Any system can be gamed, but some are harder to manipulate, or require colluding with a large and thusly brittle group.
A good article-level metric would be a representation of the assigned "PageRank" variant score. A paper linked not only to a large number of papers, but linked to a large number of well-linked papers is undoubtedly an important paper. While it's possible that it's important in a negative sense, that outcome is far less likely, and becomes less likelier still, as the score grows.
Would be very interesting to compare PageRank-oid scores with classic metrics such as number of citations, and also other newer metrics, such as number of views, downloads, tweets, likes, etc.
In other words, the world already fairly successfully relies on PageRank for the vast majority of sorting by importance or merit in information lookups, and science is merely a specific variant of this.
This post has been migrated from the Open Science private beta at StackExchange (A51.SE)