Proving or disproving these assumptions would be pretty important for strategic direction at a QA site, incidentally, since the world if I'm right looks very different from the world if I'm wrong. Since StackOverflow's data set is public, you could do it as a weekend project. In general, though, this is the kind of stupidly powerful thing you can go with the data you already have in your startup.
And so this reminds me of an idea for a more focused version of HN I'd had a while back. Let's call it 'Bluntr' for now. All stories are requests for review of work: code, designs, websites, products, whatever. All comments are supposed to be brutally direct feedback, with as much concrete backup-data/recommendations as possible.
Note that this does not mean 'nasty': no mockery or vague insulting filler words would be permitted. But it also means every thread can jump straight to the point, without traditional 'attaboy!' or 'I like the X but...' niceties slowing things down. Essentially: at Bluntr, empty compliments or insults would both be flagged/downvoted to oblivion, while steely-eyed evaluation would be promoted.
The community might not gel without those normal verbal ticks and grooming-activities, but then again these ground rules might help the really focused, strong contributors cut to the chase a lot faster. You are not allowed to be either nice or nasty, just honest -- and so you are free to be more productive. That's the initial theory to be tested, at least.
Please avoid such racist stereotyping. There are terrible and awesome programmers in every community, and it's an insult to put them in the same bin and call them the same name. I found it quite offensive myself.
'...answered by a terrible/awful Java Programmer from India'.
If he wrote following,
'...answered by a Perl hacker from Germany.' his point will still be valid. I think you are reading too much into it.
P.S.:I am an Indian.
There are alot of words on the landing page [of BCC]. Look at some of the ycombinator pages and you will notice the average site has about 45 words and 4-9 links. Don't give them so many options and so much to read. Otherewise, looks good!
The quality of your users' interactions with FeedbackRoulette is directly related to the quality of the advice they receive. To recruit and retain the best users who give the best advice, you have to be giving out advice which is worth their time. The above quoted advice is well-intentioned but does not convince me that I am likely to realize business value by continuing to be engaged with the service.
To see another take at resolving this problem, take a look at StackOverflow: by breaking the 1:1 nature of the communication, you can use people's desire to peacock to get hypercontributing members of the community to be responsible for most of the good advice on the site, and let the 90% who do no writing merely vote the good stuff to the top. Hopefully they pick the right good stuff. If, on the other hand, StackOverflow picked two random programmers with a question and connected them, the average experience of using it would be getting your query on Rails ActiveRecord validation answered by a junior Java programmer from India. That would be a much less effective StackOverflow. It would likely not maintain the interest of somebody like John Skeet (who, if you are not familiar with the SO community, is something of a local legend).
(Sidenote: if this reviewer was right and there are in fact YC companies with 45 words on the front page... how do I put this gently... there must be someone in your alumni network who you trust on SEO, right? Talk to him about whether that is a good idea. It will be a very brief conversation.)