Toby-
OTOH, let's suppose that we have individual articles on each character in addition to an article [[Characters in Atlas Shrugged]] that (I guess) summarises information on the characters.
I don't want that. I want an article that contains a *description of the characters* without all the details. If anything changes, we can just do a search/replace in that single text, instead of doing it on dozens of pages. Only for the most important characters separate articles may be appropriate (e.g. in a [[Disney character]] discussion, [[Donald Duck]] would be linked to, whereas [[Lil Bad Wolf]] would be described on the page).
In general, about the same amount of text appears under either system
Nope, you need redundant intros with separate articles. For one paragraph articles, this is substantial (about 20%). In addition, in my system, the text is on one page, making it easy to do search/replace operations.
Yes, it does, and we've already decided our standards of significance: Gnipper *is* significant enough. Thus we have an article on him.
This is not the only question where significance matters. In my view, if a subject is relatively insignificant, it is more proper to discuss it in context.
Whether an article can theoretically grow is a somewhat counter-intuitive way to *determine* significance, even for a non-dynamic encyclopedia. My policy suggestion could also be phrased as "Insigificant subjects should not have their own articles but instead be merged into longer ones", but that's more vague and can perhaps be perceived as condescending ("Gnipper is no unsigificant! He's the best dog EVER!").
So is your proposed policy about potential length or significance?
Both. Mostly it is about avoiding eternal stubs that are likely to be neglected (I'm surprised that I think of this argument only now, it is a very important one: very short articles on insignificant subjects are more likely to be neglected, because fewer people will follow all the links from an obscure page, and with more and more articles, the random page likelihood per article decreases.), and because of all the other disadvantages of short articles I have mentioned.
PS: Do you have any objection to [[en:Sarah Marple-Cantrell]], the example that got Oliver started on all this?
Yes. I would prefer this case to be discussed in the context of teenage suicides, with a redirect to that discussion on her name.
Regards,
Erik