For instance, there is a list of domain names that wikipedia sees often as spam. So if an innocent user goes to edit a page, and wants to add a new piece of information referencing a domain on that list, a bot will come along and mark it as spam (and revert it) even though its a good faith edit.
It can be frustrating and confusing for new users. The bot is often quite terse and rude, and so there is no incentive to return to wikipedia to make edits. 9/10 new edits are rejected, so no wonder they can't keep new editors.
"the neverending stream of basically anecdotal information"
should be the first clue that this is basically what everybody seems to already know except for the foundation. This is not even remotely a new issue as their data and numerous complaints, articles, comments etc. shows, but it's taken them more than half a decade to even bother to notice it.
Here's from 2010:
http://ostatic.com/blog/is-wikipedias-deletionism-out-of-con...
2009
http://schott.blogs.nytimes.com/2009/11/26/deletionists/
http://blogs.telegraph.co.uk/technology/shanerichmond/100002...
2008
http://news.slashdot.org/story/08/09/21/2342226/debating-del...
2007
http://www.roughtype.com/archives/2007/08/rise_of_the_wik.ph...
and on and on and on
Here's foundation's response during all this
http://news.bbc.co.uk/2/hi/technology/8382477.stm
It's only in the last 2 years has the foundation even bothered to acknowledge there's a problem, and only after the problem became so bad that it couldn't be ignored.
The Counter Vandalism Unit was created late 2005; so maybe it got traction during that time and over-zealous members templated the newbies?
One thing I see in the graphs is that the "survival rate" of editors who made a good-faith first edit was already in relatively steep decline by 2005, but the same rate for editors whose first edit was golden remained on a high plateau through 2006 and then just categorically dropped off a cliff. What happened then?