Divide polishing into four layers: 'scientific meaning protection, paragraph logic repair, sentence refinement, final verification.' Do not use fluency to mask evidence gaps.
Scope of Work
- Handle translation, restructuring, compression, and language calibration of existing text.
- Leave drafting new sections from materials to
qinyan-nature-writing。 - Do not add experiments, data, statistics, citations, mechanisms, or conclusions not expressed by the authors.
- Do not rewrite correlation as causation, trends as significance, or local results as universal conclusions.
Execution Workflow
- Register immutable facts. Extract values, comparison directions, samples, method names, figures/tables, and citation locations to build a fidelity ledger.
- Identify the text's role. Determine the paper type, section, target readers, target journal, and word limit.
- Diagnose priorities. Process in the order of 'argument breakpoints → paragraph function → syntax and collocation → format'; do not start with synonym replacement.
- Rebuild paragraphs. Let the topic sentence define the paragraph's task, let evidence sentences carry the facts, let explanatory sentences be constrained by the evidence, and let the closing sentence advance naturally.
- Refine sentences. Prioritize clear subjects and actions, and reduce empty nominalizations, stacked modifiers, and redundant metadiscourse.
- Calibrate tone. Choose verbs and qualifiers according to observational, associational, predictive, interventional, and mechanistic evidence.
- Check fidelity. Check values, positive/negative direction, comparators, tense, citation attribution, and scope of negation item by item.
- Run style audit. Execute
python scripts/style_audit.py <file>, manually judge each hit, and do not replace mechanically.
When the complete editing protocol is needed, read references/editing-protocol.md. When processing by section, read references/section-style.md。
Mode Selection
clean: return only the paste-ready version.annotated: return the polished draft, key revisions, and scientific risks.parallel: return a paragraph-by-paragraph comparison of the original and the rewrite.minimal: correct only errors and obviously unnatural expressions, preserving the author's voice as much as possible.
Use annotatedby default; use clean。
Default Output
润色设定
- 章节 / 目标读者 / 模式：
- 不可变事实：
润色稿
[revised text]
关键编辑说明
- [结构、语气、术语或压缩]
科学含义风险
- [原文存在歧义或证据不足之处]
AUTHOR_INPUT_NEEDED
- [必须由作者确认的事实]
Quality Thresholds
- Preserve all values, units, comparison directions, sample identities, and citation intents.
- Keep terminology and abbreviations consistent throughout the text.
- Each paragraph has a clear central task and information progression.
- Avoid clichés, promotional adjectives, false precision, and mechanical stacks of transitional words.
- Use
significant/significantly。 - Do not arbitrarily strengthen the authors' cautious expressions, and do not remove necessary limitations just to make it 'more like Nature'.
- Clearly mark issues that cannot be fixed by language alone; do not pretend the fix is complete.
Resource Routing
| Task | Read |
|---|---|
| Fidelity ledger, four-layer editing, tone calibration, Chinese-to-English translation | references/editing-protocol.md |
| Title, abstract, introduction, results, methods, discussion, conclusion | references/section-style.md |