Hypothesis: these people might not have knowledge, but they're very willing to schlep, and so—with enough effort and iteration—they can take a plan and turn it into working code.
And on the other hand, some people who are very "talented" on paper are nigh-unemployable because they just can't get things done. (They can easily tell someone else how to get the job done, but they have no drive or desire to schlep for themselves.)
The other side of this is that, at this level of developer, these are exactly the positions where offshoring makes sense. If you're going to get people whose primary talent is a willingness to schlep just adequately enough to be employable... well, I can find those kinds all over the world, and can probably do better in the skills category, for much less cash. So if I'm looking for domestic (I'm in the US) technical talent, I'm usually looking for those that are worth the higher pay commanded in the US... which means I am absolutely trying to sort out the schlepers for the most part. Naturally, offshoring has its own difficulties regardless of why you're doing it; so there are times when schleppers (I like that term :-) ) are sufficient, but at some point that strategy changes (growth, etc.)
I've literally seen engineers with 20 years of experience grind for 45 minutes on FizzBuzz, not able to make any progress.
As for companies innovating in this regard, it's kinda funny to think about it. When it first got started, I think LinkedIn was trying to be exactly this. Rather than encouraging people to add any and every person they ever thought about (or that asked), the idea was that you should only add those people that you know and trust. In this way, you could use your LinkedIn network as a network of trust for things like hiring or looking for work. That's changed I think both by LinkedIn wanting to have a large network of data to mine and by any number of participants that will add connections since they're really after a bulk network rather than a trusted network. In both cases the business case for such a network of trust doesn't seem to work well for the company providing the service and doesn't work well for the participants... so it breaks. Naturally, that's just one approach and there could be others, but I do think there have been efforts here.