There are a couple additional properties which help an architecture to use a glass box representation, uniform representation of knowledge and a meaningful representation of knowledge. A uniform representation of knowledge allows the various parts of the architecture to all use similar methods to examine and change the knowledge. A meaningful representation of the knowledge may not be strictly required, but it eases the use of a "glass box" knowledge database. One can imagine, for example, that it would be somewhat difficult to reach in and "tweak" the probabilistic network used by ICARUS in any meaningful way. In some sense whether or not a specific knowledge representation is "meaningful" or not is decided by whether or not any of the other modules already know how to utilize it.
A glass box representation of knowledge can be used to help facilitate Learning and meta-level reasoning. Learning mechanisms can reach inside the control knowledge and update it to reflect newly learned information. Meta-level reasoning capabilities, by definition, have to have access to the knowledge that the system contains.
A glass box representation does not come without costs, however. An architecture which utilizes a glass-box approach to knowledge representation usually finds that that representation is now fixed. Since so many parts of the architecture now have their fingers in the low level details of the knowledge database, the representation of that knowledge cannot be easily changed without breaking these other mechanisms.