The Repository Pattern

The repository pattern separates data access from business logic. Services talk to a Repo interface — not directly to SQLite, files, or any specific backend.

The payoff:

• Swap in-memory storage for a real database without touching your service
• Test your service without a database at all
• One interface, many implementations

Why care? This is how the cohort's expense classifier stays testable while supporting both in-memory (tests) and SQLite (production) backends.

Task

Expense and the ExpenseRepo Protocol are already defined. Implement InMemoryExpenseRepo with full CRUD:

• add(expense) — store and return the expense
• get(expense_id) — return the expense or None
• list_all() — return all stored expenses
• delete(expense_id) — remove it, return True if found, False if not