Well, I restarted Author Intrusion again. The last iteration was promising but I was spending my time working on reinventing stuff because I thought it was what I want. Earlier this week, I realized what it was and how I could use an existing library to hopefully avoid spinning wheels.

It does mean I can't use CSS selectors for phrases. Instead, I'll use XPath and C#'s XElement.

No `file.chapter word` and probably `//file[has-class('chapter')//word`.

Not as pretty but I need this working.