nature-polishing

interface:
  display_name: "Nature Polishing"
  short_description: "Polish academic prose while preserving factual meaning"
  default_prompt: "Use $nature-polishing to revise this passage into concise Nature-leaning English without changing its factual meaning."

name: nature-polishing
version: 6.6.0
description: >
  Declarative manifest for the static/dynamic split. SKILL.md uses this to
  decide which fragments to load for a given polishing request.

always_load:
  # Shared layer — common to nature-polishing and nature-writing
  - ../nature-shared/core/reader-workflow.md
  - ../nature-shared/core/paper-type-taxonomy.md
  - ../nature-shared/core/ethics.md
  - ../nature-shared/core/terminology-ledger.md
  # Skill-local core
  - static/core/stance.md
  - static/core/failure-modes.md
  - static/core/output-format.md

axes:
  paper_type:
    detect: |
      Decide what kind of paper or section is being polished. Default to research.
      Use the user's stated framing first; fall back to inference from the text.
    values:
      research:       static/fragments/paper_type/research.md
      methods:        static/fragments/paper_type/methods.md
      hypothesis:     static/fragments/paper_type/hypothesis.md
      algorithmic:    static/fragments/paper_type/algorithmic.md
      review:         static/fragments/paper_type/review.md
    default: research
    multi: false

  section:
    detect: |
      Identify which section(s) the user wants polished. The user may name
      one or several. If unclear, ask before loading. Skip this axis only
      when the user provides standalone prose with no section context.
    values:
      abstract:       static/fragments/section/abstract.md
      intro:          static/fragments/section/intro.md
      results:        static/fragments/section/results.md
      discussion:     static/fragments/section/discussion.md
      conclusion:     static/fragments/section/conclusion.md
      title:          static/fragments/section/title.md
      methods:        static/fragments/section/methods.md
    multi: true

  language:
    detect: |
      Detect the source language of the draft. Use zh-to-en when the draft
      is Chinese or shows strong Chinese-influenced English structure.
      Otherwise use en.
    values:
      en:             static/fragments/language/en.md
      zh-to-en:       static/fragments/language/zh-to-en.md
    default: en
    multi: false

  journal:
    detect: |
      Use nature only for the flagship journal Nature. Use nat-comms for Nature
      Communications and nat-mach-intell for Nature Machine Intelligence or
      NMI. If the user says "Nature-family" generically or names another Nature
      Portfolio subjournal, use generic and check that title's current guide.
    values:
      nature:         static/fragments/journal/nature.md
      nat-comms:      static/fragments/journal/nat-comms.md
      nat-mach-intell: static/fragments/journal/nat-mach-intell.md
      generic:        static/fragments/journal/generic.md
    default: generic
    multi: false

references:
  on_demand:
    - condition: polishing, restructuring, or shortening Results or full main text; deciding what belongs in main text versus captions or SI; preventing revision accretion; reducing statistical or claim repetition; or applying a paragraph necessity test
      path: ../nature-shared/core/main-text-discipline.md
    - condition: target is flagship Nature, Nature Communications, Nature Machine Intelligence, or another Nature Portfolio title and the user is polishing or restructuring Results, Discussion, or their boundary; testing claim escalation; deciding whether local interpretation or robustness advances the claim; or separating necessary recap from redundant re-demonstration
      path: ../nature-shared/core/nature-results-discussion.md
    - condition: polishing, restructuring, or auditing any scientific Discussion; repairing its reverse-funnel movement or function sequence; distinguishing interpretation from Results replay; calibrating modal verbs or hedging; making limitations claim-specific; or converting generic future work into uncertainty-resolving next tests
      path: ../nature-shared/core/discussion-argument-language.md
    - condition: target is flagship Nature, Nature Communications, Nature Machine Intelligence, or another Nature Portfolio title and the user is polishing or restructuring the Introduction or whole-manuscript narrative; building a fast problem funnel, exact knowledge gap, literature tension, question-first novelty, compact study roadmap, or Introduction–Results alignment
      path: ../nature-shared/core/nature-introduction.md
    - condition: target is flagship Nature, Nature Communications, Nature Machine Intelligence, or another Nature Portfolio title and the user is polishing or restructuring the abstract; compressing the manuscript into a discovery-centred evidence chain; selecting one main claim, one or two decisive supports, optional core numbers, or a bounded field-level payoff
      path: ../nature-shared/core/nature-abstract.md
    - condition: needs Nature/Nature Communications article-level patterns for abstracts, intros, Results openings, Discussion, conclusions, titles
      path: references/published-article-patterns.md
    - condition: empirical 2025 Nat Commun diction calibration — connector frequencies, the significantly red line, hedge/booster/tense word choices
      path: references/nat-comms-2025-diction.md
    - condition: needs section-specific move orders or phrase patterns
      path: references/section-moves.md
    - condition: needs hedging, transition, evidence, limitation, future-work phrases
      path: references/phrasebank-playbook.md
    - condition: needs academic style, register, article use, or mechanics checks
      path: references/style-guardrails.md
    - condition: polishing or proofreading a full manuscript rather than a passage, or the text has been through more than one round of editing — sweep for accumulated terminology/unit/number drift, headline counts that do not reconcile, claims contradicted by the paper's own tables, tense parallelism, and prose that restates displays
      path: ../nature-shared/core/consistency-sweep.md
    - condition: needs deeper writing-strategy principles
      path: references/writing-strategy.md
    - condition: needs LaTeX layout/typesetting help (排版) — loose or sparse pages, stranded section headings, figures that don't fill the page or split across pages, "Float too large", multi-panel figure arrangement, or Supplementary Information that looks empty
      path: references/latex-layout.md
    - condition: target is Nature Machine Intelligence and exact content-type, word/display, data/code, initial-submission, or production requirements affect the polish
      path: ../nature-shared/journal-formats/nature-machine-intelligence.md

# `nature-polishing` Skill

[中文说明](README.md)

`nature-polishing` polishes, restructures, or translates academic manuscript text into concise Nature-leaning English while preserving author meaning, evidence boundaries, and citation intent.

## What To Use It For

- Translate Chinese academic paragraphs into submission-ready English.
- Shorten long sentences and improve argument order and paragraph movement.
- Adjust abstracts, introductions, results, discussions, or titles using Nature / Nature Communications / Nature Machine Intelligence article patterns.
- Route NMI separately to check its 3,500-word main text, 150-word abstract, six-display budget, code-review duties, and substantial conference-paper extension without importing flagship Nature numbers.
- For flagship Nature, Nature Communications, NMI, and other Nature Portfolio titles, preserve claim escalation, evidence-bound local interpretation, and cross-Results synthesis, distinguishing a necessary recap from redundant re-demonstration.
- For any journal's Discussion, audit the movement from specific findings to bounded implications, coverage of the core rhetorical functions, modality against evidence strength, and whether limitations and future work resolve explicit claim boundaries and open questions.
- For every Nature / Nature Portfolio target, apply a fast problem funnel, exact gap, literature tension, question-first novelty, and Introduction–Results alignment.
- For every Nature / Nature Portfolio target, keep one main discovery, one or two decisive supports or boundaries, only necessary core numbers, and a bounded payoff in the abstract. These three defaults were initially distilled from published NMI papers, with the Results–Discussion guidance further reinforced by flagship Nature comparisons, and generalized as Nature-style guidance; they are not official policy.
- Distinguish research-paper and methods-paper writing priorities.
- Check AI-like phrasing, exaggerated claims, excessive causal language, and unnatural collocations.
- Sweep full or repeatedly revised manuscripts for terminology, unit, numeric-precision, and internal-claim drift.
- Audit Results for main-text necessity: separate core discovery and necessary support from robustness, heterogeneity, and alternative inference; allocate material across main text, captions, and SI; and make every addition trigger a deletion or replacement check.

## Method Sources

- Writing strategy: hourglass structure, reader workflow, and section roles from academic-writing course notes.
- Published article patterns: section moves from selected Nature and Nature Communications papers.
- Phrase support: expression families from Academic Phrasebank.

## Typical Requests

- "Translate this Chinese Results paragraph into Nature-style English, keeping it conservative."
- "Polish this abstract without changing facts or citation intent."
- "This introduction sounds too AI-written; rebuild its logic and language."

## What You Need To Provide

- Source text, target section, paper type, and terms that must be preserved.
- Facts, data, citations, and specialized expressions that must not change.
- Desired output: rewritten text only, or rewritten text with change notes and risk flags.

## Outputs

- Ready-to-paste English rewrite or Chinese-English paired version.
- Key change notes: logic reordering, tone tightening, terminology alignment, and claim boundaries.
- A manuscript-wide consistency risk list, with mechanical warnings inspected in context before revision.
- For main-text compression, an optional result-allocation table, deletion record, statistics-location record, and claim-repetition map.
- Facts or citation intent that require author confirmation.

## Boundaries

- The skill does not add results, mechanisms, statistical significance, or citations not provided by the author.
- It does not exaggerate novelty, causality, or generality to sound more like Nature.
- For building a manuscript section from scratch, use `nature-writing` first.

## Related Skills

- `nature-writing`: section-level drafting and argument rebuilding.
- `nature-response`: reviewer response and cover-letter language.
- `nature-statistics`: statistical text, figure legends, and statistical reviewer responses.

# `nature-polishing` 技能

[English](README_EN.md)

`nature-polishing` 用于把学术手稿文本润色、重构或翻译成更接近 Nature 风格的简洁英文，同时保持作者原意、证据边界和引用意图。

## 适合用它做什么

- 将中文学术段落翻译为投稿可用英文。
- 精简冗长句子，增强论证顺序和段落推进。
- 按 Nature / Nature Communications / Nature Machine Intelligence 论文范式调整摘要、引言、结果、讨论或标题。
- 对 NMI 识别独立路由，检查 3,500 词正文、150 词摘要、6 个 display、代码审查与会议论文实质扩展，不再误用旗舰 Nature 数字。
- 对旗舰 Nature、Nature Communications、NMI 及其他 Nature Portfolio 期刊的 Results–Discussion，保留 claim 递进、证据绑定的局部解释和跨结果综合，区分必要回顾与重复论证。
- 对任何期刊的 Discussion，检查从具体发现到有边界意义的反向漏斗、四类功能是否完整、情态动词是否匹配证据强度，以及局限和未来工作是否指向明确的 claim 与未决问题。
- 对所有 Nature / Nature Portfolio 目标的 Introduction，执行快速问题漏斗、精确 gap、文献张力、问题先行的 novelty 与 Introduction–Results 对齐检查。
- 对所有 Nature / Nature Portfolio 目标的 Abstract，保留一个主发现、1–2 个决定性支撑或边界，只在数字定义或实质支撑核心 claim 时保留，并用有边界的意义句收尾。这三组默认最初来自 NMI 语料归纳，其中 Results–Discussion 又经旗舰 Nature 论文对照加强；它们均非官方规则。
- 区分 research paper 与 methods paper 的写作重点。
- 检查 AI 味、夸张声称、过度因果表达和不自然搭配。
- 对全文或多轮修改稿执行一致性扫描，定位术语、单位、数值精度和内部声称漂移。
- 对 Results 执行主文必要性审计：区分核心发现、必要支撑与稳健性/异质性/替代推断，分配到主文、图注或 SI；每次新增文字都触发删除或替换检查。

## 方法来源

- 写作策略：学术写作课程笔记中的沙漏结构、读者工作流和章节职责。
- 发表论文模式：精选 Nature 与 Nature Communications 文章的 section moves。
- 短语支持：Academic Phrasebank 中适合学术论文的表达族。

## 典型请求

- “把这段中文结果翻译成 Nature 风格英文，保持克制。”
- “润色 abstract，不要改变事实和引用意图。”
- “这段 introduction 太像 AI 写的，帮我重构逻辑和语言。”

## 你需要提供

- 原文、目标章节、论文类型和希望保留的术语。
- 不能改变的事实、数据、引用和专有表达。
- 期望输出：只给改写版，还是给修改说明和风险标记。

## 产出

- 可粘贴的英文改写或中英对照版本。
- 关键修改说明：逻辑重排、语气收敛、术语统一和声称边界。
- 全文一致性风险清单；机械扫描结果逐条回到上下文确认后再修改。
- 主文压缩任务可附结果分配表、删除记录、统计位置记录和 claim repetition map。
- 需要作者确认的事实或引用意图。

## 边界

- 不会替作者新增结果、机制、统计意义或未给出的引用。
- 不会为了更像 Nature 而夸大 novelty、causality 或 generality。
- 如果需要从零搭建论文章节，优先使用 `nature-writing`。

## 相关技能

- `nature-writing`：章节级起草和论证重建。
- `nature-response`：返修回复信和 cover letter 语言。
- `nature-statistics`：统计文本、图注和审稿统计回复。

# LaTeX layout & float typesetting for Nature-style manuscripts

## Contents

- [0. Diagnosis workflow (do this first)](#0-diagnosis-workflow-do-this-first)
- [1. The "loose / 松散" float page — top-align the glue](#1-the-loose-松散-float-page-top-align-the-glue)
- [2. Wide-and-short figures can never fill a portrait page — fix at the source](#2-wide-and-short-figures-can-never-fill-a-portrait-page-fix-at-the-source)
- [3. Do NOT rotate to landscape (Nature house style)](#3-do-not-rotate-to-landscape-nature-house-style)
- [4. Float backlog → stranded section headings](#4-float-backlog-stranded-section-headings)
- [5. [H] and placeins — sharp edges](#5-h-and-placeins-sharp-edges)
- [6. Multi-panel main-text figures — stack, don't cram](#6-multi-panel-main-text-figures-stack-dont-cram)
- [7. Page-count is not the metric — fullness is](#7-page-count-is-not-the-metric-fullness-is)
- [8. Quick checklist](#8-quick-checklist)


Deep reference for **typesetting/layout** requests (排版): loose pages, stranded
headings, figures that don't fill the page, "Float too large", page-splitting
figures, Supplementary Information that looks sparse. This is about *placement*,
not prose. Load it when the user asks to fix layout, spacing, figure placement,
page breaks, or "make it less empty / more compact / more beautiful".

Golden rule: **change → compile → render to image → look → iterate.** Never judge
layout from the `.tex` alone. Measure, don't guess.

---

## 0. Diagnosis workflow (do this first)

1. Compile and read the log for the three signals that matter:
   - `Float too large for page by Xpt` — a float (often figure+caption) exceeds the text height.
   - `Overfull \vbox` — content ran past the bottom margin (often `[H]` placement).
   - `Reference ... undefined` / `Citation ... undefined` — re-run needed (and for `xr-hyper`, compile the SI first, then the main file twice).
2. Render pages to PNG and *look*. With pymupdf:
   `fitz.open(pdf)[i].get_pixmap(dpi=90).save(...)`. Build a **contact sheet**
   (all pages as a grid) to spot whitespace, stranded headings, and split floats at a glance.
3. Measure figure aspect ratios from the source PDFs (`pypdf` mediabox, or pymupdf):
   `aspect = width/height`. Then `displayed_height_pt ≈ textwidth_pt / aspect`.
   For a 1-in-margin `article` on US Letter, `\textwidth ≈ 468 pt`, usable
   `\textheight ≈ 620–650 pt`. Budget every page against that height.

---

## 1. The "loose / 松散" float page — top-align the glue

**Symptom:** a page holding only floats (two tables, or a figure) has a big band of
whitespace in the middle / above, content seemingly centered with gaps.

**Cause:** LaTeX's float-page glue is rubber and spreads floats to fill the page:
`\@fptop = 0pt plus 1fil`, `\@fpsep = 8pt plus 2fil`, `\@fpbot = 0pt plus 1fil`.

**Fix:** top-align floats so they pack from the top and slack collects at the bottom
(reads as an intentional page tail, not "loose"):

```latex
\makeatletter
\setlength{\@fptop}{0pt}            % no stretch at top → floats start at top
\setlength{\@fpsep}{14pt}           % fixed gap between stacked floats
\setlength{\@fpbot}{0pt plus 1fil}  % all slack to the bottom
\makeatother
% Let a page hold more float and less forced text:
\renewcommand{\topfraction}{0.95}
\renewcommand{\bottomfraction}{0.95}
\renewcommand{\textfraction}{0.06}
\renewcommand{\floatpagefraction}{0.80}
\setlength{\textfloatsep}{16pt plus 3pt minus 3pt}
\setlength{\floatsep}{14pt plus 3pt minus 3pt}
```

This single change fixed the spread-apart table pages in practice (a two-table page
went from a centered gap to a tight top-aligned block). Apply it in the SI preamble.

---

## 2. Wide-and-short figures can never fill a portrait page — fix at the source

**Symptom:** a figure (e.g. a 1×N strip of per-dataset bars) sits at full width but is
short (aspect 3:1–4:1), leaving ~40–50% of the page empty; it also tends to strand the
next heading (see §4).

**Why LaTeX can't fix it:** the figure is width-bound at `\textwidth`. A 3.3:1 figure
at 468 pt wide is only ~140 pt tall. You cannot make it taller without making it wider
than the text, and stretching (`height=...` without keeping aspect) distorts it.

**Correct fix — regenerate the figure with a taller aspect** (target ~**1.9:1–2.2:1**)
so two stacked panels + caption fill the portrait page. In matplotlib this is just the
`figsize` height, e.g. `figsize=(17, 4.1)` → `figsize=(17, 6.8)`. Then:

- **Verify faithfulness before trusting regeneration:** re-run at the *original* size
  first and pixel-diff against the shipped PDF. Only if identical do you trust that
  changing the height won't drift fonts/colors/style.
- **Only touch the SI branch.** Main-text and SI versions are often produced by
  different functions (`plot_avg_bar` vs `plot_grouped_bar`); change only the one the
  SI uses, or you'll silently alter approved main-text figures.
- **Font-size caveat:** the on-page text size is set by the *display width* (figure
  scaled to `\textwidth`). Making a figure *taller* does **not** enlarge tick/label
  fonts — it only adds vertical extent (taller bars, more breathing room). To enlarge
  labels you must *narrow* the figure (more scale-up to `\textwidth`) or bump the
  in-plot font, not add height.
- **Mind the caption budget.** A 5–6 line caption is ~80 pt. Size the figure so
  `heading + intro + 2 panels + subcaptions + caption ≤ usable height`. A long-caption
  figure (e.g. a timeline) often needs to be ~0.5 in shorter than a short-caption one.

---

## 3. Do NOT rotate to landscape (Nature house style)

Nature / Nature Communications discourage rotated/landscape figures that force the
reader to turn the page. **Prefer a taller redraw (§2) over `\rotatebox` /
`sidewaysfigure`.** Landscape is *tolerated* in SI for genuinely un-squarable figures,
but try the redraw first. If the user defers ("do what Nature does"), choose upright.

---

## 4. Float backlog → stranded section headings

**Symptom (very common in figure-heavy SI):** two or more section headings pile at the
top of a page above a huge empty gap; their figures appear pages later.

**Cause:** big floats can't be placed, so they defer; meanwhile body text and the next
`\section` keep flowing and stack up. The headings out-run their figures.

**Fix — bind each figure section into a "heading + figure" unit:**

1. Make the figure short enough to *share a page* with its heading + intro
   (`heading + intro + figure ≤ usable height`; see §2 caption budget).
2. Start the section on a fresh page: `\clearpage` immediately before the `\section`.
   This guarantees full height is available.
3. Pin the figure with `[H]` (needs `\usepackage{float}`) right after the heading/intro
   so it cannot float away:

```latex
\clearpage
\section{Per-dataset context scaling}
Figure~\ref{fig:scaling} shows ...        % short intro
\begin{figure}[H]                         % H = exactly here, no floating
  ...two stacked panels...
  \caption{...}\label{fig:scaling}
\end{figure}
```

Result: a fresh page with heading + intro + figure + caption that fills the page, with
no stranded heading.

**Also pin the *preceding* section's trailing figure.** A `\clearpage` before a heading
flushes the page early — if the previous section ended with a big floating figure whose
intro text is still on the page, you strand that text instead. Give that trailing figure
`[H]` too (and size it to fit with its intro), so the page before the `\clearpage` is
also full. (In practice: a timeline figure had to be shrunk ~10% so its intro paragraph
+ figure + long caption fit one page.)

---

## 5. `[H]` and `placeins` — sharp edges

- **`[H]`** (float package) does not float. If the figure doesn't fit in the remaining
  space it either page-breaks *before* itself (stranding preceding text) or overflows
  (`Overfull \vbox`). Safe only when you guarantee room — hence pair it with `\clearpage`
  and size-to-fit. Great for the heading+figure unit; risky mid-page.
- **`\usepackage[section]{placeins}`** puts a `\FloatBarrier` at every `\section`, keeping
  a section's floats inside it. Useful against backlog — **but** it flushes the previous
  float and can strand the *current* heading alone on a page if that section's figure is
  too tall to share. It fixed a two-heading pile-up yet produced one-heading-per-empty-page
  until figures were also shortened. Use it *with* §2/§4 sizing, not as a standalone cure.
- **Don't blanket-`\clearpage` every section.** Section-boundary clearpages create
  near-empty pages when a section is short. Use `\clearpage` surgically (figure sections,
  §4), not everywhere. (Removing 8 blanket clearpages once took an SI 23→21 pages with
  even density.)

---

## 6. Multi-panel main-text figures — stack, don't cram

- A 2×2 grid shrinks panels; for **wide** panels prefer a **single-column vertical stack**
  (`a` over `b` over `c` over `d`), each near full width, on a dedicated float page `[p]`.
  Reads larger and matches a 2-panel `a/b` stacked figure elsewhere.
- Tune inter-panel space with `\vspace` between `subfigure`s (e.g. `0.3em–0.8em`).
- If you see `Float too large for page by X pt` and X is small (<5 pt), shave the
  smallest thing first: reduce inter-panel `\vspace`, or panel width by ~2%, before
  touching the figure itself.

```latex
\begin{figure}[p]\centering
  \begin{subfigure}{0.66\textwidth}\includegraphics[width=\textwidth]{a}\caption{}\end{subfigure}
  \vspace{0.3em}
  \begin{subfigure}{0.66\textwidth}\includegraphics[width=\textwidth]{b}\caption{}\end{subfigure}
  ...
  \caption{...}\label{...}
\end{figure}
```

---

## 7. Page-count is not the metric — *fullness* is

Tightening loose pages can *reduce* count; enlarging short figures to fill pages can
*increase* it. Both are correct. A 20-page SI where every page is full beats a 19-page SI
with half-empty figure pages. Optimize for even density and no stranded headings, not for
the smallest page count.

---

## 8. Quick checklist

- [ ] Top-aligned float glue in the preamble (§1).
- [ ] Wide-short figures regenerated to ~2:1, SI branch only, faithfulness pixel-checked (§2).
- [ ] Upright, not landscape (§3).
- [ ] Each figure section: `\clearpage` + `[H]` heading-figure unit; trailing figure of the
      previous section also `[H]` and sized to fit (§4).
- [ ] No blanket section clearpages; no stranded headings (§5).
- [ ] Compile clean: no "Float too large", no "Overfull \vbox", no undefined refs.
- [ ] Contact-sheet the final PDF and eyeball every page (§0).

# Nature Communications 2025 — Diction & Connector Calibration

Use this file during the sentence-level polish pass when you want **empirical,
quantified word-choice calibration** to back up `style-guardrails.md` and
`published-article-patterns.md`. The preferences below are measured from a 2025
reading set of 20 open-access *Nature Communications* computer-science / AI
articles. They are calibration data, not rules to apply blindly — a discipline
or a specific journal house style overrides them. **Do not copy source wording.**

## 1. Connectors — observed preference order

句首大写连接词词频(5 篇子集合计):
`However` **51** ≫ `Furthermore` 22 · `Therefore` 19 · `Overall`/`Notably` 16 · `In addition` 13 · `In contrast` 6 · `Moreover` 4 · `Importantly` 2.

- **However** carries turns, gap-opening, and surprises — do not scatter weaker
  alternatives where `However` is idiomatic.
- For addition, prefer **Furthermore** over **Moreover** (22 vs 4 here).
- Cause/close with **Therefore**; summarise a block with **Overall / In summary / In conclusion**.
- Reserve **Notably / Importantly / Interestingly / Surprisingly** for the
  paragraph's key finding, not as routine sentence openers.

## 2. Boosters — the `significantly` red line

- **`significantly` appears 0 times in the corpus.** When a draft leans on
  "significantly (better/higher/improved)", first check it is a *statistical*
  claim with a test behind it; otherwise replace with **notably /
  substantially / considerably / markedly**.
- This extends the `style-guardrails.md` overclaim list: treat blanket
  intensifiers as a smell, and attach every booster to a number or a test.

## 3. Hedges — concentrated, not sprinkled

- Primary hedges are **may** and **potential**; `might / could / likely` are
  secondary. They cluster in meaning/Discussion sentences, not in Results
  reporting. Pattern marker: *"Encoding these mechanisms **may** help further
  improve the performance."*
- Keep Results sentences assertive + numeric; move the hedge to the
  interpretation sentence.

## 4. Achievement verbs — the house vocabulary

Frequent, defensible when backed by data: **achieve · demonstrate · outperform
· superior · robust · generalizable · comparable**. When the result is *weaker*
than a baseline, state it honestly and reframe with **comparable** +
concession: *"**Despite the smaller scale** of our pre-training data …, the
**comparable** performance highlights the effectiveness."* Do not upgrade
`comparable` to `superior`.

## 5. Tense & voice (polish-pass checks)

- Background / property = present; specific operation = past; figure
  description = present. Flag accidental past-tense for a standing property
  (*"CataPro demonstrates…"* not *"demonstrated"* when stating a capability).
- Active **we** for narrative and claims; passive for apparatus/method only.
  Flag passive over-use that hides the agent in claim sentences.

## 6. Sentence-skeleton phrases (calibrated to 2025 corpus)

- Abstract hinge: `Here we show/present X (FULL NAME, abbr.), a … that …`
- Gap: `However, … remains a challenge` / `the scarcity of … hinders …` /
  `Without X, Y cannot be quantified`.
- Result close: `These findings collectively affirm/confirm that …`
- Significance (soft promise, not a guarantee): `promises to / offers
  potential / paves the way`.
- Title: no number, no result — keep digits for the abstract.

## 7. 中文润色要点

- 连接词:转折/制造空白优先"然而";递进偏"此外/进一步"(对应 Furthermore);
  因果用"因此";收束用"总体而言/综上"。
- 慎用"显著"——英文 `significantly` 在本语料 0 次。无统计检验支撑时改"明显/大幅/相当/尤为"。
- 对冲(可能/提示/有望)集中在意义句,结果句保持带数字的肯定陈述。
- 战绩措辞(实现/证明/优于/稳健/可泛化/相当)需有数据支撑;弱于对手时用"相当"+"尽管规模更小"客观化,不拔高为"优于"。

# Phrasebank Playbook

## Contents

- [Evidence strength](#evidence-strength)
- [Evidence collocations](#evidence-collocations)
- [Transition families](#transition-families)
- [Paragraph linking without sounding repetitive](#paragraph-linking-without-sounding-repetitive)
- [Gap language](#gap-language)
- [Comparison with prior work](#comparison-with-prior-work)
- [Limitation language](#limitation-language)
- [Implication language](#implication-language)
- [Future-work language](#future-work-language)


Use this file after the main argument and section role are already clear. It is a phrasebank layer derived from `Academic Phrasebank`, not a substitute for deciding what the paragraph is trying to do.

## Evidence strength

Choose verbs that match the evidence.

### Strong

- `show`
- `demonstrate`
- `establish`
- `reveal`
- `identify`

Use only when the design and data justify a strong claim.

### Moderate

- `suggest`
- `indicate`
- `support the view that`
- `are consistent with`
- `point to`

Use when the interpretation is plausible but not definitive.

### Speculative

- `may reflect`
- `could arise from`
- `appears to`
- `seems likely`
- `might be explained by`

Use when moving beyond direct observation.

## Evidence collocations

Adjectives for evidence:

- weak: `limited`, `scant`, `insufficient`
- developing: `growing`, `emerging`, `accumulating`
- strong: `robust`, `reliable`, `convincing`, `considerable`

Useful patterns:

- `The evidence presented here suggests that ...`
- `The available evidence supports the view that ...`
- `Current evidence raises important questions about ...`
- `The data point to a need for ...`

## Transition families

### Contrast

- `however`
- `by contrast`
- `nevertheless`
- `despite this`
- `whereas`

### Addition

- `furthermore`
- `in addition`
- `moreover`
- `also`

### Consequence

- `therefore`
- `thus`
- `consequently`
- `as a result`
- `thereby`

### Qualification

- `notably`
- `importantly`
- `approximately`
- `in part`
- `at least in this cohort`

Prefer the smallest connective that does the job. Do not decorate every sentence with a transition word.

## Paragraph linking without sounding repetitive

Prefer these patterns over repeated `This suggests`:

- restate the noun: `Such heterogeneity ...`
- definite noun phrase: `The resulting gradient ...`
- participial summary: `Taken together, ...`
- zero-connective progression when the logic is already obvious

Limit demonstrative-led openings. One per paragraph is usually enough.

## Gap language

Use gap statements that are precise rather than dramatic:

- `remains poorly understood`
- `has not been examined in ...`
- `has received limited attention`
- `few studies have addressed ...`
- `evidence remains sparse for ...`

Avoid:

- `no one has ever studied`
- `completely unknown`
- `ignored by all previous work`

## Comparison with prior work

To align with earlier work:

- `These results are consistent with ...`
- `This finding accords with ...`
- `Our observations broadly support ...`

To mark divergence fairly:

- `In contrast to earlier reports, ...`
- `This finding differs from ...`
- `One possible reason for this discrepancy is ...`

## Limitation language

Useful patterns:

- `These findings should be interpreted with caution because ...`
- `A limitation of this study is that ...`
- `The generalisability of these results is limited by ...`
- `We cannot exclude the possibility that ...`
- `Another source of uncertainty is ...`

Pair limitation language with the actual source of uncertainty, not with vague modesty.

## Implication language

Useful patterns:

- `An implication of this is that ...`
- `These findings may help to explain ...`
- `These data support further investigation of ...`
- `This work has implications for ...`

Implications should stay within the evidence boundary.

## Future-work language

Useful patterns:

- `Further work is needed to determine whether ...`
- `Future studies should examine ...`
- `A useful next step would be to ...`
- `Larger studies are required to validate ...`

Future work should emerge from an actual limitation, uncertainty, or opportunity.

# Published Article Patterns

## Contents

- [Abstract pattern](#abstract-pattern)
- [Introduction pattern](#introduction-pattern)
- [Results pattern](#results-pattern)
- [Discussion pattern](#discussion-pattern)
- [Conclusion pattern](#conclusion-pattern)
- [Title pattern](#title-pattern)
- [Sentence-level pattern](#sentence-level-pattern)
- [Overclaim checks from article patterns](#overclaim-checks-from-article-patterns)


Use this file when polishing should improve scientific argument, not just English.
The patterns below are distilled from curated Nature and Nature Communications
research articles across materials, energy systems, construction decarbonization
and machine learning. Do not copy their wording. Use the patterns to diagnose
and reshape the user's manuscript.

## Abstract pattern

Strong abstracts usually move in six steps:

1. name the field-scale problem or opportunity
2. show why existing approaches are incomplete
3. state the specific intervention or method
4. give the decisive result with scale, comparison or constraint
5. explain why the result changes what is possible
6. close with scope, application or boundary

Polishing rule: if an abstract starts with the method, add the problem and gap
first. If it ends with enthusiasm, replace it with a bounded implication.

## Introduction pattern

High-performing introductions often use `scale -> bottleneck -> prior attempts ->
missing capability -> present study`.

- `Scale`: quantify why the problem matters, preferably with field-level stakes.
- `Bottleneck`: name the physical, computational, clinical or practical barrier.
- `Prior attempts`: acknowledge existing strategies fairly.
- `Missing capability`: explain what those strategies still cannot do.
- `Present study`: state what the paper does, not what it hopes to do.

Polishing rule: keep the gap narrow enough that the study can actually fill it.
Avoid novelty claims that depend on weakening prior work.

## Results pattern

Results sections usually work best as an evidence ladder:

1. overview of the system, workflow or design space
2. validation that the platform works under controlled conditions
3. primary performance result
4. comparison with baselines or current practice
5. mechanism, interpretation or diagnostic analysis
6. scale-up, generalization or real-world application

Each Results paragraph should begin with the question or test, then report the
observation, then give the quantitative or comparative support. Interpretation
should be brief unless the paragraph is explicitly bridging into Discussion.

## Discussion pattern

Effective Discussion writing starts from the central advance and then widens:

- what the study demonstrates
- why the evidence is credible
- how it changes an existing workflow, design rule or conceptual boundary
- what constraints remain
- what future work is enabled, without promising untested outcomes

Do not turn the Discussion into a second Results section. Use it to state what
the results mean and where that meaning stops.

## Conclusion pattern

Nature-style conclusions are compact. They usually combine:

1. the contribution in one sentence
2. the mechanism, performance or feasibility evidence that supports it
3. the scale or application implied by the work
4. one boundary condition, if the claim could otherwise overreach

Polishing rule: remove new data from conclusions. Preserve confidence, but add
scope control.

## Title pattern

Strong titles are concrete and searchable. They often combine:

- object or system
- action or capability
- application, scale or consequence

Prefer titles that reveal the central scientific move. Avoid titles that sound
like grant aims, slogans or broad fields.

## Sentence-level pattern

Published prose often looks simple because each sentence does one job:

- background sentence: field stake or known fact
- gap sentence: unresolved limitation
- method sentence: what was done
- result sentence: observed effect plus condition
- comparison sentence: baseline or previous state
- implication sentence: meaning with scope
- limitation sentence: boundary or dependency

When polishing, label each sentence by job. If two jobs compete in one sentence,
split it or subordinate one job.

## Overclaim checks from article patterns

Flag and soften claims when:

- a laboratory or simulation result is written as an immediate field-wide solution
- a single material, model or system is described as universally applicable
- correlation is rewritten as mechanism
- a comparison lacks a fair baseline
- a future application is stated as an achieved outcome

Good Nature-leaning prose can be ambitious, but the ambition must be attached to
evidence, scale and boundary.

# Section Moves

## Contents

- [Introduction](#introduction)
- [Literature Review](#literature-review)
- [Methods](#methods)
- [Results](#results)
- [Discussion](#discussion)
- [Conclusion](#conclusion)
- [Abstract](#abstract)
- [Title](#title)


Use this file only after the main section logic has been decided in `SKILL.md`. This file is for phrase-level and move-level support derived from `Academic Phrasebank`, not for deciding the paper's overall writing strategy.

## Introduction

Questions this section must answer:

1. Why does the topic matter?
2. What is already known?
3. What is still missing or contested?
4. What does the present study ask or do?

Preferred move order:

1. establish importance
2. summarize what is known
3. identify a gap, limitation, or controversy
4. state the study aim
5. indicate value or approach

Useful phrase families:

- `Recent years have seen increasing interest in ...`
- `X is a central issue in ...`
- `Previous studies have shown that ...`
- `However, the mechanisms underlying ... remain poorly understood.`
- `Few studies have examined ...`
- `Here, we investigate whether ...`
- `This work provides ...`

Avoid:

- long historical throat-clearing
- detailed results
- inflated novelty claims before the gap is defined

## Literature Review

Questions this section must answer:

1. What lines of work define the field?
2. What has been established?
3. Where do findings diverge or remain incomplete?
4. Which gap matters for the present paper?

Preferred move order:

1. describe the scope of existing work
2. identify dominant approaches
3. state what has been established
4. note disagreements or contradictions
5. isolate the missing piece

Useful phrase families:

- `A substantial body of work has focused on ...`
- `Most studies have relied on ...`
- `Previous work has established that ...`
- `Findings have been mixed regarding ...`
- `By contrast, little attention has been paid to ...`
- `No study has yet examined ...`

Avoid:

- citation-by-citation summary
- treating all prior work as uniformly weak

## Methods

Question this section must answer:

- Could another group reproduce the work from this description, or from this description plus a clearly cited protocol?

Preferred move order:

1. design or cohort
2. materials or data source
3. procedure
4. outcome measures
5. analysis and statistics
6. ethics when relevant

Useful phrase families:

- `A cross-sectional study was undertaken to ...`
- `Samples were collected from ...`
- `X was quantified using ...`
- `We used ... to assess ...`
- `Differences were analysed using ...`
- `All analyses were performed in ...`

Avoid:

- `under standard conditions`
- `using routine methods`
- `data were analysed statistically`

## Results

Question this section must answer:

- What was observed, under which condition, and with what evidence?

Preferred move order:

1. orient the reader to the figure, table, or experiment
2. state the main observation
3. add quantitative detail
4. note expected or unexpected patterns
5. compare with prior work only if it clarifies the result

Useful phrase families:

- `Figure 1 shows ...`
- `As shown in Table 1, ...`
- `The most notable finding was that ...`
- `Contrary to expectations, ...`
- `No significant difference was observed in ...`
- `These results are consistent with ...`
- `In contrast to earlier reports, ...`

Avoid:

- discussion-length mechanism explanations
- repeating every visual detail from the figure

## Discussion

Questions this section must answer:

1. What do the main findings mean?
2. How do they relate to earlier work?
3. Which explanations are plausible?
4. What limitations constrain interpretation?
5. What follows from the findings, and what does not?

Preferred move order:

1. restate the main finding
2. explain plausible reasons
3. compare with earlier work
4. note limitations
5. state implications
6. point to future work if needed

Useful phrase families:

- `Taken together, these findings suggest that ...`
- `A possible explanation is that ...`
- `This discrepancy may reflect ...`
- `These results should be interpreted with caution because ...`
- `An implication of this is that ...`
- `Further work is needed to determine whether ...`

Avoid:

- repeating the Results section in new words
- claiming mechanism when only association was shown

## Conclusion

Questions this section must answer:

1. What was the central contribution?
2. Which finding matters most?
3. What implication follows, with what boundary?

Preferred move order:

1. return to the aim
2. summarize the decisive finding
3. state contribution or significance
4. give a boundary or forward look

Useful phrase families:

- `This study set out to ...`
- `The present findings indicate that ...`
- `These results extend our understanding of ...`
- `Notwithstanding these limitations, ...`
- `Further studies are required to ...`

Avoid:

- introducing new experiments
- ending on vague praise of the work

## Abstract

Questions this section must answer:

1. What problem or gap is being addressed?
2. What was done?
3. What was found?
4. Why should the reader care?

Preferred move order:

1. broad context
2. concrete gap
3. approach
4. key result with numbers if available
5. implication

Useful phrase families:

- `X remains challenging because ...`
- `Here, we ...`
- `Using ... , we found that ...`
- `We show that ...`
- `These findings suggest ...`

Keep the abstract selective. If a detail does not affect editorial triage, it probably does not belong.

## Title

Question this section must answer:

- Which few words make the paper searchable, accurate, and interesting without overclaiming?

Target properties:

- searchable
- specific
- restrained
- defensible

Useful patterns:

- `[Core entity] in/through/by [mechanism or context]`
- `[Process] shapes [outcome] in [system]`
- `[Signature/pattern/framework] of [phenomenon]`

Avoid:

- `A study of ...`
- vague hooks
- unverified `first`
- stacked jargon

# Style Guardrails

Use this file for mechanical and stylistic checks after the main rewrite. This file should refine prose and correctness, not override the main writing strategy in `SKILL.md`.

## Academic style

- prefer cautious, precise prose over conversational confidence
- avoid contractions
- avoid rhetorical questions in polished manuscript prose
- define abbreviations on first use
- use British spelling by default if the target is Nature-style prose
- keep figure legends concise; if aiming for Nature style, `<= 300` words is a good upper bound
- if aiming for Nature style, keep titles at `<= 75` characters including spaces

## Articles

Common checks:

- first mention of a singular count noun: `a` or `an`
- later mention of the same item: `the`
- generic plural: usually no article
- unique entity: often `the`
- abstract nouns used generally: often no article

Typical repair:

- bad: `The hypoxia induces ...`
- better: `Hypoxia induces ...`

## Numbers and units

- use numerals for measurements
- leave a space between the value and the unit: `25 cm`, `3.2 s`
- keep statistical symbols and mathematical notation consistent
- use en dashes for ranges where appropriate

Do not rewrite numbers into words unless the surrounding house style demands it.

## Academic register

- avoid spoken fillers and weak evaluative language
- use `we` only when it suits the discipline and document type
- keep nominalisation useful, not excessive
- keep the prose impersonal where appropriate, but do not force lifelessness

## Sentence and paragraph checks

- each sentence should express one main proposition
- dependent clauses must stay attached to a main clause
- do not join two independent clauses with only a comma
- each paragraph needs a controlling idea and supporting material
- avoid common structure errors such as sentence fragments introduced by `although` or `whereas`

## Overclaim checklist

Flag and soften:

- `prove`
- `conclusively`
- `unprecedented`
- `best`
- `superior`
- `first`

Safer replacements:

- `show`
- `suggest`
- `to our knowledge`
- `among the strongest`
- `in this cohort`

## Integrity rules

- do not invent references
- do not alter quantitative values unless correcting an obvious typo requested by the user
- do not upgrade association to causation
- do not imply broader generalisability than the study supports

## AI boundary

Use AI for language control, not for scientific fabrication.

Allowed:

- grammar and clarity
- restructuring and hedging
- translation with terminology checking

Not allowed:

- fabricated citations or datasets
- invented mechanisms presented as fact
- unsupported claims of novelty

# Writing Strategy

## Contents

- [Core stance](#core-stance)
- [Hourglass structure](#hourglass-structure)
- [Writing order is not reading order](#writing-order-is-not-reading-order)
- [Claim, evidence, boundary](#claim-evidence-boundary)
- [Section responsibilities](#section-responsibilities)
- [Citation as positioning](#citation-as-positioning)
- [Fairness to earlier work](#fairness-to-earlier-work)
- [Overclaim control](#overclaim-control)


Use this file when the user is not just asking for cleaner English, but for better scientific writing logic. This is the layer that should govern all paragraph- and section-level rewriting.

## Core stance

Academic polishing is not only about style. It is also about making the reasoning legible. A polished paragraph that still performs the wrong rhetorical job is a failed edit.

## Hourglass structure

Most strong research writing follows a `broad -> narrow -> broad` pattern:

- `Introduction`: open the territory, narrow to the gap, then state the study
- `Discussion/Conclusion`: start from the specific findings, then widen to implications and limits

Use this pattern when deciding paragraph order and section scope. If a draft jumps between background, results, and implications without control, rebuild the progression first.

## Writing order is not reading order

The author may draft in one order and the reader may consume in another. A useful planning sequence is:

1. results
2. introduction and conclusion
3. title
4. discussion
5. methods
6. abstract

The practical rule for this skill is simple: organize around evidence and argumentative function, not around the chronology of the raw draft.

## Claim, evidence, boundary

Every important scientific statement should have three parts:

1. `claim`: what is being said
2. `evidence`: what supports it
3. `boundary`: where the claim stops, or what uncertainty remains

Typical failures:

- claim without evidence
- data without an explicit point
- implication without a scope condition
- correlation rewritten as mechanism

When polishing, repair these failures before polishing rhythm.

## Section responsibilities

### Introduction

The Introduction should answer four questions:

1. What is already known?
2. What remains unresolved?
3. What exact question does this study ask?
4. How does the study address it?

Do not summarize results or conclusions here.

### Results

Results state what was observed. They should provide:

- object or system
- condition
- quantitative support
- direct result

Do not turn Results into a Discussion section by adding long mechanistic interpretation.

### Discussion

Discussion explains what the findings mean. It should address:

- how the work fits the broader field
- what has been added to understanding
- which earlier work is being supported, revised, or complicated
- which explanations are plausible
- which limitations constrain the interpretation

Discussion is the natural home for hedging.

### Methods

Methods should pass a reproducibility test: could another group repeat the work from this description, or from this description plus a clearly cited prior protocol?

Reject vague writing such as:

- `under standard conditions`
- `using routine methods`
- `data were analysed statistically`

### Conclusion

Conclusion is not a mini-discussion. A strong closing usually does three things:

1. restates the central contribution
2. identifies the decisive evidence
3. states the implication with a boundary

Do not introduce new data here.

### Abstract

The abstract is a mini-paper:

1. context or problem
2. gap
3. approach
4. key result
5. implication

It should help the reader decide whether the paper is relevant, credible, and potentially important.

## Citation as positioning

Citation is not just a formatting issue. It tells the reader how the current work stands relative to earlier work.

Useful categories:

- `support`: prior work supports the premise
- `borrow`: current work adopts a method, framework, or protocol
- `contrast`: current work differs in result, setting, or interpretation
- `reuse/adaptation`: material, data, code, or images come from elsewhere

Always cite the source actually read and verified. Do not cite a paper as direct support if you only know it through another paper's summary.

## Fairness to earlier work

Do not manufacture novelty by flattening previous studies into a weak baseline. Prefer language like:

- `Although previous studies showed ..., their performance in ... remains unclear.`
- `Earlier work established ..., but did not address ...`

This preserves intellectual honesty while still making the gap explicit.

## Overclaim control

Watch for:

- `prove`
- `conclusively`
- `unprecedented`
- `best`
- unqualified `first`

Replace or qualify them unless the evidence is unusually strong and the scope is tightly defined.

---
name: nature-polishing
description: Polish, restructure, or translate academic prose into concise Nature-leaning English while preserving facts, evidence boundaries, terminology, and citation intent. Use for manuscript paragraphs, abstracts, introductions, Results, discussions, conclusions, titles, Methods, Chinese drafts, proofreading, language editing, and general academic or scientific writing. Also use to shorten bloated Results, allocate evidence across main text, captions, and Supplementary Information, prevent reviewer-driven revision accretion, reduce repeated statistics or claims, and apply paragraph-necessity checks. Covers LaTeX layout or typesetting fixes such as sparse pages, stranded headings, oversized or split figures, float errors, multi-panel arrangement, and sparse Supplementary Information via references/latex-layout.md. Trigger on 学术写作、科研写作、论文润色、SCI写作、英文论文润色、语言润色、润色、改写、学术英语、排版.
---

# Nature-Style Academic Polishing — Router

This skill is split into two layers:

- A **static layer** under `static/` that holds versioned, reusable content fragments (core principles, paper-type playbooks, per-section guidance, language-specific rules, per-journal style).
- A **dynamic layer** (this file plus `manifest.yaml`) that detects the request's axes and loads only the fragments needed for the current job.

Do not try to apply the polishing logic from memory or from this router. Always load fragments from disk as described below.

## Routing protocol

Follow these five steps every time the skill is invoked.

### 1. Load the manifest and the core layer

Read [manifest.yaml](manifest.yaml). It declares the axes (`paper_type`, `section`, `language`, `journal`), the allowed values, and the file paths each value maps to.

Also read every file listed under `always_load`. These hold the default stance, failure-mode diagnosis, ethics, and output format that apply to every polish job.

### 2. Detect the axis values for this request

For each axis in the manifest, decide the value using the manifest's `detect:` hint and the user's input:

- `paper_type` — research / methods / hypothesis / algorithmic / review. Default: research.
- `section` — abstract / intro / results / discussion / conclusion / title / methods. May be multiple. Ask the user if it is ambiguous and matters for the polish.
- `language` — en or zh-to-en. Detect from the draft itself.
- `journal` — nature / nat-comms / nat-mach-intell / generic. Default:
  generic. Use `nature` only for flagship Nature, `nat-comms` for Nature
  Communications and `nat-mach-intell` for Nature Machine Intelligence (NMI).
  Do not route another Nature Portfolio title through flagship Nature rules.

State the detected axis values in one short line to the user before proceeding, so they can correct you cheaply.

### 3. Load the matching fragments

For each axis value, Read the file mapped in the manifest. Skip the `section` axis only if the user has supplied free-floating prose with no section context.

Do **not** read every fragment in `static/`. Load only what step 2 selected.

### 4. Polish using the loaded material

Apply the loaded fragments in this priority order, matching the `paper type -> section job -> paragraph logic -> claim/evidence/boundary -> sentence polish` rule from `core/failure-modes.md`:

1. Paper-type playbook (architecture, writing order).
2. Section-specific job and failure modes.
3. Journal-specific framing and constraints.
4. Language-specific sentence and paragraph rules (apply last).
5. Core stance and ethics throughout.

If a paragraph's structural problem cannot be fixed without inventing content, flag it instead of papering over it.

For Results, full-main-text compression, main-versus-SI allocation, or prose
added during revision, load `../nature-shared/core/main-text-discipline.md`
before sentence polishing. Classify each result, retain the shortest sufficient
evidence chain, and require every addition to trigger a deletion or replacement
check across the affected paragraph.

For flagship Nature, Nature Communications, Nature Machine Intelligence, or
another Nature Portfolio title, load the matching shared Nature-style corpus
guidance:

- Results or Discussion →
  `../nature-shared/core/nature-results-discussion.md`
- Introduction or whole-manuscript narrative →
  `../nature-shared/core/nature-introduction.md`
- Abstract → `../nature-shared/core/nature-abstract.md`

Preserve claim escalation, the fast question funnel, Introduction–Results
alignment, discovery-centred abstract compression, evidence-bound local
interpretation, and cross-Results synthesis. These defaults were initially
distilled from published NMI papers; treat them as corpus-derived guidance, not
official policy, and obey the target journal's current rules when they differ.

For any Discussion polish or restructuring job, also load
`../nature-shared/core/discussion-argument-language.md`. Use its function labels
to remove Results replay, repair the movement from specific findings to bounded
implications, calibrate modal and reporting verbs to evidence strength, and make
limitations and future work resolve named claim boundaries. Treat it as general
writing guidance, not journal policy.

### 5. Reach for references only when needed

The files under `references/` are deep references, not defaults. Open them on demand per the `references.on_demand` table in the manifest, for example when the user explicitly asks for phrasebank-style alternatives or a stricter style audit.

When the target is Nature Machine Intelligence and exact limits, availability
sections, conference-extension disclosure or production checks affect the
revision, load
`../nature-shared/journal-formats/nature-machine-intelligence.md`.

When the job is a whole manuscript rather than a passage, or the text has already been through more than one round of editing, also load `../nature-shared/core/consistency-sweep.md`. Polishing passage by passage cannot see accumulated drift: one experimental factor under several names, the same quantity in two units, a metric at two precisions, or a superlative the paper's own table contradicts. Sweep for those before working on sentences, and repeat the sweep until a pass finds nothing new.

**Layout/typesetting (排版) requests are different.** If the user asks to fix
*placement* rather than wording — loose/sparse pages, stranded headings, figures
that don't fill the page or split across pages, "Float too large", multi-panel
arrangement, sparse Supplementary Information — skip the prose axes (paper_type,
section, language, journal) and load `references/latex-layout.md` directly. That
file is self-contained: it carries the diagnosis workflow (render → contact-sheet →
read the log), the float-glue and `[H]`/`\clearpage`/`placeins` patterns, and the
"regenerate wide figures taller at the source" rule. Always compile and visually
inspect rendered pages before and after — never judge layout from the `.tex` alone.

## Why this split

- The static layer is versioned and reviewable. Adding a new journal style or paper type is one new file plus one manifest line.
- The dynamic layer keeps each invocation cheap: only the fragments relevant to this draft enter context, instead of the full 1000-line monolith.
- The router itself is short on purpose. Update fragments, not this file, when adding scope.

# Diagnose failure mode before editing

Before rewriting, identify the main problem:

- wrong paper type logic
- missing gap or poor positioning
- claim without evidence
- evidence without a clear claim
- missing boundary or limitation
- Results and Discussion mixed together
- weak title or abstract signal
- inconsistent terminology, abbreviations, units, or notation across sections
- evidence-complete but main-text-bloated Results
- reviewer-driven additions appended without deleting redundant prose
- the same claim or full statistical report repeated across Results and captions
- sentence-level clutter only

Prioritize fixes in this order:

`paper type -> section job -> evidence placement -> paragraph necessity -> claim/evidence/boundary -> sentence polish`

Do not sentence-polish a draft whose section job is wrong. Surface the structural problem first, then polish.

For Results or full-main-text work, load
`../../../nature-shared/core/main-text-discipline.md`. Resolve evidence placement,
revision accretion, explanatory recursion, and claim repetition before editing
sentence rhythm.

Terminology consistency is a cross-cutting check that runs at every level: build the Terminology Ledger on first contact (see `../../../nature-shared/core/terminology-ledger.md`) and enforce its canonical forms throughout the polish.

# Output format

Default output:

1. The polished text as plain prose, not in a code block.
2. `Revision notes:` with `3-5` short bullets on the major structural and stylistic changes.
3. If the rewrite changed section logic, say so explicitly.

If the user asks for side-by-side revision, provide:

- `Original`
- `Polished`
- `Why changed`

If any paragraph's structural problem could not be fixed without inventing content, say so under `Revision notes:` instead of papering over it.

When the main-text discipline is triggered, add a compact
`Main-text discipline audit:` that identifies material kept, replaced,
compressed, relocated to SI/caption, or deleted; states which descriptive and
primary inferential quantities remain in the main text; and reports the
before/after word count. Do not bury the polished prose under the audit.

# Default stance

- Language serves argument. Do not polish sentences while leaving the reasoning broken.
- Write with empathy for the reader: relevance first, then novelty, then trust, then reuse, then meaning.
- There should be no mystery for the writer, but there may be one for the reader.
- Do not invent data, references, mechanisms, or novelty claims.
- Do not let AI draft the paper's core scientific argument from scratch.
- If the draft is Chinese or structurally rough, reconstruct the logic first and the prose second.
- On first contact with the draft, build a Terminology Ledger and keep terms, abbreviations, units, and notation consistent across every section. Do not introduce synonyms to vary the prose. See `../../../nature-shared/core/terminology-ledger.md`.
- Avoid em dashes in polished output by default. Prefer commas, parentheses, or full stops. Use colons sparingly unless the user explicitly asks to preserve dash-based punctuation or wants a colon-led style.

## Reader workflow

See `../../../nature-shared/core/reader-workflow.md` (loaded via manifest `always_load`) for the 5-step reader question sequence. Polishing should help the paper answer those questions in order.

## Protect the core argument

The paper's core argument includes:

- the scientific question the paper actually answers
- why that question matters
- how the work differs from existing research
- what the results imply
- how the main line of reasoning unfolds

AI may help polish, structure, or compare phrasings. AI should not invent or author the core argument. If the argument is weak or unclear, expose that weakness rather than hiding it under polished language.

# Journal: generic

Used when the user has not named a target journal, or the journal is not specifically modeled by another fragment.

## Defaults

- Apply Nature-leaning style without enforcing Nature's strictest length or significance-framing demands.
- Keep the em-dash and hedging defaults from `core/stance.md`.
- If the user later names a journal, ask whether to re-polish under that journal's fragment rather than guess.

## Things to ask the user before final polish

- Target journal and section format (structured vs unstructured abstract; word limits; reference style).
- Audience breadth: subfield vs broad readership.
- Whether the draft will go through a separate copy-edit pass (affects how aggressively to rewrite vs flag).

# Journal: Nature Communications (polishing)

## Read the shared facts first

Open `../../../../nature-shared/journal-formats/nat-comms.md` for the authoritative formatting facts: word limits, abstract rules, figure specs, reference style, mandatory statements, and common desk-rejection patterns.

The notes below are the **polishing action layer** on top of those facts.

## Audience

Open-access, broader than a subfield journal but more specialist-tolerant than Nature. The reader is typically an active researcher in an adjacent area.

## Polishing priorities

- Significance framing matters, but a less aggressive opening than Nature is acceptable.
- Word count is more forgiving than Nature, but **the ~5,000-word cap includes Methods**. If polishing brings the manuscript near the limit, surface a word-budget check rather than just compressing prose.
- Methods can stay more detailed in-line; do not strip reproducibility content just to hit length.
- Same em-dash and hedging defaults as Nature.
- Abstract is unstructured, 150 words, no citations. If the user's abstract has citations or is structured (Background/Methods/Results), restructure before sentence polishing.

## Length-aware polishing checks

When polishing a Nature Communications Article, run these word-budget checks before sentence-level work:

1. Estimate total word count **including Methods**. If over ~5,000, flag the gap and ask the user where to cut before polishing.
2. If display items > 10, flag for redistribution into Supplementary Information before deeper edits.
3. If references > ~60, flag for trimming or consolidation.
4. If the abstract is > 150 words or contains citations, restructure first.

These are diagnostic, not destructive — surface the problem in `Revision notes:` rather than silently cutting.

## Things the shared facts already cover

Do not restate facts that live in `nature-shared/journal-formats/nat-comms.md`. Reference them when the user asks. Examples Claude should respond by reading shared facts:

- "What's the reference style?" → cite the format example from shared
- "How many figures can I have?" → cite the 10-item cap from shared
- "What's the abstract limit?" → 150 words, unstructured, no citations

# Journal: Nature Machine Intelligence — polishing

## Read the shared contract first

Open
`../../../../nature-shared/journal-formats/nature-machine-intelligence.md` for
the exact, stage-aware NMI facts. This fragment controls polishing decisions,
not the submission numbers themselves.

For Results or Discussion, also open
`../../../../nature-shared/core/nature-results-discussion.md`. It is
corpus-derived Nature-style writing guidance, not an official NMI requirement.

For Introduction polishing, also open
`../../../../nature-shared/core/nature-introduction.md`. It is corpus-derived
Nature-style writing guidance, not an official NMI requirement.

For abstract polishing, also open
`../../../../nature-shared/core/nature-abstract.md`. It is corpus-derived
Nature-style writing guidance, not an official NMI requirement.

## Audience and stance

Polish for readers across AI, robotics, machine learning and the scientific or
societal application domain. Preserve technical precision while making the
question, evidence and consequence understandable outside one specialist
benchmark community.

## Structure before sentence polish

For an Article, check these before line editing:

1. Count Introduction + Results + Discussion against the 3,500-word main-text
   limit; exclude abstract, Methods, references and legends.
2. Keep the unreferenced abstract at or below 150 words.
3. Check that figures and tables total no more than six.
4. Keep the introduction unheaded; allow topical subheadings in Results and
   Methods but not in Discussion.
5. Preserve enough Methods detail for replication; NMI does not publish a
   fixed Methods word limit on the current pages.

Report an over-limit or wrong-structure manuscript under `Revision notes:`.
Do not silently delete evidence, uncertainty or reproducibility detail to make
the prose fit.

## Claim calibration

- Replace benchmark-only claims of general intelligence or deployment
  readiness with the evaluated scope.
- Separate gains due to architecture from data, supervision, compute or tuning
  differences.
- Name robustness and transfer boundaries when the prose currently hides them.
- Do not convert association, retrospective performance or simulation results
  into causal, clinical or societal effectiveness.
- Define specialist abbreviations and remove jargon that is not needed for
  technical accuracy.

## Results–Discussion calibration

- Preserve a claim-escalating Results sequence rather than flattening related
  subsections into parallel experiment reports.
- Keep a bounded local inference in Results when it directly resolves the
  current experiment; move cross-result synthesis and broader implications to
  Discussion.
- Keep conclusion-changing robustness and failure boundaries visible in the
  main text; move reassurance-only variants to SI.
- Permit a brief finding anchor in Discussion, but remove repeated evidence,
  effect sizes, and tests that merely re-demonstrate the Results claim.

## Introduction calibration

- Compress broad context until the specific unresolved problem appears early.
- Replace vague limitation language with the exact unknown, condition,
  mechanism, boundary, or literature tension.
- Reorganize citation catalogues into a known–unknown argument.
- Replace novelty adjectives with the research question and the design that can
  answer it.
- Check that each Introduction question maps to a Results answer and that the
  closing paragraph previews this route without replaying the paper.

## Abstract calibration

- Rebuild the abstract around one main discovery rather than a list of
  datasets, models, metrics, and completed analyses.
- Keep only enough method information to show why the design answers the gap.
- Retain no more than one or two decisive supporting claims or boundaries.
- Keep a number only when it defines, supports, or bounds the main discovery.
- Make the final sentence state the bounded conceptual or field-level payoff,
  not merely superior performance.

## Submission-sensitive checks

- Preserve explicit disclosure of a substantially extended conference paper.
- Preserve Data Availability and the separate Code availability section.
- Flag missing reviewer access or a missing Software Submission Checklist when
  new code is central to the conclusions.
- Do not apply a flagship Nature title, Methods or legend limit. NMI's current
  public pages state no current fixed number for those items. For figure
  legends, use the official 2018 NMI below-300-word instruction only as a
  historical advisory ceiling: count the complete legend, not each panel, aim
  for 150–250 English words, and obey any newer editor or submission-system
  instruction.

# Journal: Nature (and Nature subjournals)

## Audience

A broad, multi-disciplinary readership. A reader outside the immediate subfield must be able to grasp the claim, the gap, and the significance from the first paragraph and the abstract.

## Polishing priorities

- The opening sentence must signal significance for a non-specialist audience without overclaiming.
- Avoid jargon that does not appear in the headline of a typical Nature News piece. Define or replace.
- Abstract is unstructured prose for many Nature titles; check the target journal's current guideline if the user names one.
- Length discipline: Nature articles are unforgiving on word count. Prefer cuts to compressions.
- No em dashes in body prose. Use commas, parentheses, or shorter sentences.
- Hedging should be calibrated. Avoid both overclaim (`proves`, `definitive`) and timidity (`might possibly suggest`).

## Things to flag, not silently fix

- If the manuscript leans on a specialist mechanism the broad audience would not follow, flag it for the author rather than over-translate.
- If significance for a non-specialist is genuinely thin, surface that as a structural problem in `Revision notes:`.

# Language: English source

## Sentence rules

- In polished prose, aim for sentences in the `10-30` word range.
- Keep every sentence at `<= 30` words.
- Do not produce full sentences under `10` words unless the user explicitly asks for terse style or the item is a heading, label, or fixed technical expression.
- If any sentence exceeds `20` words, check whether it contains more than one main proposition.
- Split overloaded sentences rather than polishing them cosmetically.
- The last sentence of a paragraph often becomes the longest and weakest. Check it explicitly.
- Prefer one core subject-verb proposition per sentence.
- Do not use em dashes as prose punctuation in the polished version unless the user explicitly requests them. Rewrite with commas, parentheses, or shorter sentences instead. Use colons only when they add clear structural value.

## Paragraph rules

- Each paragraph should have one controlling idea followed by support.
- Supporting material may include data, comparison, explanation, consequence, literature, or limitation.
- If a new idea appears, start a new paragraph instead of stacking it onto the old one.
- Use thematic linking, not repetitive `This suggests ...` openings.

# Language: Chinese-to-English

When the source is Chinese or strongly Chinese-influenced English, do not translate clause-by-clause.

## Workflow

1. Extract the core propositions first. List them in plain English before drafting prose.
2. Reconstruct explicit logical links: contrast, cause, implication, limitation. Chinese academic prose often elides these connectives — restore them.
3. Verify terminology, causality, and hedging strength against the source.
4. Keep technical terms, gene/protein names, model names, dataset names, and statistical terms stable; do not "translate" them into rough paraphrases.
5. Apply the English sentence and paragraph rules from `language/en.md` only after the logic is rebuilt.

## Common Chinese-influenced patterns to fix

- Topic-comment chains rewritten as subject-verb sentences.
- Strings of short clauses joined by commas — split or add connectives.
- Vague generalizations (`many studies have shown`) — convert to specific citations or remove.
- Hedging asymmetry: Chinese drafts often understate; English Nature-style asks for precise hedging matched to evidence strength, neither over- nor under-claiming.
- Repetition of the topic noun where English would use a pronoun or omit it.

# Paper type: algorithmic or device

The argument proposes a procedure, tool, or system and must show that it performs reliably and advantageously.

## What the reader expects

- a clear problem formulation and the precise scope being claimed
- the proposed procedure or device, described to a level a peer could re-implement
- a fair comparison against credible baselines under matched conditions
- ablations or controls that isolate why the new approach works
- failure modes, runtime/cost characteristics, and applicability boundaries

## Polishing priorities

- Separate "what the system is" from "why it works" from "how well it works"; do not braid them.
- Performance claims must specify the dataset, metric, baseline, and conditions, not stand as bare numbers.
- Avoid marketing verbs (`leverages`, `enables`, `empowers`) unless they carry information.
- The Discussion should name the failure modes the experiments revealed, not only the wins.

# Paper type: hypothesis-based

The argument tries to establish or rule out a causal explanation.

## What the reader expects

- a clearly stated hypothesis, framed before the evidence
- a falsification path: what observations would refute it
- evidence presented in support, against, or as boundary conditions
- alternative explanations addressed, not ignored

## Polishing priorities

- Make the hypothesis statement explicit and locatable; do not bury it in the third paragraph of the Introduction.
- Distinguish supporting evidence from consistent-but-non-discriminating evidence.
- In the Discussion, address rival explanations before generalizing.
- Hedging should match the strength of the causal claim. Avoid overclaiming mechanism from correlational data.

# Paper type: methods

The reader of a methods paper asks:

- whether the method works
- whether it is reproducible
- whether it is better under a fair comparison

## Results section job

The Results section must show the advantages of the method over existing methods. Typical questions are:

- Is it more reliable?
- Is it faster?
- Does it require fewer resources?
- Is the comparison fair and reproducible?

## Methods section detail

The Methods section in a methods paper may need additional detail such as:

- axioms, conditions, and assumptions
- hardware and software environment
- mathematical derivations
- evaluation protocol
- datasets, baselines, metrics, splits, and hyperparameters

## Productive writing order

1. Methods
2. Results
3. Introduction
4. Conclusion
5. Discussion
6. Abstract

# Paper type: research

The reader of a research paper asks:

- why the phenomenon matters
- what was done
- what was found
- what it means

## Hourglass structure

Strong research papers mirror an hourglass:

- `Introduction`: open broadly, then narrow to the specific gap, question, hypothesis, methods, and study
- `Discussion/Conclusion`: widen again, connecting the findings back to the literature and explaining how the knowledge gap was filled

If a paragraph or section violates this architecture, rebuild it before polishing wording.

## Productive writing order

1. Results
2. Introduction and Conclusion
3. Title
4. Discussion
5. Materials and Methods
6. Authors
7. Abstract

Follow the logic of evidence and argument, not the raw order in which the user drafted sentences.

# Paper type: review

The reader asks: what is the state of the field, where is the disagreement, and what is the path forward.

## What the reader expects

- a clear scope statement (which sub-area, which time window, which inclusion criteria)
- a synthesis that organizes the literature by argument, not by paper
- explicit positioning of disagreements and gaps
- an outlook that names the most informative open questions

## Polishing priorities

- A review is not a survey list. Replace `Author A reported X. Author B reported Y.` with synthesis that groups claims by mechanism, method, or conclusion.
- Position the reviewer's own stance carefully: a review can take a view, but it must show its reasoning.
- Avoid generic transitions (`furthermore`, `additionally`). Use connectives that signal the logical relation (`in contrast`, `building on this`, `the remaining disagreement is`).
- The closing section should leave the reader with a usable map, not a summary of what was just read.

# Section: Abstract

The abstract is the manuscript's shortest evidence chain, not a compressed
Introduction:

`precise problem/gap -> answer-enabling design -> main discovery -> decisive support/boundary -> implication`

It should answer:

1. What question was addressed?
2. How was it addressed?
3. What was found?
4. Why should anyone care?

Some journals require a strict abstract format. Follow the journal if it conflicts with the generic pattern. (See the loaded `journal/*.md` fragment for journal-specific constraints.)

## Polishing priorities

- Cut sentences that summarize background that the title already implies.
- Make the gap and the contribution one short, locatable sentence each.
- Organize the abstract around one main claim and at most one or two decisive
  supporting claims or boundaries; do not inventory every Results subsection.
- Keep method detail only when it explains why the question is answerable.
- Keep a number only when it defines, supports, or materially bounds the main
  discovery; numeric reporting is not mandatory by itself.
- The last sentence should state significance, not repeat the result.

# Section: Conclusion

Use the three-part close:

1. restate the central contribution
2. summarize the key evidence or outcome
3. state the implication with a boundary

Do not introduce new data in the conclusion. Always run an overclaim check here.

## Overclaim check

- Does each claim trace back to evidence in this paper?
- Are mechanism words (`demonstrates`, `proves`, `establishes`) backed by the right study design?
- Is the scope of the implication narrower than or equal to the scope of the evidence?

# Section: Discussion

Load `../../../../nature-shared/core/discussion-argument-language.md` for the
complete Discussion function chain, modality ladder, limitation logic, and
sentence-function audit.

Discussion should answer:

- how the work fits within the broader field
- what has been added to understanding
- who should be credited for earlier work
- whether the findings support, complicate, or revise earlier results
- how the findings are interpreted
- when that interpretation may fail

Begin with a brief central-finding anchor when needed, then synthesize across
Results claims. Do not replay the comparison, effect size, statistical test,
and inference used to demonstrate each claim.

Short rule:

- `Results = the evidence chain that establishes and advances claims`
- `Discussion = what those claims mean together, how they relate to the field,
  and when the synthesis may fail`

## Sentence syntax

Discussion sentences usually interpret:

- `may reflect`
- `suggests that`
- `could indicate`
- `is likely due to`
- `may facilitate`

Hedging strength should match evidence strength. Do not promote a "consistent with" finding to "demonstrates" wording.

## Common failure modes

- Re-summarizing Results instead of interpreting them.
- Skipping rival explanations.
- Omitting boundaries: when does the interpretation stop holding?
- Expanding from a narrow result to a field-wide implication without an
  evidence bridge.
- Stacking hedges or using `must`, `cannot`, `demonstrates`, or causal verbs
  more strongly than the design permits.
- Naming a limitation without stating its consequence for a claim or scope.
- Offering generic future work that does not resolve an identified uncertainty.

# Section: Introduction

The Introduction should:

- tell the reader why the work matters
- explain what gap it fills
- explain why that gap matters
- state what is already known
- state what remains unresolved
- state what question the paper asks
- indicate how the study addresses it

Use a converging funnel:

`important problem -> specific phenomenon or difficulty -> what prior work establishes -> exact unknown -> research question or hypothesis -> study route`

Make literature serve the known–unknown transition rather than catalogue the
field. Let novelty emerge from the question and the design capable of answering
it, not from adjectives.

Do not summarize the Results section here. Do not summarize the Conclusion here.

## Common failure modes

- Opening paragraph reads as a textbook rather than a positioning move.
- The gap is implied but never explicitly named.
- The transition from "what is known" to "what this paper does" is missing.
- Methods are previewed in detail; keep that for Methods.
- The gap is defined as the absence of the author's method rather than a
  scientific unknown.
- Background paragraphs do not narrow toward a question answered in Results.
- The closing paragraph lists contributions but does not preview the study's
  question–answer route.

# Section: Materials and Methods

Methods should be specific, complete, transparent, and reproducible.

Another group should be able to determine:

- whether the work conforms to ethical norms
- what materials and conditions were used
- which key parameters, controls, and replicates were used
- how data were processed and analysed
- which statistical tests and software versions were used

It is acceptable to abbreviate by citing an earlier report only when that report truly contains the necessary detail.

## Forbidden vague phrases

Never leave wording such as:

- `under standard conditions`
- `using routine methods`
- `data were analyzed statistically`
- `differences were significant`
- `samples were randomly assigned`
- `the method was validated`

Replace them with the actual reproducible information.

# Section: Results

Results are the evidence chain that establishes and advances the paper's
scientific claims.

Load `../../../../nature-shared/core/nature-results-discussion.md` when the
target is Nature / Nature Portfolio. Rebuild the section around a discovery
loop, a core-capability validation envelope, or a capability ladder rather than
an experiment inventory.

Results writing should:

- load `../../../../nature-shared/core/main-text-discipline.md` and classify
  every result before deciding its location
- stay mainly in past tense
- report what was observed, under what conditions, and with what quantitative support
- use statistics correctly and sparingly
- keep core discovery and necessary support in the main text
- move non-central robustness, heterogeneity, provenance detail, alternative
  inference, and edge cases to SI unless they change the conclusion
- report the main descriptive quantity and primary inferential support once;
  avoid duplicating the full report in both Results and captions

End each coherent evidence unit with the bounded inference it supports. Keep
local interpretation when it directly answers the current experiment and the
supporting evidence is visible there; move cross-result synthesis, extensive
literature comparison, and broad implications to Discussion.

## Sentence syntax (Results vs Discussion)

Results sentences usually report:

- `was detected`
- `increased`
- `showed`
- `enabled`
- `achieved`

Use interpretive syntax (`may reflect`, `suggests that`, `is likely due to`)
only when the transition is intentional, evidence-bound, and calibrated.

## Common failure modes

- Adding an inline interpretation that is unsupported by the current evidence,
  or expanding a bounded inference into broad Discussion.
- Citing supplementary data when the result should stand in the main text.
- Treating every completed analysis as main-text material.
- Appending reviewer-defense sentences without checking what they replace.
- Explaining why a secondary analysis does not contradict the main result over
  several recursive sentences instead of moving the reconciliation to SI.
- Restating the same claim in a heading, transition, caption, Results paragraph,
  and closing sentence without adding a new function.
- Vague comparisons (`higher than control`) without effect size or test.
- Replaying an earlier baseline or control and drawing the same conclusion in a
  later subsection, instead of centring the new perturbation, falsification,
  stronger comparator, or boundary that creates the deeper inference.

After restructuring, apply the paragraph necessity test to every Results
paragraph and record whether text was retained, compressed, relocated, or
deleted. Do not move contradictory or conclusion-changing evidence out of the
main text.

# Section: Title

A strong title should:

- tell the reader what to expect
- avoid unnecessary technical language
- be easy to search
- be substantiated by data
- create curiosity without sacrificing credibility

Use `curiosity with credibility`, not empty cleverness. A hook is only acceptable if the claim remains fully defensible.

## Polishing approach

- If asked for alternatives, generate 3-5 candidates spanning declarative, question, and finding-led patterns; mark the most defensible.
- Strip jargon that the target journal's general audience would not recognize.
- Verify every quantitative claim in the title against the manuscript.