There is a browser tab I keep reopening. It is a comparison of book writing software, a grid of apps down one side and features across the top, with little check marks in the boxes. I have opened some version of that tab in three different years. Each time I read it carefully, pick a winner, move my draft, and feel the specific relief of a fresh start. And each time, some months later, I open the tab again.
If that is familiar, I want to be careful about what I am claiming. Choosing where to keep a book you will spend two years on is a real decision, and taking it seriously is not avoidance. The odd part is narrower: the comparison keeps coming out differently. Same apps, same writer, different winner. That is not what a well-posed question does, and it is the thread worth pulling.
The grid is unstable, and that is a fact about the grid
Run the comparison in a month when your current app lost some work, and it scores reliability first. Run it after a fortnight of fighting an outline, and it scores structure first. Run it on a laptop on a train, and portability jumps three places. The apps did not change between those runs. Your criteria did, and your criteria were set by whatever your last bad afternoon happened to be.
That instability is diagnostic. When a question has a stable answer, repeated honest attempts converge on it. When they do not converge, either the question is underdetermined or the thing being measured is not the thing that matters. Here it is the second one.
Features are close substitutes. Almost every serious writing app can hold chapters, count words, and get out of your way. Ranking them on that is like ranking pens: the differences are real, they are small, and preference is doing most of the work. A ranking built on close substitutes will swing with mood, because there is not enough signal in it to hold still.
The column that does not exist
So what would hold still? In my experience one thing predicts whether a long book gets finished, and it is not a feature. It is whether the whole book can live in one place: the notes, the research, the chapters, the versions, the people and places, the decision made in chapter three that has to still be true in chapter nineteen.
That property has no column, and it cannot have one. A feature belongs to an app, and you can put a check mark beside it. Scope is a relationship between an app and everything else you keep open, and no cell holds a relationship. It is also not fixed when you sign up. It is set afterwards, by a hundred small and individually sensible decisions, which is how an app that could have held your entire book ends up holding a third of it.
I am not going to re-argue how that unfolds, because it is written up already: the three kinds of scatter, and the inventory that shows you yours, are in Where Your Book Manuscript Actually Lives. What matters here is what it does to the comparison. The table is silent on scope, and to anyone scoring rows, silence reads as unimportant.
A comparison grid scores properties of an app. Whether your whole book can live in one place is a relationship, not a property, so it never appears as a row — and the thing you never score is the thing that decides whether you finish.
Two stories we tell about the restarts
When the search comes round again, the explanations usually offered are that the right app has not been found yet, or that the writer is simply disorganised. Neither survives much contact with the evidence. The first cannot ever be settled, since there is always another app with a feature you did not know to want until a table showed it to you. The second is contradicted by the drafts folder, the naming convention, and the running list of things to fix in chapter nine, which are not the artefacts of a disorganised person.
Both stories share a flaw worth naming: they take the grid's terms for granted and argue about who is failing them. The more useful move is to doubt the terms.
- Your shortlist has changed more than once while your book stayed the same.
- You can name the feature that won it for each app you have tried, and none of those features is why you left.
- You have never asked a candidate app where your research would go.
- You could not say, without checking, which app currently holds the newest version of chapter one.
- The last app you left is still where some part of your book is.
That last one is the one I would sit with, and it is where part two starts.
A tool you like is not the same as a place your book can live. The first is a preference. The second is a structural property, and only one of them survives a long project.
Where I might be wrong
Two limits, because this argument overreaches if you let it.
Some writers work across several apps for an entire career and finish everything they start. They tend to share one habit: a single canonical location they never violate, with everything else explicitly downstream of it. That is the same principle arriving by another route, so if you have it, none of this describes you, whatever your app count says.
And the instability I have made so much of is not always a symptom. A first comparison, run by someone who has never chosen deliberately, is not a restart and has nothing to converge with yet. What I am describing is a search that recurs, with a different answer each time, and no defect behind it that you could state in a sentence.
What follows from this
If this fits, the good news is that it is one decision rather than a character flaw. You have been answering a question that was put to you in the wrong terms, carefully, several times.
The part that should give you pause is the last diagnostic above: the app you already left is still holding some of your book. That means the searching is not a neutral prelude to the writing. Each pass takes something, and part two is the bill for the searching itself, including the item most writers never attribute to it at all. Part three is then the method: one test you can run in a single session instead of a trial month, four questions worth scoring, and how to settle the question so that it stops reopening on hard days.
