Automated extraction of structure + metadata from a URL

[metasj]

For the i3 data backend:
How much structured data can be automatically drawn from a source URL?
This would let people spend more time focusing on the fields that can't be inferred (what a source is used for, what other scripts / datasets it is used with)