A typical interaction might go as follows: The user first sees an overview of a virtual city, as if seen from high in the sky. The city is a high-level view of all the user's data. The user mouses over to one particular neighborhood and that neighborhood fills the screen.

A number of buildings are now visible with labeled awnings hanging off them saying what they represent. All the currently visible buildings will be in a certain general district (say the Novels Neighborhood, Drama Neighborhood, History Neighborhood, Literary Criticism Neighborhood, and so on).

To aid the user's memory, buildings in each neighborhood should be visually distinctive, as should all the rooms in a building, and so on. Much of memory is spatial: the thing searched for is recalled to be near the watercooler or it's in the bright red building next to the two-story Victorian dormer. Creating a variety of styles is easy to do by building in a large number of different ornaments and incidental detail and having them randomly assigned to each building when it's created. (To those who balk at the computational cost of rendering "unnecessary" visual detail, think of visiting a bookstore whose every book had exactly the same dimensions and color with only their texts to distinguish them---that's today's desktop in a nutshell.)

Selecting, say, the Novels Neighborhood building puts the user into that neighborhood. From then on, the user can only see pages related to novels; to get to nearby Neighborhoods the user must first go back to the roof. The Novels Neighborhood building may be further divided into floors, the floors into rooms, the rooms into bookcases, and so on. Each selection successively restricts the user's view of the totality of data. Protecting the user from information overexposure is what good information management is all about.

When the user find that Jane Austen novel originally searched for, other novelists similar to Jane Austen are near to hand. Further, once the user browses the generic neighborhoods a while, that user will start linking things that seem related to that user and not necessarily to anyone else. The mapping system can extract some of this new linkage information by analyzing the user's link-following behavior; it need not force the user to enter explicit directives. This information could then be used to reconfigure the neighborhood linkage map as seen by that particular user.