There's a specific kind of dread that shows up around the midpoint of a novel: you're about to write a scene that depends on something from twelve chapters back, and you're not sure anymore. Did the letter arrive before or after the funeral? Is the family estate three days' ride from the capital, or five? You wrote it. You know you wrote it. You just can't find it without derailing an hour of writing time to go looking.
Most writers solve this with a running notes document that starts diligent and quietly falls apart by chapter ten, because keeping it updated is its own separate job competing with the job of actually writing.
What a chapter summary is actually for
A summary of what you've already written isn't a writing aid in the usual sense — it doesn't touch your prose at all. It's closer to a status report: here is what has happened in this manuscript so far, stated plainly, without spoiling anything beyond where you currently are.
That's useful in two specific moments. The first is picking a project back up after a few days or weeks away, when you need to re-load the story into your head before you can write the next scene with any confidence. The second is exactly the dread scenario above — double-checking continuity before you commit a new detail to the page, instead of after a reader points out you got it wrong.
Where fact extraction does the part a summary can't
A prose summary is good at plot and pacing, but it's not built to catch a specific, narrow kind of error: a detail stated once, then quietly contradicted later. A character's eye color. A sibling's age relative to the protagonist. The year a war ended, mentioned in passing in chapter two and then mentioned differently in chapter twenty-one.
This is where fact extraction is doing a different job than a summary. Instead of narrating your plot, it quietly builds a running list of the concrete facts your manuscript has established — names, ages, dates, physical details, relationships — and flags the moment a later chapter says something that doesn't match. It's not reading for meaning, the way a summary is. It's reading for consistency, which is a more mechanical task and one that's genuinely tedious for a human to do by hand across a full manuscript.
The distinction matters because it's the difference between a beta reader catching an error on page 200 and you catching it while you're still writing page 45.
Why this only works if it's read-only
Both of these features share a constraint worth naming directly: neither one ever generates a sentence that ends up in your manuscript. A summary reports what's already there. A fact flag says "this doesn't match something you wrote earlier" — it doesn't tell you which version is correct or rewrite the sentence for you. The decision about what actually happened in the story stays with the person who's writing it.
That's a deliberate line, not a missing feature. A tool that quietly resolved the contradiction for you would be making a plot decision — which era the war ended in, how old the brother actually is — that isn't its decision to make. Flagging it and getting out of the way is the more useful, and more honest, version of that help.
In Bramblequill
Every chapter gets an AI-generated summary of your manuscript so far, refreshable once you've written more, alongside fact extraction that runs quietly in the background and surfaces a conflict the moment one shows up — right in the editor, not in a separate report you have to go check. Both are read-only by design: they tell you what's on the page and what doesn't match, and you decide what to do about it. The letter's timing, the estate's distance from the capital — that's still entirely your call. You just don't have to remember it unassisted.