The vision of a web of triple-blind, peer reviewed scientific observations, published independently of the "story" behind the research is audacious. But it's not clear what incentive researchers would have to publish this way -- especially the tantalizing, breakthrough observations. And it's also not entirely clear that observations have an independent existence outside of the scientific stories that give them context.
If this takes off, I can see this being a way for frontier-edge research being scooped. I'm not sure I see the new value-add beyond that.
What does this offer over arXiv?
I guess peer review, which arxiv doesn't have, and their LEGO model - you (and others!) can go back and add new data to your first data point, thereby slowly arriving at the "story" by linking several data points from you and others. You can only replace arxiv papers with newer versions, and then only by you.
I did not see where this will be indexed and archived. Without it, it won't take off. I have many old data sets that did not really fit into a nice story and have been sleeping on my harddrive. I would love to have an easy way to dump those figures out on the internet.
if you want to upload just raw data, throw it in datazar.com
You might want to check out figshare, it's free and you'll get doi's assigned for your work.
Disclaimer- I work for the parent company, Digital Science.
Check out osf.io, it's run by a non-profit called the Center for Open Science, and is dedicated to hosting scientific data and papers. You can even link your github / Dropbox, and choose at any point to register your project, which will copy all the files with the hope of freezing it in time.