The ten tests are for revision, not composition, and here more than anywhere: the ending is the one unit whose obligations cannot be known until the rest of the book exists, and a writer who runs the plant audit while drafting will write a book made of plants. Run them in order, because tests 1 to 4 (Retroactive, Plant audit, Earned and Inevitability) concern the ending’s relation to the book and a failure there makes the rest premature. Tests 1 (Retroactive), 2 (Plant audit), 6 (Proportion) and 9 (Reread) need the whole manuscript, and test 4 (Inevitability) needs other readers, which is why they are the ones writers skip. Record a sentence, not a verdict. An ending written first is a hypothesis about the book, and the book that got written may have falsified it: revise the ending last, whenever it was written.
Test 1 — Retroactive
Reread five early scenes chosen at random: what has the ending made of each?
Failure. Stranded scenes; either the ending or the scene is wrong, and the choice should be explicit.
Exception. The cetology chapters of Moby-Dick, which no ending converts and readers keep.
Test 2 — Plant audit
List what the ending rests on and trace each item to its first appearance: does anything load-bearing appear first in the ending?
Failure. Unearned in the strict sense; usually mended by one sentence early.
Exception. Dickens’s original ending to Great Expectations, which introduces a Shropshire doctor and a horse and has distinguished defenders.
Test 3 — Earned
Complete the situation is resolved because…: is the blank inside the book’s causal system?
Failure. New information, a new capability or, worst, a new agent.
Exception. The Threepenny Opera (1928), whose mounted messenger with a royal pardon is the play’s argument.
Test 4 — Inevitability
Can a first-time reader predict the end thirty pages out, and can a finished reader build an alternative as good?
Failure. Alternatives come easily: the ending is one of several the book never distinguished.
Exception. Great Expectations again, where the author wrote the alternative and both endings are defended.
Test 5 — Domain
Which domains does the ending deliver, which does it gesture at, which does it refuse, and does the distribution suit the book’s mode?
Failure. A domain credited on the writer’s intention rather than a passage a reader could point to.
Exception. The Grapes of Wrath, which refuses plot resolution outright.
Test 6 — Proportion
What share of the book follows the climax, does each thread’s resolution match the space it took, and how far does the book run past its last decision?
Failure. Climax without aftermath, endless aftermath, or every thread settled at equal length.
Exception. Room (2010), whose escape comes about halfway, leaving half the novel to the aftermath.
Test 7 — Termination
Ask a reader of the last two pages whether anything comes after, and what told them.
Failure. A book that stops because the pages run out.
Exception. Bleak House, whose last sentence breaks off unfinished and still closes.
Test 8 — Tonal match
Read the last page against three passages from the middle third: are they the same book?
Failure. Uplift the material does not support, or a gratuitous darkening.
Exception. Madame Bovary, which follows two deaths with the pharmacist’s medal, a change of note that is the verdict.
Test 9 — Reread
Reread three passages the ending changes: do they now yield more?
Failure. An effect that depended on the reader not knowing.
Exception. The Murder of Roger Ackroyd (1926), whose reread yields the pleasure of the concealment and nothing more.
Test 10 — Last line
Could the book end on the paragraph before the last without loss?
Failure. The last paragraph coasts on the book instead of adding to it.
Exception. Mrs Dalloway, whose For there she was means nothing alone: dependence is a last line’s condition, coasting its fault.
Reading the results together
Failures cluster. Tests 2 (Plant audit) and 3 (Earned) together mean an imported element or agent; plant it in the first third, unremarked. Test 4 (Inevitability) alone, with readers producing good alternatives, means the middle never narrowed the material, and the repair is in the middle. Test 1 (Retroactive) alone means the ending strands earlier scenes; decide which is wrong, and do not mend both. Test 6 (Proportion) split between a low aftermath and failing domains means the book needs its lusis, not more climax; a high aftermath means finding the last decision and ending within a page of it; uniform resolution means cutting resolutions rather than adding them. Test 7 (Termination) alone usually wants a formal signal, a returned image or a terminal modification, not new material. Test 8 (Tonal match) alone means changing the ending, not the book: the ending is twenty pages and the book four hundred.
The hardest result is a clean pass on every test and an ending that lands on nothing. It is the commonest outcome of a competent revision, and the tests cannot repair it, because they audit the ending and the fault lies in the middle of the book, which has not built a want urgent enough to conclude. That is a real diagnosis, and The Structural Edit (Chapter 18) starts from it.
See also: the Reference protocol, with each test’s procedure and a table of combined failures.