Every instrument in this chapter assumes that an ending should conclude. The plant audit, the last-decision rule and the domain count are calibrated to closure, and they will report the refusals of a serious body of fiction as failures. When they do, the model is reporting its own limits. Miller’s objection stands at the end as it stood at the beginning: closure ends the instability that made the book worth reading, and a novel’s last chapter is often thinner than its middle because the thing it ends was the thing that was interesting.
The answer this chapter has given is an exchange rather than a rebuttal. The reader at the last page loses the narratable and gains the whole, a shape that was unavailable at any earlier point, and the craft question is whether the shape is worth what it cost. That is why the tests can pass and the ending still fail. An ending that satisfies every audit and lands on nothing is reporting that the book never built a want urgent enough to conclude, and no repair to the last twenty pages will supply one.
It follows that the ending cannot be revised as a unit on its own. It is where a writer finds out what the book was about, and the finding often sends the revision back to the middle, where the pressure is built, or to the opening, where the contracts the ending must honour were made. That is the reason to revise the ending last, whatever order it was written in: its obligations are set by everything else, and only when everything else is settled can a writer tell whether the ending is failing or the book is.
See also: the Reference coda, on what the model does reliably and what it leaves open.