# Core method Use this guide when a draft is complex, compressed, or unclear enough that surface editing would not solve the problem. ## 1. Write the hidden contract Before rewriting, make a private note: | Question | Useful answer | |---|---| | Who is speaking? | The person or organization whose facts and judgments must be preserved | | Who is reading? | Their role, current knowledge, likely concern, and domain gap | | Why are they reading? | To learn, decide, act, trust, respond, or enjoy | | What is the one message? | The answer the draft must successfully deliver | | What must happen next? | The action, decision, understanding, or emotional result | | What cannot change? | Facts, stance, promises, terminology, uncertainty, privacy, and format constraints | A human draft fails when any one of these is generic: no identifiable speaker, no real reader, no purpose, or no concrete subject. ## 2. Rebuild before polishing Use this order: 1. **Extract** the facts, judgments, unknowns, evidence, constraints, and requested action. 2. **Write the answer** in one plain sentence. 3. **Order the support** by what the reader needs, not by the order in which the writer discovered it. 4. **Expand compressed reasoning** where the reader would otherwise have to guess. 5. **Cut material with no communication function.** 6. **Restore the right voice and form.** Do not begin by swapping words. A smoother sentence can still carry the wrong message. ## 3. Use concrete editing operations ### Keep constraints backstage unless the reader needs them Facts, uncertainty, safety, privacy, and scope still govern the rewrite. They do not all need to become prose. Weak: > I paid 140 baht on this trip. This is only one personal experience, does not represent every operator, and should not be treated as a long-term guarantee. Please verify all details yourself. Better: > I paid 140 baht that day. Check the current fare with the station before you leave. The second version preserves the changing-fare boundary where it matters. It does not make the reader sit through a legal-style waiver. If the claim is already scoped by words such as “I saw,” “that day,” “in this test,” or “I suspect,” do not repeat the same boundary at the end. ### Stop when the purpose is complete Task-oriented writing often needs an action or decision. A story, joke, observation, or personal post may not. Do not add a moral because the draft feels short. Do not turn every experience into advice, every joke into an explanation, or every open question into a neat answer. Leave a real loose edge when that is what the speaker has. ### Put the outcome before the process Weak: > After conducting a comprehensive review of the current implementation, several areas requiring further attention were identified. Better: > The current build works for the tested case, but two installation paths still fail. The second sentence tells the reader the result. The review process can follow only if it matters. ### Replace abstract stacks with actors and actions Weak: > The completion of the synchronization consistency verification has been achieved. Better: > We compared all three copies. Their file hashes match. Ask: who did what to which object, and what was observed? ### Translate labels into consequences An internal label is useful to the team that created it and opaque to everyone else. Convert it through: `label → observation → meaning → reader impact → action` Weak: > The release is PASS_WITH_FALLBACKS. Better: > The exported file passed the content and layout checks. Setup still required two manual workarounds, so the deliverable is usable, but installation is not yet one-click. ### Keep the term; add the bridge Do not delete a precise term merely because the reader may not know it. Weak simplification: > The systems check whether the files are the same. Better: > The systems compare SHA-256 hashes, which act like fingerprints for the files. Matching hashes show that the copies are byte-for-byte identical. ### Explain to the needed depth For a technical idea, choose the parts the reader needs: - what it is; - why it happens; - how to use or verify it; - prerequisites; - what failure looks like; - where the conclusion stops applying. Do not add all six mechanically. Add the missing ones that complete the reader's task. ### Keep valid human functions Not every sentence exists to transfer facts. A sentence may establish trust, acknowledge concern, disagree, reassure, persuade, or deliver humor. Delete a comment only when it does not serve the purpose or falsely attributes a view to the speaker. Do not strip a real point of view merely because it is not a fact. ### Preserve useful form Lists, headings, commands, dates, formal qualifiers, and technical terms can make a text more human because they reduce decoding effort. Do not turn a release note, contract, or resume into casual chat. ## 4. Worked examples ### Status update Before: > The synchronization has been completed, the hashes of the three copies are consistent, and runtime discovery is pending restart. After: > The same skill is now in the project repository, the team library, and the local runtime. All three copies contain the same files with matching hashes, so no version drift was introduced. The runtime will recognize the new version after it restarts. Why it works: it names the objects, explains what matching hashes prove, and tells the reader what remains. ### 中文进度汇报 改前: > 三层同步已完成,12/12 哈希一致,运行态待重启发现。 改后: > 同一份技能已经同步到项目仓、团队技能库和本机运行目录。三处共 12 个文件完全一致,说明没有漏文件或版本漂移;重新打开宿主后,它才会识别新版本。 为什么更清楚:不是只把句子写长,而是解释“三层”“哈希一致”和“待重启”分别代表什么。 ### Evidence boundary Before: > Enterprise-scale benefits are not currently advocated. After: > The workflow passed the current tests and can be described as verified in those scenarios. It has not yet produced organization-wide usage data, so do not present it as a company-wide efficiency gain. Why it works: it does not make the writer sound opposed to business value. It states what is supported and where the evidence ends. ### Recommendation Before: > In view of the associated risks, it is recommended that advancement be temporarily suspended. After: > Do not launch it yet. Anonymous users still receive a 404. Launch after the public page opens and one remote installation succeeds. Why it works: recommendation first, observable reason second, release condition last. ### Cross-domain explanation Before: > The process is idempotent. After: > The process is idempotent: running it again with the same input should leave the system in the same state instead of creating duplicates. That means a retry is safe after a timeout, provided the input has not changed. Why it works: it keeps the term, explains the behavior, and states the practical condition. ### 关系型回复 改前: > 非常感谢你的宝贵建议,我们会持续关注并优化。 改后: > 你指出的问题成立:说明页写了“支持导出”,但当前版本导出的文件仍会丢失表格格式。我会先把这个限制补进文档;修复前,不再把它写成完整导出。 为什么更像人:它回应了对方真正指出的事情,说明影响和下一步,也承担了明确立场。关系语气没有被删掉,只是从万能客套换成了真实回应。 ### Legitimate KEEP Original: > The migration changes the file format, not the stored content. Back up your customized files, run the converter, and review the diff before replacing the old directory. Decision: `KEEP`. Why: the reader can identify the change, the required action, and the safety check. Rewriting would add motion, not value. ## 5. Final tests - **Reader test**: Can a smart person outside the field explain the point back accurately? - **Action test**: If the text expects action or a decision, is it visible? - **Replaceability test**: Could any person or company say every sentence unchanged? If yes, add the real object, evidence, or point of view. - **Depth test**: Did the explanation answer the real question, or merely restate common sense? - **Preservation test**: Did the rewrite change facts, confidence, stance, obligation, or scope? - **Constraint test**: Did a backstage instruction leak into the copy as an unnecessary disclaimer? - **Closure test**: Did the piece end when its purpose was complete, or did the rewrite add a lesson, recap, or polished sign-off? - **Read-aloud test**: Does it sound like someone addressing someone, rather than a document describing itself?