The company behind Scirus is Elsevier Science, subsidary of Reed Elsevier, dutch based multinational scientific publishing giant.
Other than Northern Light Scirus concentrates on natural science content, and other than Northern Light Scirus does not charge for access to sources.
Those documents that are not freely avaliable are just retuned with a summary and source.
Scirus draws from two sources, scientific websites, and scientific journals and archives.
The non web sources are:
1.4 million ScienceDirect full-text articles
200,500 IDEAL full-text articles
1.9 million Beilstein abstracts
Over 12 million MEDLINE citations
170,000 e-print on E-Print ArXiv
550 Neuroscion news items
370 BioMed Central full text articles
800,000 patents from the USPTO
I believe, but haven't checked yet, those journals all belong to Elsevier.
The sources from the web include 69 Mill. pages, among them from
.edu-sites: ca. 40 million
.com-sites: ca. 1 million
.org-sites: ca. 1 million
.ac.uk-sites: ca. 3.5 million
Now how does Fast, erm Scirus decide which page/site offers scientific content and which doesn't? With the .edu sites this isn't too complicated, but there is also a submit option. Perhaps a technique related to Fast's dynamic categorization is in effect here?
This is how Scirus ranks:
1. Words - the location and frequency of a search term within a result account for one half of the algorithm. This is known as static ranking.
2. Links - the number of links to a page account for the second half of the algorithm - the more often a page is referred to by other pages, the higher it is ranked. This is known as dynamic ranking. Overall ranking is the weighted sum of the static and dynamic rank values.