A testing framework that rejects invalid tests before they're used, preventing bad feedback from corrupting the learning process.