The evidence against the model is inside the model
Everything in this chapter is an instrument for noticing. None of it is a rule, and the proof is in the chapter itself.
Ulysses fails the plot-resolution test entirely and is one of the most completely closed novels in English. The Great Gatsby resolves nothing on its last page, having resolved everything on the page before. Anna Karenina runs an aftermath that would be cut by any contemporary editor and cannot survive without it. Jane Eyre is resolved by a supernatural voice across a moor and by a settlement so comprehensive it accounts for a French schoolgirl’s education. Middlemarch ends with a narrator explaining what the book meant. Beloved closes on a sentence that contradicts itself twice and then on a single word. Blindness spends three hundred pages promising a payoff and converts it into a joke about the colour of the sky. Don Quixote ends by forbidding further books. And Great Expectations — the case this chapter treats as its centrepiece — has two endings, written by the same man, one of which almost certainly fails the tonal-match test and is the one nearly everybody prefers.
A model that ruled these out would not have identified a defect in them. It would have identified its own limits, which is what §3.10 said it would do and what §§8 and 9 were built to demonstrate a dozen times over.
There is a stronger version of the caution, and this chapter has been carrying it since §3.9. The whole apparatus of earned closure assumes that closure is worth having. D. A. Miller’s argument that closure eliminates what made a novel worth reading is not a marginal position, and this chapter’s answer to it — that the reader loses the narratable and gains a shape unavailable at any earlier point — is a defence rather than a refutation. It concedes the loss. Anyone who ends a book is choosing to stop something that was still alive.
What the model does reliably
What the instruments do, and it is not nothing, is convert a vague dissatisfaction into a specific question.
This ending isn’t working becomes: four of the six domains are being served in sequence and the reader can watch the machinery; or the resolution rests on a solicitor’s letter that arrives from outside the book; or this is a coda and the material wants a parallel scene; or every element is planted and none of them was ever under pressure. Specific questions can be answered. Vague ones can only be worried at, usually by rewriting the last page repeatedly, which is the least effective revision available because the last page is almost never where the problem is.
The most useful result the protocol can return is also its most deflating, and it is worth restating at the end because students reach it more often than any other: an ending that passes every test and lands on nothing is reporting that the book was never about anything urgent enough to conclude. The instruments cannot fix that, and they are valuable partly because they can find it. A writer who discovers, at the end of a plant audit, that everything traces cleanly and nothing matters has learned something no amount of work on the final chapter would have taught them.
Why the ending closes Part III
There is an obvious reason and a better one.
The obvious reason is symmetry. A textbook that treats the units of the novel from the scene outward has to reach the book’s second boundary last, having begun at the smallest thing that can stand alone and passed through the joins and the architecture on the way. Chapter 9 and Chapter 16 are the two edges, and putting them at either end of Part III is tidy. The two chapters after this one leave the single book’s units behind: Chapter 17 follows the ending’s promises out to the genre and the series, and Chapter 18 sets every test in the book, this chapter’s included, in the order in which a revision should run them.
The better reason is that the ending is where a novel finds out what it was.
That formulation is not mystical. It is a plain description of a mechanism this chapter has spent its whole length establishing. Until a book ends, its material is provisional: an element might turn out to be a cause, a theme, a red herring or a habit. Brooks’s account (§3.4) is that meaning in narrative is conferred from the terminus rather than distributed along the sequence, and Kermode’s (§3.1) is that the shape is supplied rather than discovered. Put them together and the consequence for a writer is exact and slightly alarming: you do not know what you have written until you have written the end of it.
This is why the ending is the only unit whose obligations cannot be established in advance, why it must be revised last whatever order it was drafted in, and why the plant audit — the one test in this book that requires the whole manuscript — could not be demonstrated until §9.10, after the chapters before it had deposited something to audit. It is also why writers so often report that the ending told them the book was about something other than what they intended. That is not a failure of planning. It is the retroactive lens operating on the person best placed to feel it.
And it is why this chapter’s most-repeated instruction is not about endings at all. When the ending will not work, the fault is very often four hundred pages upstream, and the discipline the chapter actually teaches is the discipline of looking there.
Trust minted, and trust spent
Chapter 9 argued that the opening pages are where a book’s trust is minted: page 3 has no account, and every claim it makes must be secured out of its own resources.
This chapter has argued the inverse. The ending has the fullest account in the book and cannot add to it. Everything it spends was deposited earlier, by scenes that did not know what they were depositing, in chapters that had their own business to attend to. The good pen went into a coat pocket on an ordinary afternoon because a man could not have explained why; it is worth what it is worth at the end because nobody made anything of it at the time.
That is the relation between the two boundaries, and it is the sentence the first three Parts of this textbook have been building towards:
The opening is where a novel asks to be trusted. The ending is where it finds out whether it should have been.
Around them lie the chapters on how the trust is kept: what a scene is for, what follows a scene, how people talk, what a described thing is doing there, how the reader is carried across a gap, where the gap is placed, when the past may be entered, how more than one time is braided, what a book may put before itself, how a middle holds, how a character changes and a theme is argued, who stands on the other side of the gap, how a second story earns its space, and when a reveal may come. Every one of those chapters offered a taxonomy, a set of tests, a list of failures and a set of exceptions large enough to embarrass the taxonomy. That was deliberate. Every instrument in these chapters is a way of noticing, and every model in them is a heuristic that great books violate. A student who finishes this textbook able to recite the models has acquired less than one who finishes it able to say, of a particular book, which model it breaks and what it gets in exchange.
What is left open
This textbook runs a single invented novel through every chapter — a shop, a ledger, a coat on a hook, three siblings, and a mother who changed the locks in March and never said why. The chapters before this one planted; this one audits, and §9.10.8 reruns the audit across all eighteen. Nearly everything came due: the pen, the phrase, the changed hand, the silence, the coat.
The mother did not. She has no scene of her own in any chapter, and the ending does not go to her. §9.10.7 defends that choice and concedes what it costs, and the concession is the right note on which to close a book about endings. A novel that closed every account would have argued that families are solvable. The one thing this one has consistently maintained is that they are not, and the ending’s last act of discipline is to leave the largest silence where it was.
That is a decision, not a discovery. Someone else would write the other book.
The last question
Chapter 9’s coda ends with a question to ask of a first page: does this sound like someone who knows what they are doing? It works because an opening’s subject is competence — a stranger deciding, in two pages, whether to trust the person holding the pen.
An ending’s subject is not competence, and the equivalent question cannot be about the writer. By the last page the reader has stopped assessing and started remembering. What they are doing, in the moment before they close the book, is finding out what all of it was.
So the last question of this chapter is one a writer can actually ask, and can answer without self-deception, and will usually find uncomfortable:
What did this book turn out to be about — and is that on the last page?
If the two answers match, the ending is doing its work, whatever the protocol says. If they do not, the last page is not the problem and never was.
And if the question cannot be answered at all, that is the most useful result available, and it arrives at the only point in the process where there is still time to do something about it: at the end, looking back, which — as every chapter of this book has argued in its own way — is where the shape of a thing finally becomes visible.