Readers of interactive narratives can proceed only on the basis of choices they make.

I question the validity of this characteristic only because I don't think "choices" is not inclusive enough. (Bear with me, I promise there's an important distinction here. Besides, hypertexans have always bickered over terminology, so why stop now?) Although a hypertext should be ergodic, it should not have to necessarily pose choices to the reader. For example, it's hard to say that Erik Loyer's Chroma gives the reader any decisions. The reader has freedom in the visual plane of the work, exploring the "performance" of the text as it responds to mouse movement. But the verbal element of Chroma straps the reader in for the ride. She/he can't prevent Duck from getting angry at Perry in Chapters 4-6. The fact that the Chapters are numbered at all gives the text a privileged order of reading. The reader does not have any "choices" about the narrative, but there are still many ways of performing the text (a newer type of multilinear reading) in addition to the constant user interaction (ergodics).