Introduction
Last time I wrote about how 540 articles had accumulated over fifteen years, and why. This time I want to write about how they are being cut down. What became clear in doing it is that this calls for a different capacity from writing.
Editing Turned Out to Mean Taking Away
What I am doing is not writing. It is editing. And editing, it turns out, is not adding but taking away. That has proved harder than I expected.
To take anything away, one needs a standard for what stays. Without a standard, every piece is something I wrote, and every one of them has its reasons. Read it back and the day it was written comes with it. On those terms nothing can go.
The standards came gradually, once the work had started. There are three at present.
- The first: anything that could put a reader at risk. Writing about health or money in a way that outruns the evidence behind it.
- The second: anything made only of summaries and links to other pieces. Something put there to gather things together, whose job has since been taken over by a page built later.
- The third: anything that does not say what matters. Announcements, news of the month, notices about fees. They meant something when they were written; read now, nothing remains.
Once the standards were there, things could go. Finding them was the hard part. And once they were found, the work moved considerably faster.
Keeping a Record of the Work
Reading 540 articles one at a time does not happen in a day. It goes on over many days, a little at a time. And after a while one cannot hold in mind what was decided about what.
So this time the sorting is being done with a generative AI, Claude, keeping the record in files as I go — Markdown files, which I will call MD files from here. The work consumes a great deal of what is called context, the working memory a model has available, which is why it takes Claude Max to run it.
There are five MD files at present. A reference file, for things settled. A worklog, for why they were settled that way. An index of titles and addresses in both languages. A backup holding the full text of every article deleted. And a plans file, where ideas for pieces not yet written are put aside. Between them they run past eight hundred thousand characters. There is now more record than there are articles.
The reference file and the worklog are kept apart for a reason. Wanting to know what was settled, and wanting to trace why, send you looking for different things. The spelling of a name or a date belongs to the first; the argument over whether to delete something belongs to the second.
Every deleted article is kept in full in the backup. This matters more than it sounds. If deleting means losing, the decision gets heavy. Knowing the full text is held somewhere makes it possible to lean toward cutting. Several times now I have deleted a piece and carried the material out of it into another article, or into the plans file.
As I said, the work is done with Claude. Every session begins by loading all the MD files. That way I pick up where I left off, and I get told about things I have forgotten. Checking whether two articles overlap, or whether something here is already carried by another piece, is not something memory can do across 540 of them.
Being able to ask Claude about WordPress matters as much. How to set up a redirect after deleting a piece, how to check whether something is indexed, how to search the full text to find what still links to it. These are things one could look up, but looking them up takes longer than the work itself. Asking on the spot keeps the work moving.
For anyone publishing on their own, I think this part transfers directly. It amounts to no more than keeping a record while you go. But with fifteen years in front of you, working from memory rather than from a record means making the same decisions over and over.
The Same Thing I Do in a Session
Partway through, it began to seem that this was the same work I do in a session.
In a Rolfing session nothing is added. There is no building up of muscle to improve posture, no teaching of a correct shape. What happens is that the conditions are set for surplus effort to come out. When it does, what was carrying the body remains. Once that begins to work, the body settles itself. It is not settled by me.
There, too, a standard is needed. What counts as settled? If it is whether the pain has gone, or whether the person feels lighter, it shifts with the day they are having. So the relation to gravity is what gets looked at — how the body is meeting gravity as it stands. Hold the standard to one thing, and what to take away becomes clear.
Editing worked the same way. Only once the standards had settled into three could anything be cut.
There is a second resemblance.
In a session, a place that has gone hard is not treated as the bad part. No muscle tightens without reason. Something is being carried by that tightness. Release it without seeing what it was carrying, and somewhere else goes hard instead.
Articles are the same. Finding two that overlap does not mean one of them can simply go. Often the overlap is partial, and the other half exists nowhere else. It does not come down to all or nothing.
Which is why the third sorting is needed — several pieces edited together into one. Taking away does not only mean deleting.
Conclusion
This piece has been about the cutting: that nothing can go without a standard, that the work is done alongside a written record, and that it turned out to have the same shape as what I do in a session.
Next time I want to write about what surprised me in the reading. I had misremembered something I wrote myself.