Yes, I think it might be an interface problem as well. Maybe tag clouds were a good idea after all?

> I think tags model all the information you want.

> other tags for documents tagged with databases (subdirectories), and documents with only databases as a tag (members of this directory)

I don't like to tag my items "databases, mysql". I just want to tag them "mysql" and I want the system to know that MySQL is a type of database. Besides, there might be too much noise in the "other tags" list once you have a large number of tags. Just because I bookmarked an article about making TLS connections to MySQL doesn't mean that "tls" belongs to "mysql" just as much as "innodb" does.

Tags don't model all the information I want. Even if it does, I feel like there's a serious impedance mismatch between the tagging model and my mental model. MySQL is a database. Give me all the items that I tagged with (any database + rails) between 20120701 and 20120731. The mountain lion is a big cat. Give me all the items I've tagged with a big cat. Some things in the world naturally fit into categories like that. The world is not a flat namespace in my mental model. A good UI would help narrow the mental gap, but the gap will always be there.

To say that a one-dimensional tagging system can model all the information I want is like saying that a key-value store can model all the information I want. Just because I can stuff my data into it and make basic queries doesn't mean that it's the best way to organize my data.