As Figure 3 indicates, Group 3 organized their document more hierarchically than Groups 1 or 2. The top node of the Group 3 document consisted of links to nodes for the main components of the document. These second-level nodes consisted of links to content nodes grouped under the index node's particular heading. Also accessible from the main page was a graphical overview of the document. Each of the second-level nodes was accessible from the clickable diagram and the graphical overview is likewise accessible from each index document.
One of these second-level indices, the cognitive architectures node, had a list of links to an index node for each architecture. This index node was subdivided into the four sections mentioned in connection with the other documents. In each of these sections, a list of keyword links to a descriptive node on that feature, as implemented in the architecture, was also made available. These specialized feature nodes included contexual links to other, related nodes in the same architecture (for instance, the connection between property Px and capability Cw in Figure 3) as well as links to meta-level discussion nodes and/or common feature nodes. All these nodes followed Group 3's ``one-page'' rule of thumb; with few exceptions, the nodes were about a monitor page in length. At the low-level nodes, an explicit link was provided to return to the architectural index page; these are shown as dotted lines in Figure 3.
The properties, capabilities and environments nodes consisted of a list of feature names linking to common feature nodes. These nodes consisted of a few paragraphs giving a description of the common topic in general terms, possibly including definitions, examples of implementation methodologies and the importance and difficulties of implementing the relevant feature. Each of these nodes also included a list of those architectures sharing the feature. Importantly, these link to the specific, relevant node for individual architectures. For example, the ``modular construction'' property includes links to nodes in PRODIGY, MAX [Kuokka1991], and RALPH-MEA [Ogasawara1991] containing individual discussions of the modular organization of these architectures.
The high-level hierarchical organization facilitated a straightforward and consistent integration of individual architectures into the document as the course progressed. In contrast, the more free-form organization of low-level content nodes allowed a dense interconnection of links at the architectural description level, tailored to the particular features of the individual architectures.
