> the scope of work is predetermined
That's bullshit/impossible for most software, and the exact opposite of being agile and catering to real needs, not imagined ones. That's the whole point of agile methodologies, embrace change.
2. Done is done - it's a binary yes/no thing. Done but not tested is not done. Done but needs documentation is not done. This varies from place to place, but code-done is not done, and coding is usually only 30-odd % of the work.
2b. If that's the case, then you're overcommitted. Next sprint, drop back to what you managed to get done this sprint. The idea is to get things repeatable and features turning out in a maintainable steady fashion.
In practice, it works - if there's some discipline from both the devs and the business. If stuff gets shoehorned in, or devs are pulled off to work on critical stuff, then it works about as well as you'd expect (ie. not very).
#2 - Tom Lasseter (Pixar) supposedly said "We release snapshots of our movies, they're never done."
#2 - I'd agree with you about "by sprint end" for unit tests and any other harnesses that devs can write. But real QA/test work? I just don't see it. When I've been in charge, the three areas (reqs, dev, test) are interleaved. Makes for nice feedback loops. If dev and test had their own sprints, with test trailing by a week or so, I could almost see that working.
#2b - I've never seen sprints finish nicely, where unfinished work (points) didn't carry over to the next sprint. Total fiction that anything gets "done".
With proper project management, imho, the scope of work is predetermined and things get dropped to meet a hard deadline, or the deadline is pushed. And the final acceptance testing (e.g. go/no-go process), without fail, takes two weeks. That's tacked on to the tail of when everyone has jointly decided that everything's "done". I've yet to see that kind of quality, rigor, professionalism using scrum.