Test-Driven Development: London & Chicago Schools

<default_to_action> When implementing TDD or choosing testing style:

1. IDENTIFY code type: domain logic → Chicago, external deps → London
2. WRITE failing test first (Red phase)
3. IMPLEMENT minimal code to pass (Green phase)
4. REFACTOR while keeping tests green (Refactor phase)
5. REPEAT cycle for next functionality

Quick Style Selection:

• Pure functions/calculations → Chicago (real objects, state verification)
• Controllers/services with deps → London (mocks, interaction verification)
• Value objects → Chicago (test final state)
• API integrations → London (mock external services)
• Mix both in practice (London for controllers, Chicago for domain)

Critical Success Factors:

• Tests drive design, not just verify it
• Make tests fail first to ensure they test something
• Write minimal code - no features beyond what's tested </default_to_action>

Quick Reference Card

When to Use

• Starting new feature with test-first approach
• Refactoring legacy code with test coverage
• Teaching TDD practices to team
• Choosing between mocking vs real objects

TDD Cycle

School Comparison

Agent Coordination

• qe-test-generator: Generate tests in both schools
• qe-test-implementer: Implement minimal code (Green)
• qe-test-refactorer: Safe refactoring (Refactor)

Chicago School (State-Based)

Philosophy: Test observable behavior through public API. Keep tests close to consumer usage.

// State verification - test final outcome
describe('Order', () => {
  it('calculates total with tax', () => {
    const order = new Order();
    order.addItem(new Product('Widget', 10.00), 2);
    order.addItem(new Product('Gadget', 15.00), 1);

    expect(order.totalWithTax(0.10)).toBe(38.50);
  });
});

When Chicago Shines:

• Domain logic with clear state
• Algorithms and calculations
• Value objects (Money, Email)
• Simple collaborations
• Learning new domain

London School (Mock-Based)

Philosophy: Test each unit in isolation. Focus on how objects collaborate.

// Interaction verification - test method calls
describe('Order', () => {
  it('delegates tax calculation', () => {
    const taxCalculator = {
      calculateTax: jest.fn().mockReturnValue(3.50)
    };
    const order = new Order(taxCalculator);
    order.addItem({ price: 10 }, 2);

    order.totalWithTax();

    expect(taxCalculator.calculateTax).toHaveBeenCalledWith(20.00);
  });
});

When London Shines:

• External integrations (DB, APIs)
• Command patterns with side effects
• Complex workflows
• Slow operations (network, I/O)

Mixed Approach (Recommended)

// London for controller (external deps)
describe('OrderController', () => {
  it('creates order and sends confirmation', async () => {
    const orderService = { create: jest.fn().mockResolvedValue({ id: 123 }) };
    const emailService = { send: jest.fn() };

    const controller = new OrderController(orderService, emailService);
    await controller.placeOrder(orderData);

    expect(orderService.create).toHaveBeenCalledWith(orderData);
    expect(emailService.send).toHaveBeenCalled();
  });
});

// Chicago for domain logic
describe('OrderService', () => {
  it('applies discount when threshold met', () => {
    const service = new OrderService();
    const order = service.create({ items: [...], total: 150 });

    expect(order.discount).toBe(15); // 10% off > $100
  });
});

Common Pitfalls

❌ Over-Mocking (London)

// BAD - mocking everything
const product = { getName: jest.fn(), getPrice: jest.fn() };

Better: Only mock external dependencies.

❌ Mocking Internals

// BAD - testing private methods
expect(order._calculateSubtotal).toHaveBeenCalled();

Better: Test public behavior only.

❌ Test Pain = Design Pain

• Need many mocks? → Too many dependencies
• Hard to set up? → Constructor does too much
• Can't test without database? → Coupling issue

Agent-Assisted TDD

// Agent generates tests in both schools
await Task("Generate Tests", {
  style: 'chicago',      // or 'london'
  target: 'src/domain/Order.ts',
  focus: 'state-verification'  // or 'collaboration-patterns'
}, "qe-test-generator");

// Agent-human ping-pong TDD
// Human writes test concept
const testIdea = "Order applies 10% discount when total > $100";

// Agent generates formal failing test (Red)
await Task("Create Failing Test", testIdea, "qe-test-generator");

// Human writes minimal code (Green)

// Agent suggests refactorings
await Task("Suggest Refactorings", { preserveTests: true }, "qe-test-refactorer");

Agent Coordination Hints

Memory Namespace

aqe/tdd/
├── test-plan/*        - TDD session plans
├── red-phase/*        - Failing tests generated
├── green-phase/*      - Implementation code
└── refactor-phase/*   - Refactoring suggestions

Fleet Coordination

const tddFleet = await FleetManager.coordinate({
  workflow: 'red-green-refactor',
  agents: {
    testGenerator: 'qe-test-generator',
    testExecutor: 'qe-test-executor',
    qualityAnalyzer: 'qe-quality-analyzer'
  },
  mode: 'sequential'
});

Related Skills

• agentic-quality-engineering - TDD with agent coordination
• refactoring-patterns - Refactor phase techniques
• api-testing-patterns - London school for API testing

Remember

Chicago: Test state, use real objects, refactor freely London: Test interactions, mock dependencies, design interfaces first Both: Write the test first, make it pass, refactor

Neither is "right." Choose based on context. Mix as needed. Goal: well-designed, tested code.

With Agents: Agents excel at generating tests, validating green phase, and suggesting refactorings. Use agents to maintain TDD discipline while humans focus on design decisions.

Gotchas

• Agent skips Red phase and writes test + implementation together — enforce "test must fail first" by running test before writing code
• London school over-mocking creates brittle tests that break on any refactor — mock at architectural boundaries, not every function
• Chicago school tests become slow as integration scope grows — keep test boundaries tight
• Agent defaults to jest.mock() for everything — prefer dependency injection for testability
• Refactor phase is where agent cuts corners most — verify no behavior changes by checking test output is identical