input)

      // Assert
      expect(result).toBe(expectedValue)
    })
  })
})

What to Test

Integration Testing

Integration Principles

• Test the contract between components, not their internals
• Use a real (but isolated) database — not mocks
• Clean up test data after each test
• Test API endpoints with realistic payloads

Database Integration Tests

# Python example
def test_create_user_persists_to_db(db_session):
    # Arrange
    user_data = {"name": "James", "email": "james@test.com"}

    # Act
    user = create_user(db_session, user_data)

    # Assert
    saved = db_session.query(User).filter_by(id=user.id).first()
    assert saved.email == "james@test.com"

API Integration Tests

• Test request/response contracts
• Test authentication and authorization flows
• Test error responses (400, 401, 403, 404, 500)
• Test pagination and filtering

End-to-End (E2E) Testing

Tool Recommendation

Use Playwright for E2E testing — it supports all major browsers, has auto-wait built in, and works well with AI agent verification.

Selector Priority (Most to Least Robust)

1. getByRole() — ARIA roles (most resilient)
2. getByText() — visible text
3. getByTestId() — data-testid attributes
4. getByLabel() — form labels
5. CSS selectors — (avoid, brittle)

E2E Test Structure

test('user can complete checkout flow', async ({ page }) => {
  // Navigate
  await page.goto('/products')

  // Interact
  await page.getByRole('button', { name: 'Add to Cart' }).first().click()
  await page.getByRole('link', { name: 'Checkout' }).click()

  // Fill form
  await page.getByLabel('Email').fill('test@example.com')

  // Assert
  await expect(page.getByText('Order confirmed')).toBeVisible()
})

What to Cover with E2E

• Critical user journeys (signup, login, core feature flows)
• Payment or checkout flows
• Authentication flows (login, logout, password reset)
• Error states visible to users

Mocking & Test Doubles

Mock External APIs

Always mock external API calls in unit/integration tests:

jest.mock('../services/paymentService', () => ({
  chargeCard: jest.fn().mockResolvedValue({ success: true })
}))

Test Coverage Guidelines

Coverage doesn't equal quality. 95% coverage with weak assertions is worse than 80% coverage with strong assertions.

Testing AI-Generated Code

1. Generate tests before or alongside the code — don't test after
2. Review AI-generated tests critically — they often miss edge cases
3. Add tests for every bug found — regression coverage
4. Test the behavior described in the spec, not just what the code does
5. Mutation testing — use tools like Stryker to verify tests actually catch bugs

Framework Quick Reference