Part one argued that the comparison grid cannot score the thing that matters, which is why it keeps producing a different winner. This part is about what the restarts cost, because it is the part nobody prices. We are used to thinking of a tool search as overhead before the work. For a long book it is not before the work. It happens several times during it, and it takes something from the manuscript every time.
A note on scope, so this does not overreach. I am not pricing the scatter itself here; that bill is itemised in an earlier piece, and it is a different argument. This is narrower: the cost of the searching, the trialling and the moving, charged to a writer who has not settled anywhere yet.
Every migration leaves a deposit behind
The story a migration tells about itself is that the book moves from A to B. That is almost never what happens.
What moves is the manuscript, because the manuscript is the part you can see. What stays behind is everything you did not think to look for at the time: the comments in the margins that did not survive the export, the folder of clippings, the outline you built in the first flush of enthusiasm and stopped updating, three drafts of a chapter you might still want. You do not abandon these deliberately. You move the book, the new app looks clean and correct, and the old account keeps existing, quietly holding four things that turn out to matter in month nine.
So the count only goes up. A writer who has switched three times does not have their book in one place, and does not have it in three either. They have it in one place plus three sediment layers, each of which still has to be searched on the day it is needed.
A tool search presents itself as the cure for scatter. Run three times, it is the largest single producer of it, because every incomplete migration adds a place your book partially lives and removes none.
The trial month is charged to the book
The second cost is easier to see once named. Evaluating a writing app properly means using it for real work, which means the trial is not run on spare time. It is run on writing time, and it is charged to the book.
The evaluation weeks are the ones where you are learning a new arrangement, so your attention is on the app rather than the chapter. You are not writing badly. You are writing while also assessing, and assessing is not free. Two or three weeks of divided attention, three or four times across a project, is a meaningful fraction of the total time you will ever spend on this book.
And the trial produces a systematically misleading result, because it is run on the wrong material. A trial always begins in the easy part of the book. You import a chapter you already like, you write a scene you were looking forward to, and everything feels wonderful. What you needed to know was how the app behaves in chapter nineteen when you have to check something you decided in chapter three, and the trial never reaches chapter nineteen. The verdict is real, it just answers a question you were not asking.
The deciding is the expensive part, not the moving
The third cost is the one I underestimated for years, and it is not the migration. It is holding the question open.
An unsettled decision runs in the background. "Should I be doing this somewhere else" is available to you at every friction point, and a long book has a friction point most days. That question is a permanently open exit from the hard part of writing, and it is a respectable one, because it looks like diligence rather than avoidance. Nobody feels guilty for researching their setup.
The tell is timing. If you look back at when your searches started, they tend to cluster at the difficult junctures of the book: the middle, the return after a long break, the chapter you cannot get right. Those are exactly the moments a solvable-looking problem is most welcome, and choosing software is beautifully solvable. It has comparison tables and clear criteria and an end state. The chapter has none of those things.
- Your last three comparisons were prompted by a hard week rather than by a defect you could name.
- You can describe the feature sets of four apps and have not opened the manuscript in nine days.
- You have an account, and some of your book, in a tool you have decided against.
- The phrase "once I've sorted out my setup" appears in your own planning.
- You have migrated to a tool you had previously migrated away from.
What the restarts do to the writing itself
The three costs above are about time and attention. There is a fourth that lands on the page, and it is quieter.
Each move resets your sense of the book's size. A manuscript in a familiar setting has a shape you know by feel: you know roughly where you are, what is thin, which sections you have been avoiding. Rebuild it somewhere new and that felt sense goes. The book becomes, briefly, a set of files again. Most writers respond by re-reading to reorient, which is the most expensive way to spend a session and the one least likely to add words.
Do that often enough and the project acquires a particular texture: heavily worked opening chapters, thinning as you go, because the early material has been re-read at every migration and the later material has not been re-read at all. That is not a planning failure or a stamina failure. It is a record of how many times the book was moved.
The search is not free time spent before the writing. It is writing time, book-shaped attention, and nerve, spent several times over, on a question that a single well-chosen answer would close.
The honest counter-case
All of that argues for settling. It does not argue for settling on the first thing, and I want to be careful not to smuggle that in.
Some searches are correct and cheap. If you have never chosen deliberately, and your book lives wherever it landed, then one real evaluation is not a restart. It is the first decision, and it is overdue. The costs here are the costs of repeated searching, and the first one is not repeated.
Equally, staying somewhere that genuinely cannot hold your book is not thrift. If there is no way to keep versions, or no way to get the text out again, the cost of leaving is real but bounded, and the cost of staying is not. What I am pricing is the search that recurs without a defect behind it.
And a caution on the diagnosis itself. "You are avoiding the chapter" is a comfortable thing for an article to say and an uncomfortable thing to be told wrongly. Sometimes the setup really is the problem. The distinction I use is whether I can name the defect in one sentence without using the word "properly". "I cannot keep versions" is a defect. "I want something that handles research properly" usually is not, because it has no failure condition, which means no app will ever satisfy it.
What part three has to do
If the search recurs, the useful goal is not finding a better app. It is making the question closeable, so that it stops being available on hard days.
That needs two things a comparison grid does not give you: a test you can run in one session rather than a trial month, so evaluation stops costing weeks, and criteria stable enough that the answer does not change the next time you are stuck. That is what part three is.
