Hacker News Reader

The cool thing is that hypothesis tries to build minimal examples that make your test fail.

The magic is in the "strategies"? How does it do that? And does it have a memory between test runs?