Four strands of theory support the method, each given only as far as the method uses it and introduced by the question a reviser is asking when it becomes useful.
Could this scene go? Removability, kernels and satellites
A writer who suspects a scene could be cut needs a test that does not depend on whether the scene is enjoyable. The oldest is Aristotle’s. In chapter 8 of the Poetics (c. 335 BCE; translated by Malcolm Heath, 1996) he argues that a plot is unified by the internal relation of its parts, so that if any part is displaced or removed the whole is disturbed; anything whose presence or absence makes no perceptible difference is no part of the whole. In chapter 9 he adds that events in a plot should follow one another by necessity or probability, because rather than merely after, and he calls the worst kind of plot the episodic one, in which episodes succeed one another without either.
Roland Barthes (1966; translated by Stephen Heath, 1977) separated the functions that open or close a consequential alternative from those that fill the space between them, and Seymour Chatman, in Story and Discourse (1978), gave the distinction the names writers now use. Kernels are branching points, moments at which the story could go one of several ways. Satellites amplify a path already chosen. Deleting a kernel breaks the story’s logic; deleting a satellite impoverishes the story without breaking it. The extraction test among the diagnostic tests is this distinction turned into a question a writer can put to a scene.
What does “dramatised” mean, if not dialogue? Duration
Writers often equate a scene with dialogue, and the equation produces scenes that are talk without event. Gérard Genette’s Narrative Discourse (1972; translated by Jane E. Lewin, 1980) offers a better criterion. He analyses duration, the relation between story time (how long the events notionally take) and text time (how many words or pages the text gives them). Four relations exhaust the possibilities. In ellipsis, story time passes and the text gives it nothing (three years later). In summary, the text is shorter than the time it covers. In scene, the two are roughly equal, so that reading feels like being present. In the descriptive pause, the text expands while the story stands still. A scene, on this account, is the mode in which the telling moves at about the pace of the events it tells. Dialogue is its usual carrier, not its criterion.
Screen drama shows the criterion in a pure form, since a filmed scene runs at the speed of its events unless it is cut. “Review”, the seventh episode of the first season of The Bear (FX, 2022), holds that ratio for a continuous take of about eighteen minutes. A pre-order system left switched on floods a restaurant kitchen with orders, the head chef turns on his staff, and the sous-chef walks out before opening; the pressure comes from the refusal to cut, which denies the viewer the ellipsis that would let the service end. Prose seldom holds isochrony, the equality of text time and story time, for so long.
The duration account also explains why “show, don’t tell” is a poor guide. The slogan descends from Percy Lubbock’s distinction between “picture” and “scene” in The Craft of Fiction (1921), and it conflates three separate variables: duration, narratorial stance (presentation against commentary) and diction (the concrete against the abstract). Wayne Booth’s The Rhetoric of Fiction (1961) showed that all narrative is told and that showing is one rhetorical strategy among several. This chapter uses only the first variable.
Why does a scene in which nothing changes leave no trace? Events and memory
The reader’s failure to remember the first draft of Nadia’s visit to the flat, in The Problem, has a partial explanation. Mieke Bal, in Narratology (1985; fourth edition 2017), defines an event as the transition from one state to another. On that definition an occurrence that leaves every state as it was is not an event, however much activity it contains: they talked for an hour is not an event, and they talked for an hour and afterwards he would not look at her contains one.
Discourse psychology reaches a compatible position from experiment. Rolf Zwaan and Gabriel Radvansky’s review “Situation Models in Language Comprehension and Memory” (Psychological Bulletin, 1998) describes readers as building a situation model, a mental representation of the state of affairs a text describes, which they update along a small number of dimensions: who the protagonist is, where events take place, when, what is causing what, and what someone intends. Work on event segmentation finds that people divide continuous experience into events at points of change; Nicole Speer, Jeffrey Zacks and Jeremy Reynolds reported brain activity time-locked to such boundaries while participants read narratives (“Human Brain Activity Time-Locked to Narrative Event Boundaries”, Psychological Science, 2007). The step from these findings to the claim that a reader’s memory of a novel is largely a record of its transitions, so that a passage with no transition leaves little to remember, is this chapter’s inference and not a result of the studies.
Bal supplies one further term. Focalisation is the separation of the one who perceives from the one who speaks: the narration may be in the third person while what is seen is what one character sees. A change of state is always a change for someone, and a scene in which the value shifts for a character other than the focaliser is a scene told from the wrong side. The Viewpoint Shift (Chapter 13) treats the problem at the scale of the book.
How far does the test bind? The value shift and the period convention
The test this chapter applies to every scene is the value shift: a scene that earns its place moves some charged value (trust, safety, knowledge, hope, standing, intimacy) from one pole towards its opposite, so that something is true at its close that was not true at its start. The term is Robert McKee’s, from Story (1997), a book about screenwriting, where each unit is under heavier pressure than in a long novel. McKee insists that the value be charged, meaning that the story has already made the reader care about it, and that requirement is what separates a value shift from mere incident. Four qualifications follow, and they govern how the test is used.
The test measures charge rather than size: a scene in which a woman notices that her husband has stopped asking about her day has shifted a value, and a car chase in which nothing is risked has not. The shift need not be visible to any character; it may happen entirely inside one mind, as in the James chapter, or entirely in the reader’s knowledge, as in scenes of dramatic irony, in which the reader knows something a character does not and so hears every line twice. Some scenes fail the test and are indispensable, because they answer to the book rather than to the scene: chapter 94 of Moby-Dick, examined among the failure patterns, is one. And the test is the instrument of a period convention, a preference that belongs to a time and a tradition rather than to narrative itself. Henry Fielding’s Tom Jones (1749) summarises months in a paragraph and reserves full dramatisation for a few occasions, and the rationing gives those occasions a force that a wholly scenic novel cannot produce, since a book in which everything is a scene has no means of marking one occasion as weightier than another. Genette found the opposite proportion in Proust, with summary almost gone. Both are deliberate designs, and applied to Fielding or Sterne the test reports failures that are not failures.
The test is therefore not a standard every scene must meet but an instrument that makes a writer’s choice visible, so that a scene that fails it can be kept knowingly, for a stated reason.
See also: the full theoretical account in Reference, including Freytag, Barthes’s code of actions and the historical table of scene–summary proportions; The Transition (Chapter 5), on ellipsis.