Self-Editing a Novel: The Four Passes, In Order

The most expensive mistake in revision is polishing a sentence in a chapter that should not exist.

Editing feels like one activity — "making it better" — and so writers do it opportunistically: fix a clunky line here, cut an adverb there, notice that the middle sags and feel vaguely worried, keep polishing. Six months later the prose is beautiful and the book still doesn't work, because the problem was never in the prose.

Professional editing is not one job. It's four, and they run in strict order, because each pass can destroy the work of the pass below it.

Pass 1: Structure

Question: does this book hold together?

You are not reading for sentences. Ideally you are not reading at all — you're looking at a summary of the book, one line per chapter, all of it visible at once.

What you're hunting for:

• Chapters where nothing changes. If you can cut a chapter and nothing downstream breaks, it isn't a chapter.
• Turns that arrive too late. The commitment to the story's central problem should happen far earlier than instinct suggests.
• Threads that go quiet. A subplot absent for eighty pages has been abandoned, whatever your intentions.
• Repetition of function. When two chapters do the same narrative work, one must die.
• The sagging middle, which is almost never a pacing problem. It's a pressure problem: nobody's situation is getting worse.

Every fix here deletes or moves chapters. Which is why nothing below this line should be touched yet.

Pass 2: Character

Question: does everyone in this book want something, and pay for it?

Now read, but read for people rather than plot.

• Does each significant character want something concrete enough to fail at?
• Do they make decisions, or do things happen to them? A protagonist who only reacts for two hundred pages is a passenger.
• Is the antagonist right about something? If not, they're weather, not a character.
• Does anyone change? Does the change cost them the thing they wanted most?
• Would the minor characters recognise themselves in how they're described? Or are they doing jobs the plot needed doing?

Fixes here rewrite scenes, not sentences. Still too early to care about adverbs.

Pass 3: Scene and prose

Question: is every scene doing work, and doing it in the right order?

Scene-level now. Read each one and ask what it's for. Whose scene is it? What do they want in it — not in the book, in this room? What's stopping them? What's different at the end?

Then the perennial faults:

• Late entry, early exit. Start the scene as late as it can start. Leave before the goodbyes.
• Explaining after showing. The paragraph after the great line that explains the great line. Cut it. It's always cuttable.
• Dialogue that agrees with itself. If two characters are cooperating, one of them can leave.
• Description that is inventory. Two specific details beat six generic ones.
• Filter words. She saw the door open → the door opened. We are already in her head.

Pass 4: Line

Question: is this the best sentence?

Last, and only last. Rhythm, word choice, repetition, the mechanical stuff a checker can find — repeated words, tics, everyone nodding before they speak, the adverb clusters.

This pass is the pleasant one. It is also the one you must not start early, because a beautifully-lined chapter is a chapter you will fight to keep, and your judgment about whether it belongs is now compromised. Sunk cost is measured in sentences.

The rule that makes it work

Never edit at a level below the highest level that still has problems.

If the structure is wrong, working on prose is not progress — it's avoidance wearing a productive face. Fix the structure, even though it's harder, vaguer, and more frightening. Every pass below it will be cheaper afterwards, because you'll be polishing a book that exists.

Where a machine helps, and where it doesn't

Passes 1 and 2 are judgment. A model can describe your book back to you — one sentence per chapter, whose viewpoint, which threads move — and that description is genuinely useful, because it's the thing you cannot see from inside. But it cannot tell you the middle is boring, because it has no taste; it has agreeableness.

Pass 4 is mechanical, and you lose nothing by automating it.

Pass 3 sits between. An honest reader, human or otherwise, can tell you that a scene's stakes are unclear. Only you can decide what they should be.

Narrale's editorial lenses are shaped around this split: separate Plot, Character, Writing, and Grammar passes over your chapters, reporting on what's there rather than rewriting it, so the four passes stay four passes — and the drafting page stays empty and quiet.

Related: Continuity errors and plot holes · Using AI to write a novel · How to use beta readers