How reliable is it? Have you encountered any tests that did not have the same result as on real Windows?

Yeah that makes me nervous. At the very least just use a VM to test your production artifacts before deployment. Must not be a very important product they're releasing.

It has never actually broken (or at least, no one has reported that if it did). But also it's an open source project, we provide the exe's best effort, if it's important you can compile it yourself. I have zero interest in using Windows even for one second if I can help it.