Principle: Write tests that give confidence

Test suite has high coverage numbers but doesn't catch real bugs. Tests are fragile, slow, or test the wrong things.

Focus on confidence per test, not coverage percentage:

The testing trophy (Kent C. Dodds):

     /\        End-to-end (few)
    /  \       - Happy path user flows
   /----\      - Critical business flows
  / Intg \     Integration (many)
 /--------\    - API endpoints with real DB
/ Unit     \   - Component rendering with events
/___________\  Unit (some)
               - Pure logic, utilities
               Static analysis (all)
               - TypeScript, ESLint

High-confidence tests:

# Tests behavior, not implementation
def test_user_can_place_order():
    user = create_user()
    product = create_product(price=10.00)
    order = place_order(user, product, quantity=2)
    
    assert order.total == 20.00
    assert order.status == 'confirmed'
    assert len(user.orders) == 1
    # Tests the whole flow, not internal methods

# Tests edge cases that matter
def test_order_with_zero_quantity_rejected():
    with pytest.raises(ValidationError):
        place_order(user, product, quantity=0)

Low-confidence tests (avoid):

# Tests implementation details
def test_internal_cache_updated():
    service._cache.clear()
    service.get_user(1)
    assert 1 in service._cache  # Who cares? Test the behavior

# Tests trivial code
def test_getter():
    user = User(name='Alice')
    assert user.name == 'Alice'  # Tests nothing useful

Ask: If this test fails, does it mean a user-facing bug exists?

100% coverage with low-confidence tests gives false security. 60% coverage with high-confidence tests catches more real bugs.

Designing test strategies for applications