3. Why Endings Carry Disproportionate Weight

The retroactive lens is asserted constantly in craft literature and almost never argued. It can be argued, and from an unusual number of independent directions — philosophical, formal, historical, political, cognitive. Endings have, in fact, the richest critical literature of any subject in this textbook, and a chapter that ignored it would be teaching intuitions that were rigorously theorised sixty years ago.

Nine strands are assembled below. They are not variants of one another. Kermode is making a claim about human temporality; Herrnstein Smith about formal structure; Brooks about desire; Aristotle about probability; the cognitive literature about memory; DuPlessis about ideology; Miller about the whole enterprise. Where they agree, the agreement is evidence. Where they conflict — and Kermode and Miller conflict fundamentally — the conflict is the most useful thing in the section.

3.1 Kermode: endings as fictions of concord

Frank Kermode’s The Sense of an Ending (1967) is the foundational modern text on the subject, and its central argument is not about novels at all. It is about why human beings need endings in the first place.

Kermode begins from apocalypse. Western culture has spent two millennia predicting the end of the world, being wrong, and adjusting rather than abandoning the prediction. The pattern interests him because it exposes a general structure of thought: we are, all of us, in the middest — born after the beginning and dying before the end, with no access to either terminus of the story we are inside. To make sense of a middle, we posit ends. Kermode calls the resulting structures fictions of concord: imagined shapes that give a stretch of undifferentiated duration a beginning, a middle and an end, and thereby make it interpretable.

His minimal model of this is the clock, and it is the most quoted image in the literature on endings.

A clock, physically, makes an undifferentiated series of identical sounds. We do not hear it that way. We hear tick-tock — a tick that is a beginning and a tock that is an end, with an interval between them that the two sounds organise into a shape. Kermode’s point is that nothing in the acoustics licenses this. The distinction between tick and tock is supplied by the listener, who requires the interval to have a form. Tick is, in his phrase, a humble genesis; tock a feeble apocalypse. And the space between them — the interval that the ending makes into a duration rather than a mere succession — is where plot lives.

Three consequences of Kermode’s argument bear directly on craft.

(1) The ending is what converts succession into duration. Without a tock, a series of events is one thing after another. This is the theoretical form of the distinction Chapter 2 (Scene and Sequel) draws between a sequence of scenes and a causal chain: causality is a shape imposed retrospectively, and the point from which it is imposed is the end. A reader who cannot see how a book will end cannot yet know which of its events were causes.

(2) Endings are impositions, and readers know it. Kermode is explicit that concord fictions are fictions — that the shape is not discovered in the material but supplied. Reality does not have tocks. Lives do not conclude, they stop; and even the deaths that end them are not endings in the formal sense, because they resolve nothing and confer no meaning on what preceded.

(3) And we want them anyway. This is Kermode’s most valuable idea for this chapter, and it is the one most often lost in summary. He distinguishes between fictions that have hardened into myth — held to be literally true, no longer felt as constructions — and fictions held with what he calls a clerkly scepticism: known to be made, used regardless. Endings, he argues, belong in the second category for any sophisticated modern reader. We are entirely aware that novelistic endings are artificial impositions of shape on material that does not naturally have one. We want them all the same. The demand does not survive because it is naive; it survives because the need it answers is not defeated by knowing it is being answered artificially.

Kermode’s principle, in the form this chapter uses it: endings are discredited but not abandoned. The reader who finishes a novel knows the shape was imposed and wants the imposition anyway. This is why the sophisticated open ending is not simply superior to the closed one, and why a novel that refuses shape on principle frequently disappoints readers who fully understand and endorse the principle.

The rest of the chapter turns on that formulation. It permits the chapter to teach the craft of closure without claiming that closure is true, and to take the modernist and postmodern refusals seriously without concluding that the demand they refuse was stupid.

Kermode also supplies a technical term this chapter will use in §6.2. Borrowing from Aristotle, he treats peripeteia — the reversal, the falsification of expectation — as the mark of a sophisticated concord fiction rather than a naive one. A novel that fulfils its predicted shape to the letter is myth; a novel that disappoints the expectation it created, and then satisfies a better one the reader did not know they had, is doing what Kermode thinks fiction is for. The craft translation is direct: an ending should be inevitable in retrospect and unforeseen in prospect, and §6.2 argues that this is not a paradox but a description of how preparation actually works.

3.2 Herrnstein Smith: closure against termination

If Kermode explains why we want endings, Barbara Herrnstein Smith’s Poetic Closure (1968) explains what an ending actually does to produce the feeling. It is a study of poems, which is why it is useful: because a poem is short enough to be analysed exhaustively, Herrnstein Smith is able to describe closural mechanisms at a level of formal precision no study of novels has matched. Her vocabulary remains the most exact available for what a last line does, and it transfers to prose with almost no loss.

Her governing distinction is between closure and mere termination.

A text terminates when it stops. A text achieves closure when the reader experiences the stopping as the appropriate and expected conclusion of the structure — when the final line produces “a sense of stable conclusiveness, finality, or ‘clinch’”.

Every text terminates; not every text closes. The difference is entirely a matter of how the last lines are constructed, and Herrnstein Smith catalogues the constructions. Four of her mechanisms are directly usable in prose.

Closural allusion. The final lines refer, at the level of content, to things that are themselves terminal: death, sleep, night, winter, silence, departure, stillness, the sea. The reference does not have to be structurally significant; the mere presence of a terminal referent at a terminal position produces a closural effect, because the reader’s sense of finality is reinforced by an image of finality. This is the cheapest closural device in the language, which is why it is also the most easily abused. A novel that ends with a character falling asleep, watching a train leave, or looking at snow is borrowing closure from the referent rather than earning it from the structure. It is also why so many endings feel right and audit badly: the allusion has done the work the domains were supposed to do.

Terminal modification. A pattern established across the text is altered at its final instance — a line shortened, a refrain broken, a rhyme abandoned, a stanza truncated. The alteration signals that the pattern is not going to continue, and therefore that it is over. In prose the pattern in question is usually structural rather than metrical: a chapter shape, a section length, a habitual sentence rhythm, a recurring scene type. The last chapter that does not do what the previous forty did is a terminal modification, and readers register it as closure even when they cannot name the pattern that has been modified. §4.6 makes this the core of the proposed sixth domain.

Failure of expectation, or the closural force of the unexpected. A conclusion may close by defeating the reader’s expectation rather than fulfilling it, provided the defeat is legible as final. This is Kermode’s peripeteia in formal dress, and it is the mechanism by which anticlimax (§6.4, no. 11) can close a book.

Unqualified assertion. A final sentence in the declarative mood, without hedging, subordination or qualification, produces closure by grammar alone. Herrnstein Smith notes the special closural force of the epigram, the maxim and the proverb — utterances whose form claims general truth and therefore admits no continuation. Middlemarch’s Finale ends this way; so, in a much stranger register, does Nineteen Eighty-Four. The prose version of the device is §6.4’s aphoristic close, and its risk is §8.6’s explained ending: an assertion that closes by telling the reader what the book meant has bought closure with the book’s own subtlety.

Herrnstein Smith’s other lasting contribution is her treatment of anti-closure in modern poetry as a positive technique rather than an absence of technique. Poems that end without closural signals are not failing to close; they are using the absence of closure as a formal effect, and they can only do so because the closural conventions exist to be withheld. The consequence for this chapter is stated once and relied on throughout:

An open ending is a closural strategy, not the absence of one. It works by withholding devices the reader knows are available, and it therefore depends on the reader’s competence in the conventions it declines. This is why an open ending in an otherwise conventional book reads as evasion, and the same ending in a book that has taught the reader to expect it reads as integrity.

3.3 Aristotle: necessity, probability, and the unravelling

The Poetics (c. 335 BCE) supplies four things this chapter uses, and the third of them is the ancient form of its central craft term.

The unity of action, and the definition of an end. Aristotle defines an end as that which naturally follows something else, either necessarily or as a rule, and has nothing following it. The definition matters because it contains both of the ending’s constitutive properties as this chapter states them in §2.1: nothing after it, and a determinate relation to what precedes.

Peripeteia and anagnorisis. The reversal (a change of the action into its opposite) and the recognition (a change from ignorance to knowledge) are, for Aristotle, the two most powerful devices of plot, and the finest tragedies combine them so that the recognition produces the reversal. This pairing is the analytic core of the classical climax, and it is the reason two of the strategies in §6.4 — the retrospective reframe and the anagnoristic close — are recognisably ancient rather than modern.

Necessity or probability. Aristotle’s requirement that the events of a plot follow one another by necessity or probability is the oldest available statement of what this chapter calls earned. He is explicit that the incidents must arise out of the structure of the plot itself, so that they appear inevitable, and that a plot in which events follow one another merely after rather than because is inferior. Chapter 2 (Scene and Sequel) states the same requirement in modern craft terms as the causal chain — the test of whether each beat “had to happen next”. §6.3 shows that the two are the same test applied at different scales.

The prohibition on the deus ex machina. Aristotle’s corollary is that the unravelling must arise from the plot itself and not from a contrivance — his examples are the machine in Medea and the business of the fleet in the Iliad. This is the direct ancestor of §8.1, the unearned rescue. Two historical qualifications are essential and are set out in full at §8.1: first, that the device is Euripides’ signature rather than his failure, and that Aristotle was expressing a preference against a living and popular practice; second, that a whole later tradition — Gay, Brecht — makes the arbitrariness of the rescue the point.

One further Aristotelian term deserves rescue, because modern craft discussion has lost it. The Poetics divides a tragedy into the desis, the tying of the knot, and the lusis, the unravelling — and the division is not at the last scene but at the turning point. Everything after the peripeteia is lusis. On this account the aftermath is not a supplement to the structure; it is structurally half the play, and a drama that ends at its reversal has performed only one of its two movements. §6.6’s proportion problem is this observation converted into a working ratio.

3.4 Brooks: the anticipation of retrospection

Peter Brooks’s Reading for the Plot (1984) provides the most direct theoretical grounding for the retroactive lens, and does so by way of an argument about desire.

Brooks’s model of narrative is dynamic rather than structural. Plot, for him, is not a shape but a force — the desire that carries a reader forward, generated by an initial disequilibrium and discharged at the end. His key phrase for the reader’s state is the anticipation of retrospection: we read forwards in the expectation of a moment from which the whole will be retrospectively legible. We tolerate the middle because we expect an end that will explain what the middle was.

Two consequences follow.

The end is the point of legibility. Brooks argues that meaning in narrative is not distributed evenly along the sequence but conferred from the terminus. Until the end, every element is provisional — it might turn out to be a cause, a red herring, a theme, or nothing. The end fixes them. This is the retroactive lens, stated as a general property of narrative rather than as a craft observation, and it explains something the craft formulation does not: why an ending can make earlier material better without changing a word of it. The earlier material was always doing that work; the reader could not yet see which work it was.

Plot seeks its own proper end. Brooks’s most contested move is his reading of Freud’s Beyond the Pleasure Principle as a model of narrative. The organism, in Freud’s account, seeks to return to quiescence but must do so by its own path — it insists on the correct death rather than a short-cut. Brooks transposes this: narrative desire seeks discharge, but a discharge that arrives too early or by the wrong route is unsatisfying. What he calls the short-circuit is the unearned ending: the right destination reached by an illegitimate path.

The craft translation is exact, and it can be stated as a rule, because it explains an otherwise puzzling fact about reader response:

Readers do not want the ending they want. They want the ending the book requires. A reader who spends four hundred pages hoping two characters will marry will nonetheless feel cheated by a marriage the book has not earned, and satisfied by a separation it has. The desire the ending answers is not the desire for a particular outcome; it is the desire for the right outcome, and rightness is a property of the path.

This is the strongest available answer to the commercial instinct that endings should give readers what they want. It also explains why the two endings of Great Expectations (§9.1) have been argued about for a century and a half: the disagreement is not about which outcome is nicer but about which one the book’s own path requires, and competent critics have reached opposite conclusions.

3.5 Torgovnick: congruence and the forms of the close

Marianna Torgovnick’s Closure in the Novel (1981) remains the standard monograph on the subject and is the source of this chapter’s §6.4 taxonomy. Two of her contributions are load-bearing here.

The typology of closural relations. Torgovnick classifies endings by their relation to the rest of the novel rather than by their content. Adapted, and with the vocabulary this chapter will use throughout:

The overt and the covert ending. Torgovnick separates endings that announce themselves as endings — an epilogue, a finale, a chapter headed “Conclusion”, a narrator stepping forward to sum up — from endings that arrive without announcement, the last scene being formally indistinguishable from any other. The distinction is useful in revision because the two carry different risks: an overt ending risks §8.6 (the explained ending) and §8.5 (the tidy-everything ending); a covert one risks §8.2 (climax without aftermath) and the failure of termination (§7.1, test 7).

Congruence. Torgovnick’s evaluative criterion, and the one this chapter borrows most heavily, is whether an ending is congruent with the novel’s own logic — whether it emerges from the book’s established terms — or imposed on it from outside, by authorial preference, publisher’s pressure, generic obligation or fatigue. This is the scholarly form of the earned/unearned distinction, generalised beyond the question of plants: an ending can plant everything it uses and still be imposed, if what it does with the material contradicts the book’s own argument. §8.12 makes the imposed ending a failure pattern in its own right, and §9.1 shows the two endings of Great Expectations being argued about in these terms.

3.6 The cognitive strand: recency, position and the peak–end rule

Chapter 9 grounded the opening’s disproportionate weight in Solomon Asch’s work on primacy. The ending’s counterpart is recency, and the empirical picture is both real and more limited than craft writing usually admits. This section states both.

Serial position. The classic finding, established across decades of free-recall experiments and set out canonically in Murdock’s 1962 study, is the serial position curve: items at the beginning of a list (primacy) and at the end (recency) are recalled better than items in the middle. The curve is one of the most robust results in experimental psychology. Its application to narrative, however, must be made carefully. A novel is not a word list; readers do not recall novels by free recall; and the recency portion of the curve is a short-term memory effect that is substantially reduced by a filled delay after the list ends. What the serial position literature licenses is a general claim — positions at the boundaries of a sequence are privileged in memory — and not a specific one about how much of a novel’s ending a reader retains.

Privileged position. Peter Rabinowitz’s Before Reading (1987) supplies the literary-critical form of the same claim, and it is the more directly usable. Readers operate with tacit rules of notice, and among the strongest are the rules of privileged position: titles, epigraphs, first and last sentences, and the openings and closes of chapters are automatically assigned interpretive weight. The last sentence of a novel is, on this account, the second most heavily weighted position in the book, and by some measures the first. It is read as a statement about the novel, whether or not it was written as one. This is the direct justification for the last-line test (§7.1, test 10) and for treating the final image as a domain in its own right (§4.5).

Note the difference between the two claims. Recency is about memory; privileged position is about interpretation. The second is the more important for this chapter, because a novel’s ending does not need to be well remembered to be load-bearing. It needs to be treated as authoritative at the moment it is read.

The peak–end rule. The most attractive empirical result in this area, and the one most frequently over-claimed, is the peak–end rule described by Daniel Kahneman and colleagues. In a series of studies of remembered discomfort — the cold-pressor experiments, and a much-cited study of patients undergoing colonoscopy with Redelmeier — participants’ retrospective evaluations of an unpleasant episode were predicted almost entirely by two data points: the most intense moment (the peak) and the final moment (the end). The duration of the episode contributed remarkably little, an effect known as duration neglect. In the most striking demonstration, an episode that was extended by an additional period of milder discomfort was remembered as less bad than the shorter episode it contained, because the added interval improved the end.

The pull of this for a chapter on endings is obvious, and so is the temptation. The peak–end rule appears to be the empirical form of this chapter’s claim that a handful of unearned pages can flatten two hundred pages of careful work. It is worth being exact about how far that goes.

What the peak–end rule supports: that retrospective evaluation of an experience is not an average of its moments, and that the final moments are weighted far out of proportion to their duration.

What it does not support: any specific claim about novels. The studies concern brief affective episodes — minutes, at most an hour — of directly experienced physical discomfort, evaluated shortly afterwards. Reading a novel is a discontinuous experience spread over days or weeks, mediated by interpretation, memory consolidation, discussion, rereading and social context. The extension from the cold-pressor task to the four-hundred-page novel is an analogy, not a finding.

The honest position is that the peak–end rule is a suggestive parallel that makes the craft intuition less surprising, not evidence for it. It is included here because students encounter it constantly in popular writing about storytelling, usually with the qualification removed, and should be able to state what it actually shows. The literary claim stands on its own evidence: the retroactive lens is supported by Brooks, by Kermode, by Rabinowitz’s rules of notice, and by the plain testimony of readers, and it does not need borrowed authority from a colonoscopy study.

One further cognitive note, on the shape of the aftermath. Richard Gerrig’s work on transportation (Experiencing Narrative Worlds, 1993), used in Chapter 9 to explain the opening’s start-up cost, has a symmetrical implication at the close: a transported reader must be returned, and return, like departure, has a cost. An ending that terminates at the peak of intensity leaves the reader in the fictional world with no exit; an ending that supplies aftermath is, among other things, providing a graduated descent. This is not a claim the experimental literature makes directly, and it is offered as a plausible extension rather than a finding — but it is the most economical explanation available for why a book that stops at its climax leaves readers feeling both satisfied and cheated at once (§8.2).

3.7 The historical strand: from the settlement to the open close

Endings have a history, and every prescription in this chapter is calibrated to a moment in it. A student who cannot historicise the advice will mistake a local convention for a discovery about narrative.

The last row is the one that requires a plain statement, because it is the one students are living inside.

The ambiguous ending is no longer brave. It is the house style of a genre.

For roughly the last half-century, in Anglo-American literary fiction and in the workshop culture that supplies it, the unresolved close has been the default rather than the deviation. A novel that ends on a character standing at a window, a gesture left uncompleted, a sentence that trails, or a final image offered without commentary is not refusing a convention; it is observing one. This has three practical consequences that a writer should hold simultaneously.

1. The open ending no longer carries its historical charge. When Joyce declined to conclude, he was declining something the reader expected. When a contemporary literary novel declines to conclude, it is meeting an expectation. The gesture is the same and its meaning has inverted.
2. Ambiguity is therefore no longer self-justifying. It must now do the work any other strategy must do, and it must do it against the suspicion that it was chosen by default. This is why §8.9’s ambiguity-as-evasion is a live failure pattern now and would have been an incoherent complaint in 1925.
3. The genre distribution is stark and should be named. Crime, romance, thriller and much science fiction retain strong resolution conventions; literary fiction has largely abandoned them. This is a fact about markets and readerships, not about narrative. A novel that resolves fully is not thereby commercial, and one that does not is not thereby serious.

Two supplementary historical observations, both useful in seminar.

On the wind-up chapter. The Victorian settlement is easy to mock and worth understanding. It served a real function under the conditions of serial publication: a novel issued in nineteen monthly parts over eighteen months is read by an audience whose memory of part four is dim, and the concluding survey of the cast is partly an act of retrieval. Chapter 6 (The Chapter Break) describes how the part-issue shaped segmentation; the same economics shaped the close. Trollope, in Barchester Towers, breaks off to remark on the awkwardness of the novelist’s final duty — the gathering of threads that must be done and cannot be made interesting — which is a nineteenth-century novelist telling us plainly that the convention was already felt as a chore.

On Henry James, who saw it clearly. James’s essay “The Art of Fiction” (1884) contains the most quotable indictment of the settlement in the language: he objects to the expectation of a final “distribution at the last of prizes, pensions, husbands, wives, babies, millions, appended paragraphs and cheerful remarks”. And his preface to Roderick Hudson supplies the deepest single sentence written about endings by a practitioner:

Really, universally, relations stop nowhere, and the exquisite problem of the artist is eternally but to draw, by a geometry of his own, the circle within which they shall happily appear to do so.

That is Kermode’s argument, forty years early, from inside the workshop. Relations do not stop; the artist draws a circle within which they appear to. The ending is the circumference, and James’s word happily is doing a great deal — the appearance must be satisfying, not merely asserted.

3.8 DuPlessis: whose endings were available

Rachel Blau DuPlessis’s Writing Beyond the Ending (1985) converts a craft problem into a historical and political one, and it belongs in this chapter as a matter of intellectual honesty rather than as decoration.

Her argument begins from an observation about the nineteenth-century novel that is easy to verify and hard to unsee. For a female protagonist, the plot offered two endings: marriage, or death. A woman’s story concluded either in the social settlement that removed her from the narrative by absorbing her into a household, or in a death that removed her from it altogether. There were essentially no third options that the form recognised as endings. Elizabeth Bennet marries. Jane Eyre marries. Dorothea marries — twice, and the second marriage is treated by the Finale as a diminishment (§9.2). Emma Bovary dies. Anna Karenina dies. Maggie Tulliver drowns. Edna Pontellier walks into the sea. Tess is hanged.

DuPlessis’s term for this structure is the romance plot, and her claim is that it is not merely a recurring subject but a narrative ideology: a set of assumptions about what constitutes an ending at all, which determines what a female life can be represented as having been. A plot that can only conclude in marriage or death cannot represent a woman’s vocation, work, friendship, political life or intellectual development as terminal values, because it has no closural form for them.

Her book is about the strategies twentieth-century women writers developed for writing beyond the ending — for continuing past the point at which the nineteenth-century plot would have stopped, and for inventing closural forms the tradition did not supply. The strategies she identifies include: displacing the romance plot with a plot of vocation; ending on the female protagonist’s relation to a community rather than to a husband; reparenting and the recovery of a maternal line; the collective or communal protagonist, which dissolves the individual settlement; and simply refusing the marriage-or-death alternative and closing elsewhere.

Three things follow for this chapter.

(1) The “mismatched final note” is sometimes a historical fact rather than a craft error. §8.4 treats a tonal mismatch at the close as a failure. DuPlessis’s work shows that some mismatches are the sound of a writer forcing available material into an unavailable shape. When a nineteenth-century novel’s heroine is married off in terms the preceding four hundred pages do not support, the diagnosis “the ending does not match the book” is correct, and the explanation is not incompetence.

(2) Closural conventions encode assumptions about what a life is for. This generalises well beyond gender. An ending in which a colonised character’s story concludes with accommodation to the coloniser, or in which a working-class character’s story concludes with respectable advancement, is doing the same structural work. Things Fall Apart is the definitive demonstration: the novel ends by handing the last paragraph to the District Commissioner, who is contemplating the paragraph he will devote to Okonkwo in his book, The Pacification of the Primitive Tribes of the Lower Niger. Achebe closes his novel by showing the closural form that would have contained it, and the effect is to make the reader feel the violence of summary itself.

(3) A writer choosing an ending is choosing among available forms, and the availability is historical. This is the point of the whole strand for a working novelist, and it is more useful than it sounds. When a draft’s ending feels wrong and every local repair fails, the diagnosis is frequently that the material is being poured into a closural shape it does not fit — usually one of two or three shapes the writer absorbed by reading. §6.5’s substitution testing is the practical remedy, and DuPlessis is the reason the exercise is worth doing at the level of the whole book rather than the last page.

The seminar question this generates is set at §12.2, no. 4, and it is a live one: whether the contemporary preference for the unresolved close is a liberation from the settlement, as its advocates hold, or merely the settlement’s successor convention, equally unexamined and equally likely to be seen through in fifty years.

3.9 Miller: the narratable, and the case against closure

Everything above assumes that closure is desirable and that the craft problem is how to achieve it well. D. A. Miller’s Narrative and Its Discontents (1981) denies the assumption, and the chapter is obliged to put the denial at full strength before answering it — because it is the strongest available argument against the whole apparatus of earned resolution, and because a student who has not felt its force will apply this chapter’s instruments mechanically.

Miller’s central concept is the narratable: the class of things that can be narrated, which he identifies with instability, deviance, insufficiency, transgression and unfulfilled desire. Narrative, on his account, arises from and consists of the narratable. What cannot be narrated is quiescence — the state in which nothing is lacking, nothing is deviant, nothing is in motion. The nonnarratable is not a subject but a silence.

The consequence is structural and, once seen, difficult to dislodge:

A novel exists because of its disequilibrium. Closure is the elimination of the disequilibrium. Therefore closure is the elimination of the novel’s own condition of existence.

Miller works this through the nineteenth-century novel and finds, repeatedly, that the ending does not so much resolve the narratable as declare it over. The marriage at the end of an Austen novel does not answer the questions the novel raised about the relation between feeling and social form; it announces that those questions have stopped being asked. The settlements of Middlemarch do not reconcile Dorothea’s ardour with the world’s provision for it; they record that the ardour has been absorbed. On Miller’s reading, the traditional ending is characteristically thinner than the book it terminates — shorter, flatter, more schematic, written in a different and duller register — and this thinness is not a failure of execution but a structural necessity. The end must be less interesting than the middle, because interest is the narratable, and the end is where the narratable stops.

Three of his subsidiary observations sharpen the argument and are stated separately here, because each has a direct craft consequence.

• The ending is where the novel’s language changes. Miller notices that closural passages frequently abandon the novel’s characteristic mode — free indirect access, scenic presentation, particularity — for summary, report and generalisation. The reader is moved from inside to outside. This is a real observation about real texts, and it is the mechanism underneath §8.5’s tidy-everything ending and §8.6’s explained ending: both are cases of a novel finishing in a register it did not use.
• Closure is retrospectively suspicious. Because the ending is where the novel becomes legible (Brooks), and because the ending is also where the novel’s disruptive energy is contained, the retroactive lens is not innocent. It reads the whole book from the position of its own containment. A reader who accepts an ending has, in a sense, agreed to interpret four hundred pages of instability as having been a detour on the way to stability.
• The demand for earned endings is therefore ideological, not technical. If closure is containment, then “earning” it means constructing the containment persuasively — making the reduction feel inevitable. The craft vocabulary of preparation, plant and payoff is, on this reading, a vocabulary for making the ending’s violence invisible.

That is the case, and it should be allowed to stand for a moment without a rebuttal attached. Miller is not a sceptic on the margins; he is describing something readers routinely feel and rarely articulate — the flatness of last chapters, the sense that a book has become smaller in the act of finishing, the disappointment that follows even a competent conclusion.

Three responses are available, and this chapter takes the third.

The first response is refusal. Accept Miller entirely and conclude that novels should not close: that the open, the fragmentary and the interminable are the honest forms. This is a real position with real practitioners behind it, and its weakness is Kermode’s: readers who fully accept the argument continue to want endings, and a theory that describes the desire as false consciousness has not explained why it survives explanation (§3.1).

The second response is dismissal. Note that Miller is describing the nineteenth-century novel specifically, that his examples are chosen for the argument, and that many endings are not thin. This is partly true and mostly evasive: the phenomenon Miller identifies is real in a very large number of books, including good ones.

The third response, and this chapter’s, is to accept the diagnosis and dispute the prescription. Miller is right that closure ends the narratable and right that the ending is where a novel’s disruptive energy is contained. He is wrong that this makes the ending a betrayal, because he has assumed that the value of a novel is identical with its narratability. It is not. A novel is also an object with a shape, and the shape is available only from the end. The reader who reaches the last page loses the narratable and gains the whole — and the whole is not a consolation prize for the loss of instability; it is a different order of thing, unavailable at any earlier point.

The chapter’s working position can therefore be stated as a concession and a claim in the same breath, and it is the qualification this chapter’s confident vocabulary most needs:

Every ending is a loss. The craft question is not how to avoid the loss but what the reader is given in exchange for it. An ending that closes the narratable and offers nothing in return is the tidy ending (§8.5), and Miller has described it accurately. An ending that closes the narratable and delivers, in exchange, a shape the reader could not previously see is what the rest of this chapter is about.

This reframing has a practical yield, which is the test of whether it is more than a rhetorical escape. It predicts that the strongest endings will be those that preserve some narratable material while closing the book — that they will leave one thing running, one question live, one relation unsettled, not as evasion but as an acknowledgement that the containment is partial. That prediction is testable against the readings in §9, and it holds: Middlemarch leaves Dorothea’s unspent ardour explicitly unaccounted for; Beloved leaves the question of what is to be done with the story unresolved by contradicting itself; Nineteen Eighty-Four’s Appendix reopens the very despair its last line closes; and the household narrative in §9.10 deliberately leaves the mother unresolved. In each case the novel closes and something is still moving.

3.10 The retroactive lens, restated — and qualified

The chapter’s first principle can now be stated in full.

The retroactive lens. The ending is the position from which the whole book first becomes available as a whole, and the reader’s final act of interpretation is therefore performed upon everything, not only upon the last pages. Because there is no subsequent unit to correct it, the ending’s account of the book is the account that stands; and because meaning in narrative is conferred from the terminus rather than distributed along the sequence, the ending reaches backwards and determines, retroactively, what every earlier element was doing.

Three corollaries, each of which generates an instrument later in the chapter.

1. An ending may spend only what the book has already deposited. (§6.1; test 2, the plant audit.)
2. An ending’s failures are not local. A weak final chapter does not damage the final chapter; it damages the book. (§1.1; test 1, the retroactive test.)
3. The ending is the only unit whose obligations cannot be known until the rest exists. Hence: revise it last, whenever it was written. (§7.0.)

And now the qualification, which is where the more interesting teaching lies and which the chapter will not withdraw.

The retroactive lens describes a mechanism; it does not license a demand. The mechanism is real: endings do reach backwards, and readers do interpret from the terminus. But the inference usually drawn from it — that an ending must therefore be earned, in the sense of resolving what the book raised — smuggles in an evaluative premise the mechanism does not contain. It assumes that the retroactive act should produce coherence. Miller’s argument (§3.9) is that this assumption is the ideology of closure rather than a fact about narrative, and Kermode’s (§3.1) is that the coherence is in any case supplied by the reader rather than discovered in the material.

The honest formulation, and the one on which the rest of the chapter proceeds:

The demand that an ending be earned assumes that closure is the goal. A serious tradition holds that closure is what kills the book. This chapter teaches the craft of earned closure because it is teachable, because the great majority of failing drafts are failing at it, and because a writer who cannot do it cannot meaningfully decline to. It does not claim that the endings which refuse it are failing. Where the model and a great novel disagree, the model is reporting its own limits — and §§8, 9 and 16 are constructed to make sure the student sees that happen at least a dozen times.