A draft that is not landing has two possible faults. The sentences are wrong, or the plan the sentences are serving is wrong. Almost nobody checks the second one. They reach for the sentences, because the sentences are the part you can see.
That reach is the whole problem. The long version of the argument, with the two failures that taught it to me, is in the companion essay. This is the short version. Five things revision is for, one thing it is not, and a test you can run against your own draft for each of them. Read it with a draft open.
Contents
- 1. Rewriting the spec, not editing the output#
- 2. Naming the spec failure, the broken section#
- 3. Restructuring into modules, cheap revision#
- 4. Letting the model audit, delegating the check#
- 5. Knowing when you are editing instead, polish is not craft#
- The one thing revision is not, better is a byproduct#
1. Rewriting the spec, not editing the output#
Every draft is an answer to a question you asked yourself before you started. Sometimes you wrote the question down. Usually you did not. Either way it is in there, and every sentence you have written is obeying it.
Revision goes back and changes the question. Editing leaves the question alone and sands the answer.
The test. Say out loud what this piece is supposed to do for the person reading it. One sentence. If you cannot say it, you have no question to revise, only an answer to sand. Write the sentence first. Then read the draft against it.
Almost nobody runs this test, which is how a person spends a career sanding answers to questions they never checked.
2. Naming the spec failure, the broken section#
A failing draft rarely fails everywhere. It fails in one place and bleeds outward, and the bleeding is what you notice. So the whole thing feels rotten and you start over, when four fifths of it was fine.
The signs are legible if you look. Section titles that could sit on top of any essay. A close that summarizes instead of deciding. A paragraph two readers read two different ways. None of that is a sentence problem in a sentence disguise. That is the plan showing through the paint.
The test. Point at one section and say what it is doing wrong, in words, hands off the keyboard. If you can point, the repair is now a chore. If you cannot point, every repair you attempt will be guesswork, because it is.
3. Restructuring into modules, cheap revision#
A section trying to do five things does none of them, and it cannot be repaired in place, because there is nothing there to repair. There is only a decision nobody made.
Build in parts instead. One part, one job. Parts can be pulled, swapped, thrown out, or set down in a different piece for a different reader without dragging the rest along behind them. A house built in rooms can lose a room. A house poured as one slab cannot lose anything at all.
The test. Take a paragraph and try to move it three paragraphs down. If the piece survives, you built in rooms. If everything around it caves in, you poured a slab, and every change from here costs you the whole thing.
4. Letting the model audit, delegating the check#
You cannot see your own draft on the fourth read. You see what you meant. The machine has no such handicap. It has never met your intentions, and it does not get bored.
So hand it the standard in writing, and make it check each line against the standard rather than against its taste. Not “is this good.” Good is a matter of opinion, and the machine will just agree with you. Ask whether line eleven does the job you said line eleven was for. That question has an answer.
The test. Could you hand your standard to something that has never read your draft, and would it know what to reject? A standard that only makes sense inside your own head is not a standard. It is a mood.
5. Knowing when you are editing instead, polish is not craft#
There is a specific feeling worth learning to catch. You are in the third section again. The sentences there were already fine. You are making them slightly finer. You feel productive, and you are not.
That loop is what a writer does instead of facing the part that is actually broken. It is comfortable because it cannot fail. Nobody ever ruined an afternoon by improving a sentence that did not need improving.
The test. Ask what changes for the reader if this paragraph comes out perfect. If the honest answer is nothing much, close the file and go reread your one sentence from the first test. The work you are avoiding is upstream.
The one thing revision is not, better is a byproduct#
Revision is not for making the draft better. Better is a byproduct. Revision is for getting the spec right. When the spec is right, the draft is better. When the spec is wrong, no draft will be.
Which is why “make it better” is useless advice, and why it is the advice nearly everyone gives. It names the byproduct and skips the cause.
The choice sits one step earlier than the sentence, in the same place everything else that matters sits. Decide, then write. No sentence rescues a decision you never made.
The painter polishes. The architect revises. Most revision changes which wall the painter is asked to paint. Most writers do not know it is a choice.
···