Novel craft and story design

How to keep a long story consistent

August 16, 2026 · The StoryProp team

Somewhere past the middle of a long draft, the small facts start to slide. A scar moves from the left hand to the right. A character who buried her phone in a drawer in chapter six answers it in chapter nine. A minor reverend is Ellison for a hundred pages and Ellisen after that. None of this is a failure of imagination — the imagination is busy elsewhere. It is a bookkeeping problem, and bookkeeping problems have bookkeeping solutions. Scale is what makes the system necessary: FriesenPress's analysis of Amazon's top 50 bestsellers puts the average book at 398 printed pages across 2020 to 2024, and rising — 378 pages in 2020, 442 in 2024. That is an estimate rather than an industry statistic, but every page added is more surface for a fact to slip on.

This piece is about the practice of continuity: what actually drifts in a long story, the short documents that catch drift early, and how to repair a contradiction without tearing new holes. It is craft-side advice, and every part of it is bookkeeping that something else can carry for you.

What actually drifts (and why it's the middle that goes)

Drift concentrates in four places. Physical facts: appearance, objects, geography — the chipped cup that lives on Mara's shelf in one chapter and beside the sink in another, the staircase that changes sides of the house. Names and spellings, especially for minor characters, streets, and businesses. Timelines: seasons that stall, injuries that heal overnight, a pregnancy that quietly runs fourteen months. And knowledge states, the subtlest of the four: who has learned what, and when. When Daniel discusses the contents of a letter no scene ever showed him reading, most readers can't name what's wrong — they just stop trusting the book.

Notice what doesn't drift. You will not forget your protagonist's name, the death at the heart of the plot, or the town the story lives in. Those facts get rehearsed every session you write. Drift lives in the middle tier: details important enough that a reader registers them, minor enough that you never repeat them to yourself. That is exactly why raw memory fails as a strategy — the middle tier is what memory drops first — and why the fix is a handful of small, boring documents.

The canon list: facts the story has committed to

A canon list is a running record of facts your story has committed to on the page — not what you are planning, not what your notes wish for, but what a reader could quote back at you. The distinction matters. Your worldbuilding file might say Mara left town at eighteen; if chapter four has her saying “I was seventeen and furious,” then seventeen is canon now and the note is just history. Each entry should carry the fact and where it was established: “Mara's kitchen faces east — ch. 3, morning light on the chipped cup.”

Keep it ruthless. A canon list that tries to hold everything becomes a second manuscript nobody reads. Add an entry when the page commits to something you will plausibly need to keep straight later: physical layout, dates, wounds, objects with histories, anything numeric. Update it at the end of each session, while the day's pages are fresh. If you have been wondering how to keep track of characters and storylines without drowning in paperwork, this is most of the answer — one short document, kept current, honest about what the book has actually said.

The same discipline applies to research. Wherever your notes live — a folder of clippings, a spreadsheet, an AI tool that organizes novel notes — keep what you have found separate from what the story has committed to, because evidence can be revisited but canon has witnesses. And if you draft inside a general chatbot such as ChatGPT, Claude, or Gemini, know that the tool has its own version of this problem; why chatbots lose the plot covers those mechanics, and they are not a moral failing on your part or the machine's.

Timeline discipline and the who-knows-what table

Two more ledgers earn their keep. The first is a timeline, and the trick is to count days rather than assign dates. “Day 1: Daniel arrives. Day 3: the funeral. Day 9: Mara finds the letter.” Relative time is what readers actually track, and it is what breaks: a chapter opener that says “two weeks later” when the count says four days, a bruise that fades before it could, a Sunday service three days after the last one. Pin a weekday or a calendar date only when the story commits to one on the page — and then treat it as canon like anything else.

The second is a who-knows-what table: secrets and revelations across the top, characters down the side, each cell holding knows, suspects, or doesn't know, plus the scene where that changed. It sounds fussy until it catches its first error, which is usually a character acting on knowledge they don't have — or, stranger, failing to act on knowledge they do. If Daniel learns in chapter eight that the cup was his to break, every later scene in which he watches Mara wash it means something different. And every earlier scene must not.

Plot holes are broken promises

A plot hole is a promise broken: the story committed to a fact, then behaved as if it hadn't. The alibi that depends on a train the timeline says had already left. The locked house holding a man nobody let in. The question of how to fill plot holes has a procedure, and it starts with stating the hole as two sentences that cannot both be true: “The house was locked from the inside. Daniel was inside without a key.” Until you can write both sentences plainly, you do not understand the hole yet.

Then trace what leans on each sentence. List every scene that depends on the locked house, every scene that depends on Daniel's presence, and choose the commitment the story needs more — usually the one carrying the most weight downstream. Rewrite the scenes that lean on the losing sentence, and record the repair in the canon list. The classic mistake is the local patch: one explanatory line inserted at the site of the hole — a hidden key, a conveniently unlatched window — that quietly contradicts a third scene you forgot was watching. Fills that don't consult the ledgers dig new holes.

When the new page is right

Contradiction is a signal, not automatically an error. Sometimes the fresh page fights the canon list because the page is better. Suppose canon says the cup was chipped in the move, established in a throwaway line of chapter two — and then, in a scene you draft months later, Daniel confesses he chipped it the night he left, and the confession is the best moment in the book. The new page wins. But it wins by decision, not by drift: flag the conflict, choose deliberately, then sweep every earlier reference so the book agrees with itself again.

The rule is that canon serves the story, and changes to canon happen on purpose, in daylight. That is also the test worth applying to any AI tool that claims to check story consistency: when the new page and the old page disagree, does it surface the contradiction and let you rule, or does it silently “fix” one side? A tool that quietly harmonizes your manuscript is making canon decisions without you. StoryProp meets that test the plain way: it surfaces the contradiction and leaves the ruling to you. What a writing tool can genuinely remember is a separate question with a precise answer — what manuscript memory actually means sorts those claims out.

The consistency pass: how to organize a messy manuscript

Before you revise for prose, read once for continuity alone. This is the fastest way to organize a messy manuscript: a single quick pass in which you fix nothing and log everything — every name, object, date, wound, and secret — against your canon list, timeline, and who-knows-what table, building them as you go if they don't exist yet. Mid-project is not too late. The ledgers are easier to reconstruct from a draft than to invent from nothing, because the draft has already made its commitments; you are taking dictation from your own book.

The pass produces a punch list: contradictions found, decisions needed. Resolve the decisions first — which version of each disputed fact is canon — then make the fixes, tracing each one the way you would fill any plot hole. Only after that, revise for language. Sentence-level revision keeps returning to the same scenes, and if the facts underneath are still moving, you will polish contradictions until they shine.

This division of labor is where software can carry real weight without taking the wheel. StoryProp's approach is to make the ledger part of the writing itself: what you agree to in the project conversation gets promoted to records — canon, characters, story shape — and the agent reads those records before it drafts, surfacing contradictions rather than silently correcting them. The feature set is built around that separation between suggestion and commitment. Every draft is kept, grouped by session and pass, so a continuity sweep can be read side by side with what it replaced and undone a whole pass at a time if the ruling turns out wrong.

A chipped cup should stay chipped in the same place for four hundred pages. If you would rather not carry every such fact in your head, StoryProp keeps the ledger with you — decisions you confirm become records the agent consults before it writes a word.

← All posts