Long-time Scrivener user, first-time poster — and a continuity problem I finally solved

Latte Other Software & Development
1

Hi all, long-time Scrivener user, first-time poster here.

I write epic science fiction — a multi-planet story spanning millions of years of in-universe history — and for years, my biggest editing headache wasn’t prose, it was continuity.

The project started in Microsoft Word with OneNote for reference material. When I discovered Scrivener, I migrated everything over and did a round of revisions, got a few people to look over the writing, though not everyone had the bandwidth to get through it all. Around that same time, I started tracking the timeline in Excel, trying to make sure events lined up across multiple planets.

Eventually, I bought Aeon Timeline and reconstructed the whole thing there — and that’s when the real rabbit hole started. Making sure events lined up across multiple planets, accounting for repeating cycles that involved more math than I’d anticipated, I was up late more than a few nights on that alone. Then I got a little carried away with the worldbuilding more broadly, and several years passed. When I finally came back to the manuscript itself, the story in the timeline didn’t quite line up with what I’d actually written anymore. The bible had grown in one direction while the draft sat waiting in another.

I tried the usual fixes to close the gap — re-reading with a checklist, color-coded comments in Scrivener, and more readers. Nothing scaled. So I used Claude AI (and later Claude Code) to handle the reconciliation work itself — generating scene synopses from the Scrivener documents, importing them into Aeon Timeline as events, and pulling the relevant details from each scene so the timeline could finally sync with the manuscript. I didn’t let it touch the actual prose. Around that same time, I moved all my reference material out of OneNote and into NotebookLM notebooks, one per planet.

That process worked, but it also made the problem obvious: the real issue wasn’t that I hadn’t tried hard enough. It was that there was no tool that compared the manuscript against the bible automatically.

So I built one. The idea is simple: point it at your Scrivener project and your lore folder (including an Aeon Timeline CSV export), and an LLM compares any scene you select against your notes for character consistency, timeline conflicts, world-rule breaks, and plot logic gaps. It’s not a replacement for a human editor — it’s a sanity pass for the kind of small contradictions that are hard to spot when you’re 90,000 words deep in your own head.

I called it StoryCheck. The Scrivener integration was the first thing I built — you open a .scrivx directly and click any document in the binder tree to load it for a check. Read-only; it never touches the project.

I’m posting here partly because L&L users are exactly the people I built this for, and partly because I’d genuinely like to hear how the rest of you handle this. Do you maintain a separate bible, or do you re-read with a checklist? I’m curious whether this is a widespread pain point or whether I’m just uniquely disorganized after twenty-plus years building the same universe.

(If anyone wants to poke at the tool itself, I can drop a link in a follow-up — I don’t want this first post to read like an ad. Happy to answer questions in the thread.)

— Silvercat

3 Likes
2

I’m not going to comment on the LLM side of things, as I’m not letting a plagiarism-powered ecosystem destroyer anywhere near anything of mine.

On the general problem, though, I think the first question is, what is “truth” in your world? Is it the manuscript? Is it the worldbuilding bible you’ve so laboriously constructed? Or does it even matter?

While you, as the creator of the world, are all-seeing and all-knowing, your characters are probably not. They can have incomplete or incorrect information. They can lie or exaggerate. They can misinterpret what they see.

Is it a plot hole? A continuity error? Or an unreliable source being unreliable?

7 Likes
3

That’s the sharpest version of the question for this kind of work, and honestly, the one I keep running into.

In my own draft, truth lives in the bible — the manuscript is just voices telling pieces of it, sometimes wrongly. I’ve got a POV character working from cultural propaganda who deliberately misinterprets a major historical fact. That’s not a continuity error; it’s the whole point of the chapter. A few other characters across the planets carry incompatible versions of the same events for the same kind of reason.

Which is exactly the problem with this kind of automation. The easy thing is to flag “manuscript says X, bible says Y.” The useful thing is “is that a slip, or is the character supposed to be wrong?” Right now, I just go through the flags one by one and trust myself to know the difference. You’re making me think the output itself should read “this could be a continuity error or an unreliable source — your call” rather than just flagging it as a contradiction—something to think about.

How do you handle this in your own work — do you track per-character knowledge somewhere, or keep all of it in your head?

4

My scope is much smaller than yours, which does make it easier. OTOH, since mine is a pre-modern setting, I need to allow for travel time/non-trivial communication time. For any given event, I need to know both whether it has happened and whether the characters in the current scene know about it yet.

I mostly depend on a timeline and a map to keep track of how people and information move around the world. Those have – as far as I know – been sufficient so far.

2 Likes
5

Personally for me, I find that if I try to create a story bible first, and then fit the story into it, I will definitely go down a rabbit hole that makes me lose all momentum. For me, as I write the story, I’ll note it in the Story Bible in the relevant section. And I’ll note any contradictory views on things in it. I’ll have the story bible broken down by regions/cities/colonies, then I’ll tag that location or event (or item if it’s important enough) as a Keyword in the scrivener file. That way if I do want to go back and double check for actual plot holes/continuity errors, I know which chapters to look at.

I think it also helps to review the actual writing for the piece of world building to see the context it’s being mentioned. Would the LLM be able to distinguish Character A lying to Character B about an event or fact, and Character B believing it and spreading it around?

Aside from the fact that I don’t like LLMs, I’m not really looking to have yet another program to have up while writing or editing a story. I work on one (rarely 2) screens, there’s only so much real estate for programs and switching back and forth is inefficient and also slowls momentum.

3 Likes
6

@Lizbeth — that’s honestly how I should have approached it. I built the bible separately for years and ended up with two things that didn’t quite agree with each other. Tagging by region with Keywords as you go sounds a lot saner than what I did. I might actually steal that idea for narrowing down which parts of the lore get checked against a given scene — that’s something I haven’t figured out yet.

On the lying vs. ignorant question — the tool can’t tell on its own, and I didn’t try to make it. What it can do is work from whatever I’ve put in the bible. What I’m moving toward is output that reads something like “diverges from canon, but may be consistent with this character’s stated POV,” when I’ve noted that the character is working from propaganda. If I haven’t noted anything, the tool can’t guess. The bible still has to carry the judgment — the tool’s job is to spot the inconsistency I’m too close to the manuscript to see myself.

On the screen space — it’s not meant to stay open while you write. You run it on a scene, get a plain text report back, and close it. You can paste the output into a Scrivener note if you want to keep it. I built it that way partly because I didn’t want something sitting in the middle of my writing process either.

The LLM unease I share. The model itself can be touchy about fictional violence, even when it’s clearly just part of the story. That’s a real headache for SF or fantasy writers, and it’s a limitation I haven’t found a perfect way around yet.

One thing I’m curious about — does the Keyword approach handle the per-character-knowledge problem well enough for you, or do you still find yourself keeping a lot of that in your head?

— Silvercat

7

Generally, I have an idea but I also use the metadata field for which character is in which scene. The way I’ve split what is in metadata and what’s in the keyword field is based on how necessary it is for scene continuity from one scene to the next (since I write one scene per document in Scrivener). So for me, Characters (POV and others in the scene), Location (Region and Specific Setting, like a bar), and Date. I leave keywords for things that are mentioned in the scene and I need to keep track of (Items, Events, etc).

Because I do close 3rd, the POV metadata helps me when I’m editing the prose for factual errors based on the keyword. For Characters involved in the scene that are not the POV one, I just focus on dialogue or action.

1 Like
8

That’s a cleaner line than I’d drawn. I’d been treating Keywords as one bucket. Your split — metadata for what has to carry into the next scene, keywords for things mentioned in the scene that still need tracking — is the part I was missing. “Does it need to survive into the next scene” is a good test for deciding which is which.

The close-third thing lines up with how I check my own draft, too. I lean on the POV tag — if I know whose head a scene is in, I can hold the prose against what that character knows instead of against everything that’s true in the universe. A character can be wrong on the page without it being a mistake, and the POV tag is what tells me which is which.

The part I hadn’t thought about is tagging the other characters in the scene, not just the POV one. If I had that, I could pull the notes for whoever’s actually present instead of the whole cast every time. Right now I read more than I need to. I might steal that.

The catch is keeping that field current across a few hundred scenes, and I’m not always disciplined about it 90,000 words deep. But that’s on me, not the method.

Do you fill the Characters field in as you draft, or add it in a later pass? That’s the part I always lose.

— Silvercat

9

Kind of both, really. My first draft (the one that no one should ever see) is handwritten and I’ll make a note of the characters as I write. When I put it in Scrivener, I’ll update the metadata after I finish the scene. Because the scenes generally aren’t too long, it’s easy to remember.

I do have to go over it in a second or third pass for super minor characters though. Ones who get named just to hand off something, because a bigger character says people’s names, or something like that. It’s not that I’ll add them to a story bible or character list, but more to make sure I’m not reusing names in throwaway characters, or that names match the style they need to

I will say that once you start keeping track of the characters involved, it becomes easier to do and I find myself writing them down less and less (before putting it on Scrivener).

The setting, though, I do usually do during a different pass. It doesn’t change as much as the characters do, unless it’s like moving to a pub from a park or something like that, but the city and region tends to stay the same for multiple chapters in a row, not just a few scenes.

I have a friend who uses one of the status options with the view turned on in the binder (so it shows as a colored dot next to the scene in the binder) to indicate the POV. It’s not my visual preference, but it’s another idea on how to keep track of things to get an idea for cross reference.

2 Likes