109. Process: draft, rest, cut

A sentence can be beautifully finished while the paragraph around it is still deciding what to say. This is an agreeable way to spend a morning and an expensive way to discover, after lunch, that the paragraph belongs on another page.

A working process helps you make the large decisions before investing too much in the small ones. Gather the material, find its shape, draft, return with some distance, and revise. The order can bend to the writer and the task. The result still needs to hold together when someone arrives without knowing what you intended.

Decide what the page must do

Begin with a reader, a question, and a useful change. After reading, what should this person understand, choose, or be able to do?

A concept page might help a new user distinguish a saved preset from an exported file. A procedure might help someone complete a supported export and recognize its result. A decision page might help a buyer assess whether a documented limitation matters to their work.

These are different finish lines. Write yours in a working note and keep its scope honest. If the available evidence covers only the format-selection dialog, do not promise a complete export procedure yet.

A useful definition of done names something you can inspect: the relationship the explanation makes clear, the choice the evidence supports, or the workflow that can be followed. “Sounds polished” is a preference without a stopping point.

The page may find a better purpose as you work. Change the working note when that happens. An explicit change of scope is easier to review than a draft that quietly becomes three documents.

Gather material before making promises

Keep the supplied facts, exact strings, references, and unresolved questions together. A small packet may be enough; a consequential workflow may need product observations, version information, and test results.

Include what constrains the writing as well as what enables it:

Material	Why it belongs in the packet
Reader and task	Determines which facts need explanation and which can be linked
Product behavior and conditions	Limits what the page can promise
Exact labels, commands, identifiers, and quotations	Protects text that must not be improved by paraphrase
Observed results and their scope	Separates a checked case from a general claim
Source and date or version where relevant	Makes later checking possible
Known gaps	Prevents an incomplete account from looking complete
Existing voice and requested edit depth	Helps an editor preserve what already works

Mark a gap as a gap. “Where are presets stored?” is a research question. “Presets are probably stored in the project” has already supplied an answer the reader may remember.

For a real account, collect the details that might make it vivid: a file, a choice, an unexpected result, a remark you have permission to quote. Do this before deciding what the story ought to prove. Otherwise the outline can become a polite request for evidence to agree with it.

Keep illustrative inventions separate. Fictional examples are useful teaching tools when identified as such; they do not establish how a real product behaves.

Choose a drafting route you can use

Some writers discover a subject by drafting through it. Others need to arrange the material before they can write a useful sentence. Both routes can work.

If you draft to discover, begin with the part you understand. Write a rough explanation, leave questions where the knowledge ends, and keep going far enough to see the larger shape. You can write the introduction after you know what it introduces.

If you need structure first, group the material by the reader's questions. Put prerequisites before dependent actions, evidence beside claims, and the substantial explanation where it becomes useful. Draft one coherent section at a time.

Neither route requires a ritual number of headings or a ban on correcting a sentence as you go. The practical danger is losing the whole page while tending a comfortable corner of it. When that happens, step back to the purpose and the missing material.

A rough draft may contain fragments, repeated terms, and bracketed notes. Roughness is permitted; unsupported certainty is not. An unfinished sentence can tell the truth about what remains unfinished.

Let the draft discover a voice

Once the core material is visible, notice where the writer's attention becomes particular. Perhaps a small object explains a large process, a careful qualification changes the conclusion, or a long chain of preparation ends in a very ordinary task.

These are opportunities for voice. Develop the useful ones before reducing everything to its shortest possible statement.

For an explanatory paragraph, try this movement: show the concrete situation, explain the relationship, then let the consequence arrive. A longer sentence can carry the reasoning. A short sentence can mark what has changed.

For a narrative, decide what the viewpoint can know and where scene deserves more space than summary. For a dry observation, write the plain fact first and then adjust the angle or timing. For an analogy, name the relationship it explains and the point where the comparison stops.

Do not add one of each device to every page. A task guide may need only a clear sequence and a patient explanation. A long essay has more room for a narrator who notices things. Let the material and the reader's situation decide how much room to use.

Return with some distance

A familiar draft invites you to read the sentence you meant rather than the sentence on the page. Time away can help. So can changing the presentation, reading a printout, or asking another person to explain what they understood.

Use the distance the schedule allows. There is no universal minimum rest period that certifies a draft as ready. An urgent correction may need immediate review by someone who has not been composing it. A longer article may benefit from returning the next day.

On return, read for the page's purpose before fixing commas. Can you still identify the reader's question? Does the argument advance? Where do you have to supply knowledge that is present only in your notes?

If you ask a reviewer for a paraphrase, record what they actually understood. Agreement with your intention is useful evidence from that reading, not proof that every reader will follow the same route.

Revise from claims to presentation

Separate passes make the work easier to inspect. They need not be rigidly isolated: a factual correction may force a new structure, and a sentence edit may reveal a missing fact. Return to the earlier concern when necessary.

Facts and provenance. Trace product behavior, conditions, quantities, causes, quotations, and attribution to their sources. Preserve the difference between what was observed and what was inferred. Run relevant procedures when you can; record the actual environment and result.

Structure. Check the reader's route. Move prerequisites and consequential conditions before the decisions they govern. Give sections distinct jobs. Keep reasoning connected where splitting it would conceal a relationship.

Style and development. Examine viewpoint, detail, rhythm, transitions, analogies, and the ending. Preserve a useful voice. Give an explanation enough room to develop, and cut a flourish that draws attention away from it.

Sentences and words. Clarify actors, verbs, references, and relationships. Remove repetition and empty emphasis. Combine choppy sentences when the connection belongs together; divide overloaded ones where the thought actually turns.

Mechanics and presentation. Check spelling, punctuation, exact strings, links, heading hierarchy, lists, tables, and the rendered surface. A source file can look orderly while its published table is difficult to read.

This sequence puts the expensive questions early. It is easier to improve a sentence once you know it is staying.

Check claims without quietly replacing them

A fact check compares the draft with evidence. It does not merely replace an implausible claim with a more modest-sounding guess.

“Most exports finish in seconds” cannot be repaired to “Small exports usually finish in seconds” unless there is support for that narrower claim. Adding small and usually changes the assertion; it does not supply a measurement.

Likewise, an authentication failure does not establish a wrong password, a missing file does not establish deletion, and a submitted job is not necessarily a completed job. Preserve the state the evidence actually identifies.

Check provenance separately from behavior. A feature can work as described while the account of who requested it, built it, or first used it is wrong. Accurate attribution matters to the story as well as to the people in it.

When a test covers one version or one input, report that scope. A successful run can support a bounded procedure check. It does not prove every possible environment, input, or recovery path.

Make the structure visible

Write a brief description of what each paragraph does: introduces a situation, defines a term, explains a cause, supplies evidence, gives an action, or qualifies a claim. This reverse outline often reveals repetition that the prose has made pleasant enough to overlook.

If three paragraphs each introduce the same idea, combine them or give the later ones a new job. If a claim has no evidence, return to the source packet. If a prerequisite appears after the step that needs it, move it.

Read the headings and link text separately. They should offer useful destinations. Then read the paragraphs continuously. The page needs both a route for scanning and a coherent account for someone who stays.

Do not force a miniature summary into the first sentence of every paragraph. A developed essay may begin a paragraph with an observation and arrive at the point through reasoning. The question is whether the movement helps the reader, not whether every paragraph wears the same uniform.

Edit for character as well as economy

A severe cut can leave all the nouns in place and remove the relationship that made them worth reading about. A lighter edit can preserve a vivid detail while removing the explanation that follows it unnecessarily.

Ask what each expressive choice contributes:

• Does the detail help the reader picture or understand the situation?
• Does the longer sentence carry a useful chain of thought?
• Does the short sentence land a real change of emphasis?
• Does the aside answer a likely question or reveal a relevant discrepancy?
• Does the analogy clarify a particular relationship?
• Does a repeated image gain meaning when it returns?

Keep the choices that work. The aim is a lucid, observant account whose writer seems to have noticed the subject, not a page assembled entirely from approved sentence shapes.

For marketing passages, chapter 310 develops cadence and narrative movement. For technical writing, the same attention can make an explanation more companionable while the actual steps remain literal and easy to follow.

Cut what the reader no longer needs

Remove duplicated points, throat-clearing, unsupported praise, and explanations that have outlived the problem they were meant to solve. Preserve conditions, uncertainty, attribution, and useful context.

A percentage can be a useful exercise or an explicit output constraint. It is not a general quality measure. The right revision may be shorter, the same length, or longer because it restores a missing relationship.

For this fictional warning, the source says that unsaved work may be lost:

Save your changes before exporting. Unsaved work may be lost.

Changing may be lost to will be lost would increase certainty. Changing it to will be omitted from the export would describe a different behavior. Neither is a stylistic shortcut.

Keep a private cut file if it makes removal easier. Label a passage's reason for leaving: duplicate explanation, wrong page, unsupported claim, or distracting aside. A later editor can then distinguish useful material awaiting another home from a claim that still needs evidence.

The separate TLDR contract has its own compression target and three-round procedure. Follow those requirements when using that skill; do not import a general preference for brevity as permission to change them.

Read for rhythm and reference

Read the near-final prose aloud when practical, or listen to an appropriate reading. Mark places where the intended relationship becomes hard to follow, where a pronoun has several possible owners, or where every sentence arrives with the same emphasis.

A stumble is a reason to inspect. It is not automatic proof that the sentence is wrong. Unfamiliar terminology, the reader's delivery, and the medium can contribute. Check the actual construction and preserve necessary technical terms.

Read the short sentences in sequence. If they sound like a series of instructions to admire the prose, combine or soften them. Read the longer sentences for a visible subject and direction. Length can carry thought; it should not make the reader hold several unresolved beginnings.

Check the last sentence of a paragraph. It may supply a consequence, an important limit, a restrained turn, or a transition. It does not need to repeat the paragraph in grander language.

Turn feedback into a specific repair

A reviewer saying “too long” has identified a reaction, not yet a cause. They may be encountering repetition, missing navigation, dense syntax, or information for a different audience.

Ask what the reader was trying to find or understand at the point of difficulty. Inspect that location. A proposed fix can be unhelpful while the reported confusion is real.

“Make this friendlier” may call for a clearer next action or a less accusatory description of the state. It does not authorize a promise that the file is safe. “Make this more vivid” may call for a supplied concrete detail. It does not authorize an invented customer anecdote.

When reviewers disagree, compare their tasks and prior knowledge. Sometimes the page needs a short main path with deeper explanation. Sometimes the draft has chosen the wrong audience. Resolve the reader problem before voting on sentences.

Work with a model from an explicit brief

A model can propose structures, draft from supplied material, compare revisions, or help shape a difficult paragraph. Give it the reader's task, source packet, protected text, register, edit depth, and required output.

Specify the stylistic operation as well as the desired mood. “Make this engaging” leaves many choices unresolved. “Develop the explanation through a concrete example, preserve the qualification, and vary sentence length where the thought turns” gives a reviewable direction.

If fictional examples are allowed, say so and require them to be identified. If the task is editing supplied text, preserve its facts, perspective, and effective voice within the requested edit depth. Do not invent experience to make the result feel personal.

Inspect the output against the packet. A tool-using agent may have checked a command or page; look at the evidence and scope before accepting that claim. A fluent assertion that verification occurred is insufficient.

Use the relevant editing prompt for its complete contract. The final artifact still needs the same factual, structural, and stylistic review as any other draft.

Diagnose prose without guessing its author

Repetitive phrases, vague praise, empty contrasts, and interchangeable examples are editable defects. They do not establish whether a person or a model wrote the passage.

A search for troublesome wording can help locate passages to inspect. It is not a verdict. A real comparison may need a contrast, a real process may contain three stages, and a technical term may be the correct word even if it appears on a general ban list.

Name the fault precisely: an unsupported outcome, an unclear actor, a missing condition, a needless repetition, or an example that explains nothing. Repair that fault. Do not add deliberate errors, arbitrary sentence variation, or fabricated personal details to pass for a particular kind of author.

The useful question is what the prose lets the reader understand or do.

Use the finished page and record the limit

A final check should match the page's purpose. Follow a procedure in its stated environment. Use a reference entry to answer a lookup. Check whether a concept explanation supports the distinction it promises. Inspect a decision page for the evidence and conditions needed to choose.

Use safe sample material for actions that alter files or state. Check relevant failure and recovery paths when they are within the task and available environment. Do not claim those paths passed because the successful path did.

For a prose revision, compare the result with its source facts and requested voice. Check the published presentation, links, and examples. A word count and a clean diff can support a record of what changed; they do not establish that readers understood it or enjoyed it.

Keep a short record of the surface checked, method, observed result, and remaining limits. For example: “Compared the revised paragraph with the supplied packet; all three conditions remain. Read the rendered table. Product workflow not tested.”

That note is modest enough to be useful later.

Practice: revise in two distinct passes

Use this fictional packet:

• Saving a preset stores the current export settings.
• Applying it later restores those settings.
• Applying it does not start an export.
• No time saving has been measured.

Rough draft:

Presets are a powerful way to streamline your workflow by saving settings and making future exports effortless, and they are very useful when you need to do this again.

First revise for the relationship:

A preset stores the current export settings so you can restore them later. Applying it prepares the settings; it does not start the export.

Then revise for an explanatory voice:

Save the current export settings as a preset, and you can restore that set of choices later. The settings return together. Applying them does not start the export.

The second version develops the distinction through a connected setup and two shorter observations. It adds no measured saving and does not claim that saving the preset performs the export.

Now write a neutral reference version and a more discursive teaching paragraph. Mark every capability, condition, and outcome in both. Remove any claim the packet cannot support. Then inspect whether the teaching paragraph contains a useful observation or merely more words.

Finally, remove the statement about whether applying the preset starts export. Both versions must stop asserting that behavior. Keep the unresolved question in the review notes. A sentence that loses its source loses its place, however well it reads.

Checklist

• The page has a stated reader, purpose, and inspectable finish line.
• Source facts, exact strings, and unknowns remain distinguishable.
• The draft develops the material without inventing certainty.
• Facts and structure are checked before fine polish.
• A separate style pass considers detail, viewpoint, rhythm, and relationships.
• Cuts preserve conditions, uncertainty, attribution, and useful voice.
• Feedback leads to a named repair rather than a vague change of mood.
• Model output is reviewed against the same contract as other prose.
• The finished surface has been checked for its intended use.
• The verification record states what passed and what remains untested.

When those checks reveal no remaining task, stop. The sentence you improve next may be the one that makes tomorrow's reviewer ask what changed.