Testing

Testing Pyramid

• 70% Unit Tests — fast, isolated, test business logic
• 20% Integration Tests — API endpoints + database
• 10% E2E Tests — critical user flows

Quick Start

cd backend
pytest                    # Run all tests
pytest -v                 # Verbose output
pytest -m unit            # Only unit tests
pytest --cov=app          # With coverage

Database Testing

We use SQLite for local testing (fast, no Docker dependency) with a PortableJSON type adapter. This is a pragmatic trade-off:

Aspect	SQLite (Current)	PostgreSQL (Better)
Speed	2 seconds	10-30 seconds
Setup	None	Docker required
Accuracy	80%	100% matches production

Migration path: Testcontainers with real PostgreSQL when the project matures.

FastAPI Testing Pattern

# Use dependency_overrides, NOT unittest.mock.patch
app.dependency_overrides[get_db] = _override_get_db
app.dependency_overrides[get_current_active_user] = _override_auth

yield TestClient(app)
app.dependency_overrides.clear()

Model Evals

Systematic tests for AI model outputs — accuracy, hallucination rate, cost, latency. Custom eval framework with:

• Pydantic-typed eval datasets
• Domain-specific metrics (calorie error, protein error)
• Multi-model comparison (OpenAI vs Claude vs Gemini)
• CI/CD integration (trigger with [evals] in commit message)

See the Evals page for the full guide.

CI/CD

Tests run on every push/PR via GitHub Actions:

1. Backend pytest (Python 3.9, 3.10, 3.11)
2. Frontend typecheck + lint
3. Model evals (main branch only)
4. Build check

Current coverage threshold: 30% (goal: 80%).