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

Scenario	Priority
Happy path (expected behavior)	Must have
Edge cases (null, empty, boundary)	Must have
Error paths (exceptions, failures)	Must have
Invalid inputs	Must have
Performance-critical paths	Nice to have

---

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