8. Failure Patterns

Each pattern below is presented with its symptom, its mechanism (what actually goes wrong, as distinct from what the surface looks like), its persistence (why competent writers keep producing it), its remedy, and — importantly — its exception: a published work that commits the “error” and succeeds.

The exceptions are not decorative, and in this chapter they carry an unusually heavy load. Endings are the site of the most durable critical disagreements in the discipline, and several of the “failures” below are committed by works whose endings are among the most admired in the language. In every case, the pattern fails because of a mechanism, not because of a surface feature, and where the mechanism is absent the surface feature is harmless.

8.1 The Unearned Rescue

Symptom. Resolution arrives through an element the book never integrated: a sudden power, a stranger’s intervention, a document that appears, a fact withheld from the reader until the moment it is required.

Mechanism. The ending’s causal chain is broken. Aristotle’s formulation (§3.3) is exact and better than the modern one: the objection is not to improbability but to externality. The resolution is produced by something outside the causal system the book built, which means the preceding four hundred pages did not, in the end, determine the outcome. The retroactive damage is severe because it applies to the whole: if the ending did not follow from the book, the book was not going anywhere.

A secondary and equally damaging effect is that the protagonist is removed from their own ending. A rescued character has not acted; and since character is legible chiefly through choice under pressure, an unearned rescue does not merely resolve the plot badly — it dissolves the character domain at the exact moment it was due.

Persistence. Three causes, all structural rather than moral. The writer has painted the situation into a corner and the rescue is the only exit. The writer has fallen in love with a set-up whose logic produces an outcome they do not want. Or — most commonly — the writer has not noticed that the resolution requires a fact the reader does not have, because the writer has it.

Remedy. Test 3. Complete the sentence *the situation is resolved because ___* and check whether the thing in the blank belongs to the book’s causal system. Where it does not, there are two moves, and the first is nearly always right: plant it earlier, unremarked, so that the intervention becomes the arrival of something the reader already knew existed; or change the outcome to one the existing system produces.

The exception, and it is a large one. The deus ex machina is Euripides’ signature device, not his failure. Medea ends with Medea escaping in the chariot of the Sun; several other plays resolve by divine intervention. Aristotle disliked it and said so in the Poetics — but Aristotle was expressing a preference against a living, popular and deliberate practice, and later criticism has increasingly read Euripides’ machines as ironic rather than lazy: a god arrives to impose an order the human action has shown to be unavailable, and the visible artificiality of the arrival is the point.

The line from there to modern practice is direct and explicit. John Gay’s The Beggar’s Opera (1728) halts before Macheath’s execution so that the Beggar and the Player can argue about the ending, the Player insisting that the audience requires a happy one — whereupon Macheath is reprieved. Brecht and Weill’s The Threepenny Opera (1928) makes the device explicit: a mounted messenger arrives with a royal pardon, and the text observes how seldom such riders come in reality. In both, the unearned rescue is the argument. The device is not a failure when the arbitrariness is the content.

A third and different exception, from inside the realist novel: Jane Eyre (1847) resolves its central dilemma when Jane, on the point of accepting St John Rivers, hears Rochester’s voice calling her name across an impossible distance. This is supernatural intervention arriving at the precise moment of decision. It works — and generations of readers have found it the novel’s most powerful passage — because Brontë has spent the whole book establishing a world in which presentiment, dream and vision are operative, and because the voice does not decide for Jane; it interrupts a decision she then makes herself. The rescue is unearned in mechanism and earned in preparation, which is the discrimination the pattern requires.

8.2 The Climax Without Aftermath

Symptom. The book ends at the peak of conflict. The plot concludes and stops; character, theme, catharsis and the final image are left technically unaddressed.

Mechanism. Only half the structure has been executed. In Aristotle’s terms (§3.3) the desis is complete and the lusis absent. The consequences are specific rather than general: a value shift occurring at maximum intensity is felt but not registered (Chapter 1’s terms), because registration requires a beat at lower pressure; catharsis cannot occur, because release requires the pressure to have stopped; and the reader, transported, is not returned (§3.6).

The characteristic reader response is diagnostic and peculiar: satisfaction and dissatisfaction at once. That was exciting. Is that it?

Persistence. It is the correct fear applied at the wrong point. Writers are warned, rightly, against the endless aftermath (§8.3) and against the Victorian wind-up, and the contemporary literary norm for aftermath is the shortest in the historical range (§6.6). A writer avoiding one error over-corrects into its opposite. It is also, more simply, a stamina problem: the climax is the most exciting thing to write and the aftermath the least, and drafts frequently stop where the writer’s energy did.

Remedy. Not more climax. Ask what only the aftermath can do (§6.6’s first table), and supply the minimum that does it. Very often this is one scene and sometimes one paragraph. The operational rule: find the last decision — the aftermath’s job is to reach one, and it may be very small.

The exception. The Handmaid’s Tale (1985) ends its narrative proper with Offred stepping into the van, into the darkness within or else the light, with no aftermath whatever. It is a pure arrested ending (§6.4, no. 12) and the four unserved domains are left unserved — by the narrative. Atwood then supplies them at a two-hundred-year remove, in the Historical Notes, which deliver plot (Gilead fell), theme (the handling of women’s testimony), formal closure (§4.6) and a bitter substitute for catharsis. The lesson is not that aftermath is dispensable but that it may be displaced: the domains are served, in another register, by a device the novel’s formal contract permits.

The Grapes of Wrath (1939) is the cleaner case. It stops on Rose of Sharon’s gesture with no aftermath at all — no consequence, no future, no commentary — and it is complete, because the gesture is the aftermath: an act of consequence rather than an event of conflict. A climax that is itself a decision does not require a further beat to reach one.

8.3 The Endless Aftermath

Symptom. So much material follows the climax that momentum dies. The reader’s attention is exhausted before the last page.

Mechanism. Narrative interest is generated by unresolved tension; after the climax there is none, and the aftermath is therefore running on residual goodwill, which decays. The specific damage is that the actual final image arrives to a reader who has stopped attending, so the book’s most privileged position is spent on someone who is skimming.

Chapter 7 (Backstory and Flashback) supplies the precise concept in the interest deficit: a passage that does not generate its own forward pull is borrowing against reader patience, and the debt compounds. An aftermath is by definition low-tension material, and long aftermaths are therefore the largest single interest deficit most novels run.

Persistence. Two causes. The first is the writer’s reluctance to leave a world they have lived in for years — a real and sympathetic motive that produces real damage. The second is a misunderstanding of obligation: the belief that every thread must be accounted for (§8.5 is the same error at a different scale).

Remedy. §6.6’s rule. Locate the last decision any character takes; end within a page or two of it. Then apply the §5.5 audit and check whether the final pages are discharging domains or merely continuing.

The exception, and it is essential. Anna Karenina’s Part Eight is an endless aftermath by any modern standard: an entire eighth part, after the death that ends the title character’s story, largely devoted to Levin’s spiritual crisis and its resolution, plus Vronsky’s departure for the Serbian war and a good deal of argument about Slavic questions. Readers and editors have complained about it since publication, and Tolstoy’s own publisher declined to print it in the serial.

It is nonetheless indispensable, and §9.4 argues the case in detail. The short form: the novel has two arcs, and the second cannot complete at the first’s climax. A book that ended with Anna under the train would be a different and much smaller novel — one about a woman’s destruction rather than about two ways of living with the same problem. The general principle the exception establishes:

An aftermath is not too long if it is completing an arc. It is too long if it is reporting consequences.

War and Peace pushes the same principle to its limit with two epilogues, the second of which is not narrative at all but a treatise on historical causation. That one is harder to defend and is frequently skipped, which is a fair verdict — but note what it demonstrates: an aftermath can fail and leave a great novel undamaged, because a reader who abandons an epilogue has still read the book.

8.4 The Mismatched Final Note

Symptom. The last image or line does not match the tone the book has earned — a grim novel closing on uplift, a comic one closing on unmotivated darkness, a quiet one closing on a flourish.

Mechanism. A genre-contract violation at the exit. Chapter 9 §4.2 establishes that the opening pages contract with the reader about what kind of book this is; the ending is where that contract is finally honoured or broken. Because the ending occupies privileged position and is read retroactively, a mismatched note does not merely disappoint — it reassigns the book, and the reader leaves with an impression at odds with the four hundred pages they actually read.

The symmetry with Chapter 10 is exact and is recorded in this textbook’s map: Chapter 10 §8.2’s false advertising and this chapter’s mismatched final note are the same violation at the two ends of the book. One promises a book the reader will not get; the other retracts a book the reader has already had.

Persistence. The dominant cause is not craft failure but authorial fatigue and authorial mercy. A writer who has spent years inside a bleak book wants to be released from it, and grants the release on the last page. A secondary cause is external: editorial or market pressure towards a more consoling close, which is §8.12’s territory and is very old.

Remedy. Test 8. Read the last page against three random middle passages and ask whether they belong to the same book. Where they do not, change the ending, not the book — the ending is twenty pages and the book is four hundred.

The exception, and it is the one the chapter returns to most. The revised ending of Great Expectations (1861) is arguably a mismatched final note, and most readers prefer it. Dickens’s original ending has Pip and Estella meet briefly in a London street, exchange a few words, and part; the revised ending places them in the ruined garden at Satis House and closes on Pip seeing no shadow of another parting. On the tonal-match test, the case against the revision is strong: the novel has spent five hundred pages dismantling Pip’s expectations, and a close that grants him the one expectation he should have lost can be read as the book conceding what it argued.

And yet the revision is the version nearly every reader has read, it is the version most editions print, and a great deal of serious criticism defends it as the better ending. §9.1 sets out the disagreement in full and declines to resolve it. What the case establishes here is a limit on the test: the tonal-match test is a strong instrument for detecting a change of register and a weak one for adjudicating it. A tonal shift at the close can be a failure of nerve or a final modulation the book has earned, and no mechanical procedure separates them.

8.5 The Tidy-Everything Ending

Symptom. Every subplot and question resolved with equal, mechanical thoroughness, regardless of what mattered.

Mechanism. Uniform resolution destroys hierarchy. Emphasis is relational: a thread’s resolution lands as significant because other threads’ resolutions are smaller or absent. An ending in which everything is settled to the same degree communicates that nothing was more important than anything else — which contradicts the weighting the book spent four hundred pages establishing.

This is Chapter 9 §8.5’s overstuffed first page, at the other end of the book and with a different surface. There, uniform density destroyed emphasis; here, uniform resolution destroys hierarchy. The underlying principle is the same for both: significance is produced by difference, and a text in which everything is significant contains nothing significant.

There is a second mechanism, which is D. A. Miller’s (§3.9): the comprehensive settlement is where a novel most visibly administers its own material, and the register shift from scene to inventory is what makes last chapters read as thinner than the books they end.

Persistence. A sense of obligation, and it is not baseless: readers do ask what happened to people. The obligation is real and the response is disproportionate. It is also, in genre fiction, an explicit convention that many readers want.

Remedy. Cut resolutions; do not add them. The productive exercise is subtractive: remove the resolutions of the two least-weighted threads and read the ending again. Almost always, nothing is lost and the remaining resolutions gain force. Where something is lost, that thread was more load-bearing than the audit suggested and should be restored.

The exception. Jane Eyre (1847) resolves everything, at length, in a final chapter that opens with the most famous settlement sentence in English — Reader, I married him — and proceeds to account for Adèle’s schooling, Diana and Mary’s marriages, Rochester’s partial recovery of sight, the birth of a son, and finally, in a passage occupying the novel’s last position, St John Rivers’s missionary death in India. By modern standards it is the tidy-everything ending in its pure form.

It succeeds for two reasons that generalise. First, the resolutions are not uniform: they are graded, and the grading is a moral argument, culminating in the deliberate strangeness of giving the novel’s final words not to Jane’s happiness but to St John’s fatal self-denial. Second, and more fundamentally, the novel is a first-person retrospective narration by a woman explicitly addressing a reader — a form in which the accounting for everybody is what the narrator would do. The tidiness is in character.

The general rule the exception yields: comprehensive resolution is safe when it is graded, and when the comprehensiveness is motivated by the narration rather than by the author’s sense of duty.

8.6 The Explained Ending

Symptom. A character, or the narrator, summarises what the book has meant. Often in dialogue; often in a penultimate scene; often beginning I suppose what I’ve learned is.

Mechanism. The book’s thematic work is transferred from the reader to the text. Theme, per §4.3, is a question the novel tests; an explanation converts it into a proposition the reader may accept or reject, and either response is worse than the experience the novel could have produced. There is also a specific structural harm: an explanation is a resolution, so the ending’s most privileged position is spent on a paraphrase of the book rather than on the book.

The second mechanism is a voice-contract violation (Chapter 9 §4.1). A novel that has been rigorously scenic and then, on its last pages, produces a narratorial generalisation has changed its terms at the moment it can least afford to.

Persistence. Anxiety, and it is the most sympathetic motive in this section. The writer has spent years on a book whose meaning is subtle and cannot bear the thought that a reader might miss it. The explanation is insurance. It is also the case that explanations are easy to write at the end of a book, because by then the writer knows what it means.

Remedy. Delete the explanation and read the ending without it. In most drafts, by this book’s estimate, nothing is lost. Where something is lost, the missing content should be converted into an act or an object per §4.3, not restored as speech.

The exception. Middlemarch (1871–72) ends with a Finale in which the narrator explains, in general terms and at some length, what the reader should conclude — about Dorothea, about the conditions of a woman’s ardour, and about the moral economy of unrecorded lives. It is one of the greatest endings in English (§9.2).

Two things license it. First, the narrator has been generalising for eight hundred pages: the explanatory register is not introduced at the close, it is the book’s native mode, and its arrival at the end is a formal completion rather than a rupture. Second — and this is the discrimination the pattern actually needs — Eliot’s generalisation does not paraphrase the novel; it extends it. The Finale does not tell the reader what Dorothea’s story meant; it tells them what cannot be known about lives like hers, and hands them a proposition the preceding book has made available but has not stated. The failure is not explanation as such. The failure is explanation that replaces the book with a summary of itself.

Anna Karenina is the harder case and worth setting beside it: Tolstoy ends on Levin explicitly formulating what he has come to believe, in his own consciousness, at length. Whether this is thematic closure or authorial preaching is a live critical question, and it is set as a seminar question at §12.2.

8.7 The Ending That Resolves the Wrong Question

Symptom. The plot concludes, comprehensively and competently, and the reader closes the book with the sense that the thing they cared about was not the thing that got settled.

Mechanism. A mismatch between the novel’s stated dramatic question and its operative one. Most novels raise a plot question early and then develop, over their length, a different and deeper interest — usually relational or moral. If the ending answers the first and ignores the second, every domain is technically served and the reader’s actual investment is unaddressed.

This is a failure of the ending to notice what the book became, and it is closely related to §6.1’s Claim 2. It is also the pattern that most often produces the reader complaint it was fine, I don’t know, it just ended.

Persistence. The stated question is the one in the outline, the synopsis and the pitch. The operative question emerged during drafting, is unrecorded, and is frequently invisible to the writer, who is still working to the plan.

Remedy. Ask two readers, separately: what did you want to happen? Not what did you expect — what did they want. The answers identify the operative question, and they are frequently not the plot’s.

The exception, which is the most contested ending in American literature. Adventures of Huckleberry Finn (1884) spends its length on a moral question — whether Huck will betray Jim — and answers it definitively in the celebrated passage in which Huck decides to go to hell rather than turn him in. The novel then continues for a dozen chapters at the Phelps farm, in which Tom Sawyer stages an elaborate and cruel pantomime of a rescue for a man who is, it emerges, already legally free.

Hemingway’s verdict — that the book should be read up to the point where Jim is stolen and that the rest is cheating — is the standard indictment, and the standard version of this failure pattern: the novel resolves an adventure and abandons its moral question. But the defence is serious and long-standing. T. S. Eliot and Lionel Trilling both argued for the ending’s necessity; and a substantial modern reading holds that the Phelps farm sequence is the point rather than the evasion, since it dramatises what Jim’s freedom is worth in a society that will convert his liberation into a boy’s game. On that reading the novel does not resolve the wrong question; it answers the right one with a bleakness that readers have wanted to escape for a century and a quarter.

The case is set as a seminar question at §12.2. What it establishes here is a caution: the accusation that an ending resolves the wrong question is sometimes the reader refusing the question the book actually asked.

8.8 Death as Resolution

Symptom. A central problem is resolved by the death of a character — usually the character posing the problem, occasionally the protagonist.

Mechanism. Death is the most available terminal event in fiction and the most frequently substituted for a decision. Its structural attraction is its weakness: it is unarguable, it is final, and it requires nothing of anyone. When a novel resolves through a death that no character chose and none of the book’s forces produced, the ending has been outsourced to mortality.

E. M. Forster’s remark in Aspects of the Novel (1927) that without death and marriage the novelist would scarcely know how to conclude is usually quoted as a joke, and it is a diagnosis: these are the two events the form treats as automatically terminal, which makes them the two most likely to be used as substitutes for structure.

The specific harm is to the character domain. A character who dies has not chosen anything, and the reader’s accumulated investment in their capacity to choose is discharged into an event rather than an act.

Persistence. It works, at the level of feeling, almost regardless of preparation, because readers respond to death. This makes it the most reliable way to produce an emotional effect at a close without having earned one, and therefore the most tempting.

Remedy. Test 3 and the substitution question: if the character had lived, would the book’s central problem be resolved? If not, the death is doing structural work it should not be doing. The usual correction is to place the death before the ending, so that the ending contains a decision made in its light.

The exception. The Death of Ivan Ilyich (1886) is a novella in which death is the entire content, announced in the first pages and arrived at in the last, and it is not a failure because the death is not a resolution — it is the situation the whole work examines, and what resolves is Ivan’s relation to it. The discrimination generalises: death fails as an ending when it resolves; it succeeds when it is the condition within which something else is resolved.

Don Quixote (1605–1615) is the more interesting case and is treated in full at §9.8. Cervantes ends by killing his protagonist — and Alonso Quixano’s death is not the end of a plot but the withdrawal of a capacity: he recovers his wits, renounces the books of chivalry, makes a will, and dies as a man who can no longer be Don Quixote. The death resolves nothing in the plot and everything in the book, because what dies is the possibility of further adventures. §9.8 argues that this is closure by ending the character’s availability, and that it is one of the very few endings in fiction designed to prevent other people from writing more.

8.9 Ambiguity Used as Evasion

Symptom. The book declines to settle its central question, and the refusal reads as an absence of decision rather than as one.

Mechanism. Per §3.2, an open ending is a closural strategy that works by withholding devices the reader knows are available. It therefore requires the withholding to be legible. When the withholding is not legible — when there is no formal signal that the openness is chosen — the reader cannot distinguish it from the writer not knowing.

The historical dimension (§3.7) makes this worse than it used to be. In 1925, an unresolved close was a deviation from a strong convention and therefore self-marking. In the present, the unresolved close is itself the convention in literary fiction, which means it no longer signals anything by its mere presence. Ambiguity must now be specific — the reader must be able to say what is undetermined and why the book cannot determine it — or it reads as default.

Persistence. It is the easiest ending to write, it is fashionable, and it is unfalsifiable: a writer challenged on it can always claim intention, and no reader can prove otherwise.

Remedy. Two questions. What, in one sentence, is ambiguous? — an ambiguity the writer cannot state in one sentence is not a design. And what would resolving it cost the book? — if the answer is “nothing”, the ambiguity is not doing work. Where the ambiguity survives both, mark it: some formal signal, some registration that the question exists and is being left (§4.1).

The exceptions, and there are two kinds. The first is the ambiguity that is the subject. Blindness (1995) ends by withholding the confirmation the reader has been waiting three hundred pages for, and the withholding enacts the novel’s argument about seeing (§9.9). Beloved ends by contradicting itself about whether the story should be passed on, and the contradiction is the only honest position available on its subject (§9.6). In each, the reader can state what is unresolved.

The second kind is more instructive because it is contested at source. Villette (1853) ends by declining to say whether M. Paul survives his voyage: Charlotte Brontë intended to drown him, her father wished for a happy ending, and the published close is a deliberate compromise which describes a storm, invites the “sunny imagination” to hope, and stops. It is an ambiguity produced by external pressure — the definition of §8.12’s imposed ending — and it is nonetheless one of the most admired closes in Victorian fiction, because the compromise happens to produce the refusal the novel’s material requires. The origin of an ending is not evidence about its quality, which is a point the Great Expectations case (§9.1) makes at greater length.

8.10 The Last-Minute Reversal

Symptom. A disclosure in the final pages changes the meaning of the whole book: the narrator was lying, the events did not occur, the protagonist is not who they appeared to be.

Mechanism. The strategy is legitimate — it is no. 13 in §6.4 — and it fails in a specific and identifiable way: when the reversal replaces the reading experience rather than adding a layer to it. A reframe that invalidates everything before it converts four hundred pages into a set-up for a trick, and the reader who feels this has been given a worse book than the one they were reading.

The diagnostic is test 9 (the reread test). A reframe must make the second reading better, not merely different. If the earlier chapters yield less on rereading, the reversal spent the book.

A secondary failure is that a reversal is enormously expensive in trust: it teaches the reader that this narrator’s assertions need not be believed, at the exact moment the book most needs to be believed about its own conclusion.

Persistence. The reversal is intoxicating to plan and produces, in the writer, a strong feeling of cleverness that does not survive contact with a reader.

Remedy. Test 9, and the substitution question in its strongest form: would this book be worth reading if the reader knew the reversal from page one? Where the answer is no, the reversal is carrying the book. Where it is yes, the reversal is doing what it should.

The exception. Atonement (2001) reverses in its final section: the novelist Briony discloses that the reunion of the lovers, which the reader has just read, did not happen, and that she has written it as a form of restitution. This is a maximum-strength retrospective reframe, and it survives test 9 outstandingly — the novel is substantially better on a second reading, because the reader now sees the earlier sections as a made thing and can watch the making. What makes it work is that the reframe is the book’s subject: the ethics of a novelist granting a consolation she cannot grant in fact. The reversal does not invalidate the reading; it interrogates it.

Contrast The Murder of Roger Ackroyd (1926), where the reversal is a formal joke of the highest order and the reread yields the pleasure of watching the concealment operate — but where nothing thematic is added. Both are successful; they succeed at different things, and the difference is worth teaching.

8.11 The Series Hook

This edition calls the pattern the series hook, since the book keeps sequel for the reaction, dilemma and decision of Chapter 2.

Symptom. The book declines to close a major question in order to create an appetite for a further volume. A last-page arrival, a survivor revealed, an unopened letter, a returning threat.

Mechanism. The strategy converts a structural position into a commercial one. The cost is precise: the ending’s most privileged position (§3.6) is spent on a proposition rather than a payoff, which reverses the whole relation this chapter describes. Instead of closing accounts, the last page opens one — and because of the retroactive lens, the reader’s final act of interpretation is applied not to the book they read but to a book that does not yet exist.

The secondary harm is to the formal domain (§4.6). A hook declares that this artefact is not complete in itself, which retrospectively demotes it from a book to an instalment.

Persistence. Commercial pressure, and it is real: series sell, and publishers ask for hooks. It also feels generous — the writer believes they are offering the reader more.

Remedy. Distinguish an open question from a hook. A novel may end with things unresolved and be complete, provided the unresolved material is not the thing the book was about. The test: does the last page close the volume’s own central question? If yes, a further question may be raised. If no, the volume has not ended.

The exception. Don Quixote, Part I (1605), ends with an explicit promise of a third sally and an invitation to whoever might write it. That invitation was accepted — by Avellaneda, whose spurious continuation appeared in 1614 — and Cervantes’ response in Part II is a celebrated reaction to a series hook (§9.8). The episode is worth teaching because it shows the hook doing what a hook does: declaring the work incomplete, and thereby licensing others to complete it.

The serial novel is the systematic exception. Dickens’s part-issues, documented in the number plans discussed in Chapter 6 (The Chapter Break), end on hooks by design, and the practice is not a defect but the form’s engine. What distinguishes the serial instalment from the failing standalone is that the serial’s reader has contracted for continuation. Chapter 9 §10.4 makes the same observation about series openings: the conventions of a form the reader has entered knowingly are not violations.

8.12 The Imposed Ending

Symptom. The ending is congruent with something other than the book: a genre convention, a publisher’s preference, a reader’s expectation, an author’s fatigue, a moral requirement external to the material.

Mechanism. This is Torgovnick’s category (§3.5), and it is the most abstract failure in this section because it can coexist with technical competence. An imposed ending may plant everything it uses, serve all six domains, land its final image, and still be wrong — because what it does with the material contradicts the book’s own argument.

It is worth distinguishing carefully from the unearned rescue. An unearned ending uses material the book did not supply. An imposed ending uses the book’s own material to reach a conclusion the book does not support. The first is a failure of preparation; the second a failure of congruence.

Persistence. Every external pressure on a novel arrives at its ending, because the ending is where a book declares its position and positions are what publishers, readerships and censors care about. The pressure is frequently invisible in the finished text and legible only in the archive.

Remedy. Test 1 and test 8 together. Where the ending strands earlier material and shifts register, imposition is the likely diagnosis. The corrective question is Torgovnick’s: whose logic does this ending obey?

The exception, which is also this chapter’s centrepiece. The revised ending of Great Expectations is an imposed ending in the documentary sense: Dickens wrote one close, showed the proofs to Edward Bulwer-Lytton, who objected, and wrote another. The textual history is documented in the scholarly editions, and Dickens himself recorded his expectation that the alteration would make the story more acceptable.

By the criterion above, that should settle it. It does not. The revised ending is the one nearly every reader has read, and a century and a half of criticism has failed to reach agreement on which is better — with distinguished advocates on both sides, including George Bernard Shaw, who printed the original in his own edition, and equally distinguished defenders of the revision. §9.1 sets the case out in full.

The general principle, which qualifies every pattern in §8:

The provenance of an ending is not evidence about its quality. An ending written under pressure may be the right one; an ending written in perfect freedom may be imposed by a convention the author never examined. Congruence is a property of the relation between ending and book, and it must be assessed by reading, not by biography.

8.13 Summary: symptom to mechanism