"Sleep at least n" is too broad of a spec to be useful. It basically guarantees that users of that function will come to expect or rely on behavior that is not covered by that spec, such as expecting that their program will be woken up again in a relatively short period of time rather than next week, or that the program will actually sleep for a non-trivial amount of wall time instead of immediately waking back up. Microsoft deserves at least some of the blame when application programmers come to rely on behavior that isn't officially part of the spec, and Microsoft definitely deserves some of the blame when they change that behavior without even doing those programmers the courtesy of documenting the change.

"Windows 10" is too much of a moving target, and it's about time Microsoft stopped trying to pretend to the public that it's all one operating system. If they're not going to usefully document potentially breaking changes, they should at least do us all the courtesy of bundling these changes up into a release that increments the major version number, and gives us the option of not upgrading for a few years instead of having the software as a service model forced upon us.

(Given what I've written above, it probably won't surprise you to find out I recently spent a decent chunk of time trying to find a workaround for a different undocumented change brought by Windows 10 version 2004).