The idea is to describe what new code does for a user by saying what the rules of the world are after the function of the new code has been made available in the form of a capability N. The new style of description rigorously denies the existence of internal parts of X except in an abstracted form. This is precisely parallel to the description of a machine instruction set that omits description of the many internal registers used to make the machine run faster, and speaks only of those registers whose contents are relevant to the meaning of a program. It is a specialized ontology which is uniquely suitable for the programmer as he writes code. It is a relative ontology—what there is depends on where you stand.

In the context of defining a CPU this is old hat. In the context of defining an OS kernel it is fairly common. But when the Gnosis manual says that “N will be the only key to a new node” we are lying except in an ontology which omits the innards of the space-bank that incidentally still has a key to that node. What is new and probably disorienting is that learning a capability platform continuously introduces these new relative ontologies. Each time a new facility is introduced, the ontology in which one can explain how the code for that facility works, is different from the ontology suitable for the user of the facility.

David Hopwood uses “capability system” where I use “ontology” here.

Here is a kindred dilemma.