One metric that could be calculated automatically: how many articles in the book contain hyperlinks to the articles that appear later in the book, or don't appear in the book?

I guess we should try to make that number as small as possible, for convenient linear reading, but of course it has to be balanced against other concerns (such as keeping articles with similar topics together, so we can make chapters with a unified topic).

Another idea: identify the main topics of the book (they roughly correspond to the chapters you have now: "map and territory", "words", "politics", "reductionism", "quantum physics", "metaethics", "community", etc.), and mark each article with the corresponding tags.

The goal of this is to reduce the cognitive load while sorting the chapters. Instead of thinking "what was the Fake Morality article about, again?" (and trying to keep all those articles in your head), reduce all chapters to a short information: article name (or identifiers), article tags, names of referenced articles. And then only use this information to sort the articles. -- Perhaps by describing every articles in one line of a text file, easily moving them around in the text editor, and calculating the "non-backwards hyperlinks" metric automatically.