There's still some disagreement on, essentially, how strict to make the homework policy, but more clear to me is that we have near-consensus that the only questions that should be closed are the genuinely bad ones--and we have similar ideas, most of us, on what constitutes bad.
Given that, I think it remains for us in private beta to finalize our first version of the FAQ, agree on tags, and seed lots more questions in each of the mentioned broad categories of question type we'd welcome (applied math/modeling, contest math, and more things on both ends - lower pre-college and closer to upper-division undergrad, including homework-type questions we think are fine). This last is perhaps the hardest to do, but will give the strongest signal on what we're about to those who would make quality contributions but aren't sure they're interested, given the homework etc. questions that will surface en masse.
I'd also ask that we put up more honest questions of the type we'd truly want to see, instead of focusing on taking on various personas to see how various questions should be dealt with. The public beta will provide lots of questions that will test our policies, with far more range than our poor imaginations and acting abilities have thus far handled. The questions we do want are the ones that won't take care of themselves.
The discussion on what's appropriate and how to deal with various problems will continue through public beta and beyond. If it's a mess at first, we'll probably mostly agree that it's a mess (we have more disagreement on whether it will be than what a mess looks like), and I think we have several people here willing to spend a fair bit of time monitoring and moderating the site in its very early stages--not that we have this kind of time to put into it in the long run, but more work now will make for less work--and the site's continuing existence and utility--later.
Those who are skeptical at how bad questions can get will see examples of what those who feel the need to warn mean, and will adjust their expectations accordingly--or they won't see those, and that won't happen, more's the better. Those who are worried will have a real live example of something that's not exactly MO but looks like it'll work all right anyway, and can breath a sigh of relief and focus on good questions and answers--or their worst fears will come true and they can go into damage control mode, which is already what they are bracing for.
Though it doesn't entirely jive with my previous answer, I don't think there are so many decidable policy issues at stake for the moment. It's hard to do, but we really do need to mostly wait and see--just having braced ourselves for the range of unexpected or unwanted things that could come to pass.
Thanks to Kaestur, Jonathan and several others who've helped me clarify what I think about all this coherently and refocused me away from certain needlessly hypothetical policy questions.