My take is that some specifications can be written out in a linear way where you can start reading at the beginning and work to end and not feel like you need to read ahead.

Some specs have a minor discontinuity, I remember perceiving it in the K and R book on C but it seemed like there was just one kink in it and if you read the book twice you’d do OK.

Books in C++ are worse and have numerous topics that resist being put in the right order. It’s not unusual for “resource acquisition is initialization” to be repeated hundreds of times before it is defined, for instance.

That circularity is both a function of the domain and also a function of the text, I think a certain amount of circularity is inherent to many domains, but frequently you can bootstrap a domain by dividing it into numerous layers and put the circularity into a layer built just to manage the circularity.

XSLT, XMLSchema, and many XML specs have that kind of circular structure, you are left wondering what exact kind of machine is required to implement it so you can look at the spec and have a hard time understanding how to do easy things and no grasp of the hard-looking things that are actually easy. Couple that with numerous sharp edges in XML such as numeric values not being allowed in ID or IDREF fields (hate to break it to them but numeric identifiers are rampant in the jndustry) and it is no wonder people would rather use deeply lame ‘standards’ like JSON that lack comments, aren’t really clear about the semantics of numbers, and don’t have the moral authority to say “quit screwing around and just use ISO 8601 dares.

Now I finally realized the OWL spec is perfectly clear in the sense that you can understand what it really does by understanding the mapping of OWL axioms to first order logic, but the trouble is that logic is the most treacherous branch of mathematics.