I think my original point still went unaddressed...I think we're still going about using mistaken ideas about judging the current state of the articles collectively. In particular, I strongly do *not* believe that hitting the "random page" button 20 times (or even 20,000 times) is a very valid way of assessing the success of the project, for specific reasons. I don't mean to sound like I'm attacking those who've done this...after all, who hasn't played around with the random page button?? I know *I* certainly have. But, we should remain critical.
There are many objections. First, as Charles pointed out, in any system which is growing even roughly *exponentially*, you would expect the relative *percentage* of a particular type of article to remain roughly constant over time. In other words, it should come as neither a great shock nor a great threat that the *percentage* of stubs, bad articles, etc. has remained constant over time. This is what is expected, given the growth, and given that we're nowhere near an "equilibrium".
Here is a more equitable procedure: hitting the "random page" button essentially assigns the same probability to each article. But, is this really useful? What would be *more* accurate would be to assign a probability to each article proportional to the actual number of hits it gets from flesh-and-blood readers. By assigning equal probability to every article, this puts some obscure fan fluff stub article on an equal par with [[United Stated of America]], even though the latter may command 100s or 1000s of times the actual page views/day of the former.
But an even stronger objection I have to the "hit the random page button 20 times" approach is that it collects an essentially *static* view of the process. To put it mathematically, we are only looking at the 0th derivative, instead of the *FIRST* derivative, let alone the 2nd or higher-order derivatives. Of course -- in order to consider anything beyond the "0th derivative" you have to have more than a static view. You have to know what's going on in some time-region. You have to have a *longitudinal* view.
Here's what I would suggest would be more enlightening: go back 24 months, when the English wiki had roughly 200,000 articles. *Then*, hit the random page button 20 times. Then follow and see how those 20 articles have improved or degraded in the following 24 months. If many of the worst of these 20 articles show little or negligible improvement in the past 24 months, then I think this is something that needs to be thought about. But as it is, we're ignoring the possibility that many of the good (or even brilliant) articles of today might have been classified as stub or atrocious 24 months ago. And then we're pointing to the stubs and atrocious articles of *today* as evidence that the project is failing?? Does the irony of this make sense?
There are many other issues which are completely ignored by taking a static view. For instance, someone noted that general articles on basic issues are often in a much worse shape than highly-specialised articles on obscure topics. This has sometimes been an issue in math. It's something that needs to be thought about.
darin