Everything in this chapter is an instrument for noticing. None of it is a rule, and the evidence for that is contained in the chapter itself.
Consider what the model, mechanically applied, would remove. It would remove “A Squeeze of the Hand”, because no value shifts. It would remove the second flogging in The Trial, because the reader has already had that scene. It would remove nineteen chapters of Waterloo, because a set piece must not run away with a book. It would remove Eliot’s narratorial intervention in the middle of the Rome chapters, because a scene should not be interrupted by an essay. It would remove Levin’s day in the meadow, because one domain is not enough. It would compress chapter 42 of The Portrait of a Lady into a paragraph, because nothing happens and nobody speaks — and James, who wrote it, thought it the best thing in the book, and said that it threw the action forward further than twenty incidents would have done.
A model that rules out that list has not identified defects in those scenes. It has identified its own limits.
But the sharpest version of the argument is not that great scenes fail the test. It is that the test was never a test. A test has a pass mark and a consequence for failure. What §7 actually offers is a set of questions that convert an unusable dissatisfaction into a usable one. This scene isn’t working becomes: the want belongs to the wrong character; or this is a satellite at kernel length; or the reader banked this shift eighty pages ago; or this is summary that has been inflated into dialogue because I only know how to write scenes. The first sentence can only be worried at. The others can be acted on.
The value-shift test’s function is not to identify which scenes are good. It is to make a writer’s choices visible to the writer.
A scene that fails the test may be the finest thing in the book. What matters — the only thing that matters diagnostically — is whether its author knows that it fails, and can say what the book is buying with it. Melville can. Kafka can. Hugo, arguably, cannot, and Les Misérables survives anyway, which is a further and salutary datum. The writer who cannot answer has not made a choice; they have kept a scene because it was there.
There is a deeper reason the scene resists formula, and it belongs at the end. The instruments in this chapter all measure change, because change is what can be measured. But the experience of reading a scene is not an experience of registering alterations in a state vector. It is the experience of being somewhere, with someone, while something is at stake — and the sensation that the writer knows how much attention this occasion is worth and is spending neither more nor less. Every test in §7 is an oblique way of detecting that judgement in someone else’s manuscript, and in one’s own.
Which suggests the last and least mechanical question, to be asked after the other ten:
If I read this scene in a book I admired, would I feel that the writer had chosen to spend these pages here?
If the answer is yes, the scene has a job, whether or not the chapter can name it.