New text to address some of the issues The scalability (milk-carton) fallacy ===================================== (The "let us stress that ..." didn't suffice. We need to talk more about that one.) Computer scientists are always trained to ask "how does X scale" for any value of X. For algorithms and data structures, this makes a lot of sense since an exponentially scaling algorithm gets much worse than a logarithmically scaling one with quite few data items to process. However, there are clearly situations where some scalability questions make no sense at all. For instance, when buying a refrigerator for one's home, saying "Yes, that one seems like a quite nice fridge, ... but what if I want to buy 500 000 milk cartons" makes no sense at all. In home use, buying that many milk cartons is never sensible; they'd spoil before they could be consumed. And for institutional use, a wholly different arrangement (an industrial-scale refrigerator or even a milk tank) is appropriate. *The limitations, i.e., the limited consumption rate of the human users made the scalability question irrelevant*. Several people have taken issue with buoys about their scalability. "What if there would be a million buoys - would it work?" Naturally, it wouldn't. But asking the question already betrays a lack of understanding about item-based systems and user interfaces. **For any single user, a million items connected to a single item would never be relevant.** As we have said already, items are things that *the user* cares about. No user can care about a million separate things that would be connected to a single thing. Rather, for large amounts of things connected to one thing, a human would mentally model them as a tree, classifying their relationships. ... We claim that for any single user, the global information space can be divided into two parts: the explored and the unexplored part. The explored part consists of articles the user has read; webpages that the user has visited or weblogs the user follows; email the user has received or sent, and the user's own data. On the other hand, the unexplored part consists of articles the user has not read (or has just seen the abstract and decided not to thoroughly read it); websites the user has not visited. We make several conjectures about the natures of the explored space: - the local degree of connectivity is relatively low (a single item is linked to a relatively low number of others), with the exception of links that express the *type* of some object; And even then, the structure should be normalized by *classifying* the links. For instance, if there are 200 meetings with Carli in the past, they are not all connected directly to Carli but rather, the user would prefer concepts like "Meetings with Carli about Article X" "Meetings with Carli about Grant application Y", "Group meetings", ... We have seen this in our personal experience with FenPDF: whenever too many transclusions from articles come to one place, there is an urge to organize and classify them - in the classification appropriate to that place. - Zipf's law applies to the use of the explored space; time-variation: at different times, different parts are most used, but over any timescale, Zipf's law would apply (this is why we see the unique textures as being useful in the explored space) - A freely associative network structure is a relatively appropriate representation for the explored structure - The explored structure may not have much in common with the "official" structure of the items. These conjectures are what our interfaces are based on; with the interfaces presented in this article, we hope to conduct practical research with the goal to confirm or reject these conjectures. The user interfaces presented in this article are intended There is much work to be done in how other peoples' structures could help the exploration of the unknown space. The buoy user interface is specifically intended for