deliberation-tester

Test-driven development for AI deliberation starts with writing a failing unit test that specifies the expected behavior of your feature—such as a new adapter or convergence detector—before writing implementation code. This red-green-refactor workflow ensures your feature meets requirements and integrates correctly with voting and transcript systems.

Mock adapters and HTTP responses using pytest fixtures and mock libraries to test adapter logic without invoking real external services. This approach isolates unit tests, accelerates test execution, and allows you to simulate failures and edge cases in your CLI and HTTP adapter implementations.

Organize tests into three layers: unit tests for individual components like adapters and decision graph nodes, integration tests for real adapter invocations and multi-component workflows, and end-to-end tests for complete deliberation cycles. This structure ensures coverage across voting, transcript, and convergence detection logic.

Use VCR cassettes to record and replay HTTP responses during testing. This technique captures real adapter responses once, then replays them in subsequent test runs, ensuring consistent and reproducible integration tests without depending on external LLM services or network availability.

Writing failing tests first clarifies expected behavior, prevents implementation rework, and catches bugs early. For adapters serving voting systems or transcript handling, test-first development ensures reliability, reduces debugging cycles, and validates that new adapters integrate correctly with existing deliberation components.