Imported from lix965996-art/MMM (resources/app/skills/comp-paper-zh/SKILL.md). Install upstream with npx skills add lix965996-art/MMM --skill comp-paper-zh. Copyright stays with the author.

Competition Paper Writing (Chinese)

Write a competition paper based on modeling results: $ARGUMENTS

Constants

• COMPETITION — stats = 统计建模, huazhong = 华中杯, wuyi = 五一杯, mathorcup = MathorCup, others = 数模竞赛 (cumcm/huawei/etc.)

• MAX_PAGES — Default 20. Body pages (chapter 1 through conclusion) must be ≥ MAX_PAGES.

• CUSTOM_REQUIREMENTS

Inputs

1. PROBLEM_ANALYSIS.md, MODELING_REPORT.md, RESULTS.md

2. figures/, code/

Load shared rules

cat _utils/writing_rules.md 2>/dev/null || cat skills/shared-scripts/writing_rules.md

<paper_structure>

Paper Structure by Competition Type

数模竞赛 (cumcm/huawei/mathorcup/huazhong/etc.)

Template: templates/cumcm/main.tex (国赛/华为杯), templates/mathorcup/main.tex (MathorCup), templates/huazhong/main.tex (华中杯), templates/wuyi/main.tex (五一杯)

⛔ MathorCup 必须使用 MathorCupmodeling.cls 文档类（模板文件夹已包含 cls）。使用 \bianhao{}、\tihao{}、\timu{} 设置队伍信息，\keyword{} 设置关键词。摘要用 \begin{abstract}...\end{abstract} 环境。参考文献用 \begin{thebibliography} 环境。

⛔ 华中杯必须使用 cumcmthesis 文档类（模板文件夹 huazhong/ 已包含 cls + 字体）。华中杯模板使用 \begin{abstract}...\keywords{}\end{abstract} 环境写摘要（不是手动排版），参考文献用 \begin{thebibliography} 环境（不是 \bibliography{}）。

摘要（~1页，含关键词）

1 问题重述

2 模型假设

3 符号说明

4 问题一的建模与求解（each sub-problem gets its own chapter）

5 问题二的建模与求解

6 问题三的建模与求解

7 灵敏度分析与模型检验

8 模型评价与推广

参考文献

附录 A：代码

统计建模 (stats)

Template: templates/stats/main.tex

Chapter structure is driven by research content, not fixed templates.

Award-winning stats modeling papers vary wildly in structure — some organize by model, some by analysis step, some by research question. There is no "standard structure". Claude must design chapters autonomously based on the actual content in TOPIC_PLAN.md / MODELING_REPORT.md.

Fixed skeleton (must keep)

表格清单

插图清单

摘要（中英文，分页）

一、绪论/前言（研究背景 + 文献综述/研究现状 + 研究目标/内容）

  ↓

  [Middle chapters: content-driven, Claude designs autonomously, typically 3-5 chapters]

  ↓

N、结论与建议（结论 + 建议/展望 + 创新与不足）

参考文献

致谢

附录（代码）

Middle chapter design guide

After reading TOPIC_PLAN.md, design middle chapters following these principles:

Principle 1: Chapter titles must be specific, not generic

• ✗ "四、模型构建" → ✓ "四、基于 CNN 的水质预测模型构建与评价"

• ✗ "五、实证分析" → ✓ "五、生育意愿的影响因素——集成学习模型"

Principle 2: Organize by research logic chain, not by textbook methodology

• If the research has multiple sub-problems/models, each model can be its own chapter

• If the research uses a single method with deep analysis, organize by analysis steps

Principle 3: Data and method chapters can be merged or separated

• Simple data (one dataset) → merge into "数据与方法"

• Complex data (multi-source, heavy preprocessing) → separate chapter "数据描述与预处理"

Award-winning paper structure examples (reference only, do not copy)

Example A — Classification + Path Analysis (fertility intention):

前言 → 模型构建 (introduce ensemble + Bayesian network) → 数据说明和预处理 → 探索性特征分析 → 生育意愿的影响因素 (ensemble results) → 生育意愿影响路径 (Bayesian network results) → 结论与建议

Example B — Mixed Modeling (data factors & economic growth):

研究背景+文献 → 研究思路和模型介绍 → 理论分析 → 模型构建 (production function + regression + ARIMA, each a section) → 模型应用 (GDP prediction) → 总结与建议 → 创新与不足

Example C — DEA Evaluation (economic sustainability):

绪论 → 文献综述 → 研究区域概况 → 评价指标体系构建 → 数据优化处理 (normalization + PCA) → DEA 模型建立及求解 → 结论及建议

Example D — Deep Learning Prediction (water quality CNN):

绪论 → 模型构建思路与创新 → 数据描述及预处理 → 主成分分析 → CNN 模型构建与评价 (with model comparison) → 结论与展望

Key observations:

• No award-winning paper uses the "baseline regression → robustness → heterogeneity" causal inference structure (unless the topic IS causal inference)

• Chapter count varies from 5-7, determined by content

• "Model introduction / theoretical basis" can come before or after the data chapter

• "Innovation & limitations" can be inside the conclusion or a standalone chapter

Chapter design checklist (self-check before writing)

• Does every chapter title contain specific research content (not generic "模型构建")?

• Does the chapter order follow the research logic chain (reader can follow naturally)?

• Do core analysis chapters (model results) occupy 40-50% of the paper?

• Is there a dedicated data description / exploratory analysis chapter (reviewers value data understanding)?

• Does the conclusion include "innovation" and "limitations" (reviewer bonus points)?

Use Chinese numbering (一、二、三...) with sub-sections (一)(二)(三). Do not use 1、2、3 or 1.1、1.2 format.

Fixed sections that must be kept: 表格清单, 插图清单, 中英文摘要, 绪论, 结论, 参考文献, 致谢.

</paper_structure>

Workflow

Step 0: Backup + resume check + upstream validation

⛔ 上游输出完整性检查（写论文前必做）：

echo "=== 上游输出完整性检查 ==="

UPSTREAM_OK=true



# 1. 核心文件是否存在

for f in PROBLEM_ANALYSIS.md MODELING_REPORT.md RESULTS.md; do

    if [ -f "$f" ]; then

        sz=$(wc -c < "$f")

        echo "✅ $f ($sz 字符)"

        [ "$sz" -lt 500 ] && echo "  ⚠ 文件过小，内容可能不完整"

    else

        echo "❌ $f 不存在！"

        UPSTREAM_OK=false

    fi

done



# 2. 子问题覆盖度：赛题分析 vs 建模报告 vs 代码结果

PROB_COUNT=$(grep -c '问题[一二三四五六七八九十]' PROBLEM_ANALYSIS.md 2>/dev/null || echo 0)

MODEL_COUNT=$(grep -c '问题[一二三四五六七八九十]' MODELING_REPORT.md 2>/dev/null || echo 0)

RESULT_FILES=$(ls figures/problem_*_results.json 2>/dev/null | wc -l)

echo "子问题数: 分析=$PROB_COUNT, 建模=$MODEL_COUNT, 代码结果=$RESULT_FILES"

[ "$MODEL_COUNT" -lt "$PROB_COUNT" ] && echo "⚠ 建模报告覆盖子问题数少于赛题分析"

[ "$RESULT_FILES" -lt "$PROB_COUNT" ] && echo "⚠ 代码结果文件数少于子问题数"



# 3. 图表文件是否存在

PDF_COUNT=$(ls figures/*.pdf 2>/dev/null | wc -l)

echo "PDF 图表: $PDF_COUNT 个"

[ "$PDF_COUNT" -eq 0 ] && echo "⚠ 没有 PDF 图表，论文将缺少图片"



# 4. all_results.json 是否存在

[ -f figures/all_results.json ] && echo "✅ all_results.json 存在" || echo "⚠ all_results.json 不存在，论文数值可能不准确"



# 5. latex_includes.tex 是否存在

[ -f figures/latex_includes.tex ] && echo "✅ latex_includes.tex 存在" || echo "⚠ latex_includes.tex 不存在，图表嵌入代码缺失"



echo "=== 上游检查完成 ==="

Back up existing paper/. Check for incomplete sections from previous runs:

echo "=== 断点续写检查 ==="

if [ -d "paper/sections" ]; then

    for f in paper/sections/*.tex; do

        [ -f "$f" ] || continue

        chars=$(wc -c < "$f")

        if [ "$chars" -lt 500 ]; then

            echo "⚠ 占位符: $(basename $f) ($chars 字符) — 需要续写"

        else

            echo "✅ 已完成: $(basename $f) ($chars 字符)"

        fi

    done

fi

Resume rules: only write placeholder sections (<500 chars or contains "待补充"/"placeholder"), skip completed ones (>2000 chars). Save each chapter immediately — do not accumulate in memory.

Step 1: Copy template (based on COMPETITION type)

mkdir -p paper/sections

# Select template based on competition type — copy entire folder (tex + cls + fonts)

TMPL_BASE="_templates"

[ -d "$TMPL_BASE" ] || TMPL_BASE="templates"



if [ "$COMPETITION" = "stats" ] || echo "$ARGUMENTS" | grep -qi "统计建模\|stats"; then

    echo "Using stats template"

    cp "$TMPL_BASE/stats/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "mathorcup\|MathorCup\|mathor" || grep -qi "mathorcup" CLAUDE.md 2>/dev/null; then

    echo "Using MathorCup template"

    cp "$TMPL_BASE/mathorcup/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "huazhong\|华中杯" || grep -qi "huazhong\|华中杯" CLAUDE.md 2>/dev/null; then

    echo "Using huazhong template"

    cp "$TMPL_BASE/huazhong/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "huawei\|华为杯" || grep -qi "huawei\|华为杯" CLAUDE.md 2>/dev/null; then

    echo "Using huawei template"

    cp "$TMPL_BASE/huawei/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "wuyi\|五一杯" || grep -qi "wuyi\|五一杯" CLAUDE.md 2>/dev/null; then

    echo "Using wuyi template"

    cp "$TMPL_BASE/wuyi/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "cumcm\|国赛" || grep -qi "cumcm\|国赛" CLAUDE.md 2>/dev/null; then

    echo "Using cumcm template"

    cp "$TMPL_BASE/cumcm/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "changsanjiao\|长三角" || grep -qi "changsanjiao\|长三角" CLAUDE.md 2>/dev/null; then

    echo "Using changsanjiao template"

    cp "$TMPL_BASE/changsanjiao/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "huashu\|华数杯" || grep -qi "huashu\|华数杯" CLAUDE.md 2>/dev/null; then

    echo "Using huashubei template"

    cp "$TMPL_BASE/huashubei/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "diangong\|电工杯" || grep -qi "diangong\|电工杯" CLAUDE.md 2>/dev/null; then

    echo "Using diangongbei template"

    cp "$TMPL_BASE/diangongbei/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "dongsansheng\|东三省\|辽宁" || grep -qi "dongsansheng\|东三省\|辽宁" CLAUDE.md 2>/dev/null; then

    echo "Using dongsansheng template"

    cp "$TMPL_BASE/dongsansheng/"* paper/ 2>/dev/null

elif echo "$ARGUMENTS" | grep -qi "shuwei\|数维杯" || grep -qi "shuwei\|数维杯" CLAUDE.md 2>/dev/null; then

    echo "Using shuweibei template"

    cp "$TMPL_BASE/shuweibei/"* paper/ 2>/dev/null

else

    echo "Using default template"

    cp "$TMPL_BASE/default/"* paper/ 2>/dev/null

fi

# Rename to main.tex if needed

[ -f paper/main.tex ] && echo "Template copied: $(wc -l < paper/main.tex) lines" || echo "ERROR: template not found!"

⛔ 模板复制后立即验证（必须通过才能继续）：

echo "=== 模板完整性验证 ==="

if [ ! -f paper/main.tex ]; then

    echo "❌ CRITICAL: paper/main.tex 不存在！模板复制失败"

else

    # 通用检查

    grep -q 'documentclass' paper/main.tex && echo "✅ documentclass 存在" || echo "❌ 缺少 documentclass"

    grep -q '\\input{sections/' paper/main.tex && echo "✅ sections input 存在" || echo "❌ 缺少 sections input"

    grep -q 'thebibliography\|bibliography{' paper/main.tex && echo "✅ 参考文献结构存在" || echo "❌ 缺少参考文献"

    grep -q 'appendices\|\\appendix' paper/main.tex && echo "✅ 附录结构存在" || echo "❌ 缺少附录"



    # ⛔ 模板指纹校验：对比 paper/main.tex 和模板原文，确认是复制的不是 Claude 自己写的

    TMPL_BASE="_templates"

    [ -d "$TMPL_BASE" ] || TMPL_BASE="templates"

    # 找到当前使用的模板原文

    TMPL_MAIN=""

    for d in "$TMPL_BASE"/*/; do

        [ -f "${d}main.tex" ] || continue

        # 用 documentclass 行匹配

        TMPL_CLS=$(grep 'documentclass' "${d}main.tex" 2>/dev/null | head -1)

        PAPER_CLS=$(grep 'documentclass' paper/main.tex 2>/dev/null | head -1)

        if [ "$TMPL_CLS" = "$PAPER_CLS" ]; then

            TMPL_MAIN="${d}main.tex"

            break

        fi

    done



    if [ -n "$TMPL_MAIN" ]; then

        echo "模板原文: $TMPL_MAIN"

        # 对比前 20 行（preamble 部分），如果差异超过 5 行说明被重写了

        DIFF_COUNT=$(diff <(head -20 "$TMPL_MAIN") <(head -20 paper/main.tex) 2>/dev/null | grep -c '^[<>]')

        if [ "$DIFF_COUNT" -gt 10 ]; then

            echo "❌ CRITICAL: paper/main.tex 的 preamble 和模板差异过大（$DIFF_COUNT 行不同）— 可能被重写了！"

            echo "⛔ 强制从模板重新复制..."

            cp paper/main.tex paper/main.tex.broken 2>/dev/null

            TMPL_DIR=$(dirname "$TMPL_MAIN")

            cp "$TMPL_DIR"/* paper/ 2>/dev/null

            echo "✅ 模板已重新复制（旧文件备份为 main.tex.broken）"

        else

            echo "✅ 模板指纹校验通过（preamble 差异 $DIFF_COUNT 行，在允许范围内）"

        fi

    else

        echo "⚠ 未找到匹配的模板原文，跳过指纹校验"

    fi

fi

Read the copied template to understand its structure before writing:

cat paper/main.tex

⛔ 你必须完整读取 main.tex 模板内容。后续写章节时，只修改 sections/*.tex 文件，不要重写 main.tex。

Use the template as-is. Only modify:

• Replace bracket placeholders ([论文标题], [中文摘要内容], etc.) with actual content

• Replace \input{sections/...} filenames if your section files have different names

• Fill in the abstract and keywords

⛔ CRITICAL TEMPLATE RULES (violation = broken PDF):

1. NEVER rewrite main.tex from scratch — the template has 100+ lines of carefully tuned preamble (fonts, margins, section numbering, citation format). Writing from scratch will break fonts, margins, and formatting.

2. NEVER replace \listoftables/\listoffigures with hand-written text lists — the template uses LaTeX auto-generated lists. If you write "表1.xxx\n表2.yyy" manually, the list won't update when tables change.

3. NEVER add a separate cover page for MathorCup — MathorCup 官方格式没有独立封面。队伍编号表格+标题+摘要都在第一页。不要自己加 \maketitle 或写一个大标题封面。直接用模板里的格式：表格 → 分隔线 → 标题 → 摘要。

4. NEVER change the cover page year/届数 format — replace [竞赛年份] with the actual year (e.g., 2026), [届数] with the actual number (e.g., 十二).

5. Verify after writing: diff paper/main.tex against the template — only bracket placeholders and \input lines should differ.

6. MathorCup 模板验证：检查 main.tex 使用 \documentclass{MathorCupmodeling}，\bianhao{}/\tihao{}/\timu{} 已填写。

7. 华中杯模板验证：检查 \documentclass[withoutpreface,bwprint]{cumcmthesis} 未被修改，摘要使用 \begin{abstract}...\keywords{}\end{abstract} 环境，参考文献使用 \begin{thebibliography}{99}...\end{thebibliography}，附录使用 \begin{appendices}...\end{appendices}，不要自己加 \usepackage{geometry}。

8. 五一杯模板验证：检查承诺书页完整保留，摘要使用手动排版（\noindent \textbf{关键词：} 在上，\noindent \textbf{摘\quad 要：} 在下，不用 \begin{abstract} 环境），参考文献使用 \begin{thebibliography}，附录第一节是文件列表表格。⛔ 五一杯额外检查：(a) 不要加 \usepackage{cite}（和 natbib 冲突会导致编译错误），(b) 不要重复加载 \usepackage{subcaption} 和 \usepackage{float}（cls 已包含），(c) 不要加 \maketitle（五一杯用手写承诺书页，不用 cls 的 maketitle），(d) 不要删除 withoutpreface 选项。

The cover page uses \fzxbsong (方正小标宋) and \fsgb (仿宋GB2312) fonts with \cline{2-2} underlines in a tabular. These fonts have fallback definitions — if the .ttf files are not installed, they fall back to \heiti and \fangsong.

When replacing cover page placeholders, ONLY change the text inside [...] brackets. Do NOT touch \cline{2-2}, \\, &, or any LaTeX commands:

% Template has (和 cover station.tex 格式一致):

参赛学校：&[学校名称]\\\cline{2-2}

% Replace ONLY the bracket text:

参赛学校：&某某大学\\\cline{2-2}

% NEVER remove \cline{2-2} (creates underline) or & (column separator)

Do NOT modify:

• \documentclass line (font size, paper size)

• \usepackage[...]{geometry} (page margins)

• \ctexset{section=...} (chapter numbering format)

• \pagestyle / \fancyhf (header/footer)

• \usepackage[...]{natbib} or \usepackage[...]{gbt7714} (citation format)

• Cover page structure (封面布局)

• \listoftables / \listoffigures (表格与插图清单 — 自动生成，不要手写)

• Anything before \begin{document} (preamble)

• The cover page, abstract, TOC, and bibliography sections in main.tex — only replace placeholder text within them

Write all chapter content in separate paper/sections/*.tex files. The main.tex \input{sections/...} lines load them automatically. Do not paste chapter content directly into main.tex.

If you need to add packages, add them after the existing \usepackage block, before \begin{document}.

Do not write main.tex from scratch — the template handles fonts, spacing, margins, citation format, headers/footers.

Note: the stats template uses natbib with [numbers, square, super] format (superscript [1]), while the cumcm template uses gbt7714 with superscript format. Use whichever the template provides. In body text, use \cite{key} — the super option automatically renders it as superscript.

Do not redefine LaTeX built-in commands (\sin, \cos, \tanh, \log, \exp, \max, \min, etc.) in math_commands.tex.

Step 2: Abstracts

⛔ CRITICAL: Do NOT write the abstract now. Skip this step entirely. Write a placeholder [摘要待正文完成后填写] in the abstract section. The abstract MUST be written LAST (after Step 5) because it needs specific numerical results from all chapters. Writing it first = making up numbers.

Come back to fill the abstract in Step 5 (final check), after all body chapters are complete. At that point, read RESULTS.md and all section .tex files to extract the actual numbers.

<abstract_format>

The template uses manual typesetting for abstracts. Do not use two \begin{abstract} environments — ctexart shows "摘要" as title for both.

⛔ 华中杯例外：华中杯模板基于 cumcmthesis 文档类，使用 \begin{abstract}...\keywords{}\end{abstract} 环境。不要改成手动排版格式。正确写法：

\begin{abstract}

[中文摘要内容，400-600字]

\keywords{[关键词1]\quad [关键词2]\quad [关键词3]}

\end{abstract}

⛔ 五一杯说明：五一杯模板有承诺书页（第一页）和封面页（第二页）。五一杯摘要使用手动排版（不用 \begin{abstract} 环境）：关键词在上（\noindent \textbf{关键词：}...），摘要在下（\noindent \textbf{摘\quad 要：}）。这是五一杯官方格式，跟国赛/华中杯不同。承诺书页不要删除。

Correct format for cumcm/stats templates (already in template):

% === 中文摘要 ===

\begin{center}

{\zihao{3}\heiti 摘\hspace{2em}要}

\end{center}

\zihao{-4}\songti

[中文摘要内容，600-800字，写满一整页]

\noindent\textbf{关键词：}...



% === 英文摘要（新页）===

\newpage

\begin{center}

{\zihao{3}\bfseries Abstract}

\end{center}

[English abstract, 400-600 words, faithful translation]

\noindent\textbf{Keywords:} ...

数模竞赛摘要: 400-600 字, every sub-problem must have specific numerical results. ⛔ 必须按问题分段：第1段背景概述，第2-4段分别针对问题一/二/三（方法+数值结果），第5段模型评价。每段用 LaTeX 空行分隔。五一杯虽然是手动排版摘要，分段规则同样适用。

统计建模摘要: 500-700 字, aim to fill most of one page but leave 3-4 lines margin at bottom — overflowing onto a second page looks worse than being slightly short. Content chain: 研究背景与意义 → 现有方法不足 → 本文方法 → 数据来源与处理 → 关键数值结果 → 应用价值与政策建议. English abstract: 350-500 words, same structure and all numerical results, also fit on one page.

</abstract_format>

Step 3: Figure inventory + mandatory embedding plan

Before writing any chapter, build a complete inventory of available figures:

echo "=== Available PDF figures ==="

ls -la figures/*.pdf 2>/dev/null || echo "No PDF figures found"

echo ""

echo "=== Available LaTeX table files ==="

ls -la figures/TABLE_*.tex 2>/dev/null || echo "No TABLE files found"

echo ""

echo "=== latex_includes.tex content (figure→PDF mapping) ==="

cat figures/latex_includes.tex 2>/dev/null || echo "No latex_includes.tex"

echo ""

echo "=== TikZ architecture diagrams ==="

[ -s figures/tikz_architecture_examples.tex ] && echo "YES — must embed" && grep -c 'begin{figure}' figures/tikz_architecture_examples.tex || echo "No TikZ diagrams"

⛔ MANDATORY: Build a FIGURE EMBEDDING PLAN before writing any chapter:

FIGURE EMBEDDING PLAN:

1. fig_desc_stats.pdf → 第三章 描述性统计 → caption: "图X 核心变量描述性统计分布"

2. fig_model_comparison.pdf → 第四章 模型对比 → caption: "图X 模型性能对比雷达图"

3. TABLE_regression.tex → 第四章 回归分析 → caption: "表X 回归分析结果"

4. tikz_roadmap (from tikz_architecture_examples.tex) → 问题重述章节末尾 (1_restatement.tex) → caption: "图X 研究技术路线图"

Rules:

• ⛔ 图表 caption 必须是中文（中文论文）。如果 latex_includes.tex 里的 caption 是英文，嵌入时必须翻译成中文

• ⛔ 必须使用 latex_includes.tex 里的 figure 代码块，不要自己写 \includegraphics。直接复制整个 \begin{figure}...\end{figure} 块，只改 caption 为中文

• ⛔ TikZ 图必须嵌入：从 figures/tikz_architecture_examples.tex 复制每个 \begin{figure}...\end{figure} 块到对应章节

• ⛔ 图片路径必须是 ../figures/xxx.pdf（因为 sections/ 在 paper/ 下面）

• Only embed figures whose PDF files actually exist — do not create figure environments for PDFs that don't exist

⛔⛔⛔ DrawIO 图嵌入规则（最容易被遗漏，必须逐条检查）：

DrawIO 图（技术路线图、求解流程图、Pipeline 图等）的 \begin{figure}...\end{figure} 代码块在 latex_includes.tex 的末尾部分（由 paper-figure-drawio 步骤追加）。你必须把它们嵌入到正确的章节：

| DrawIO 图类型 | 嵌入位置 | 章节文件 |

|--------------|---------|---------|

| 技术路线图 (fig_roadmap) | 问题重述章节末尾 | 1_restatement.tex 或 1_introduction.tex |

| 子问题求解流程图 (fig_flow_q1/q2/q3) | 对应子问题的"模型建立"小节开头，先写 2-3 句引导文字（概述本问题的求解思路和主要步骤），再放流程图 | 5_problem1.tex、6_problem2.tex 等 |

| 数据处理 Pipeline (fig_pipeline) | 数据预处理章节 | 数据处理相关章节 |

| 模型架构图 (tikz_*) | 对应模型章节的"模型建立"小节 | 模型构建相关章节 |

写完所有章节后，必须运行以下检查确认 DrawIO/TikZ 图全部嵌入：

echo "=== DrawIO/TikZ 嵌入检查 ==="

for pdf in figures/fig_roadmap.pdf figures/fig_flow_*.pdf figures/fig_pipeline*.pdf figures/fig_framework*.pdf; do

    [ -f "$pdf" ] || continue

    bn=$(basename "$pdf")

    if grep -rq "$bn" paper/sections/*.tex paper/main.tex 2>/dev/null; then

        echo "✅ $bn 已嵌入"

    else

        echo "❌ $bn 未嵌入 — 必须立即修复！"

    fi

done

# TikZ 图检查

if [ -s figures/tikz_architecture_examples.tex ]; then

    for lbl in $(grep -oh '\\label{[^}]*}' figures/tikz_architecture_examples.tex 2>/dev/null); do

        if grep -rq "$lbl" paper/sections/*.tex 2>/dev/null; then

            echo "✅ TikZ $lbl 已嵌入"

        else

            echo "❌ TikZ $lbl 未嵌入 — 必须立即修复！"

        fi

    done

fi

如果有任何 ❌，必须立即修复后再继续写下一章。不要等到最后才修复。

Also scan figures/*.tex for all \begin{figure} / \begin{table} blocks with their \label{}. After writing, verify all are embedded:

grep -oh '\\label{[^}]*}' figures/*.tex 2>/dev/null | sort -u > _tmp/all_fig_labels.txt

grep -oh '\\label{[^}]*}' paper/sections/*.tex paper/main.tex 2>/dev/null | sort -u > _tmp/embedded_labels.txt

comm -23 _tmp/all_fig_labels.txt _tmp/embedded_labels.txt  # should be empty

TikZ figures from figures/tikz_architecture_examples.tex: paste each \begin{figure}...\end{figure} code block directly into the corresponding section (not via \input). Map by label/caption:

• 技术路线图/研究框架图/问题关系图 → 问题重述章节末尾（1_restatement.tex），不要放到后面的子问题章节

• 各子问题求解流程图 → 对应子问题章节开头（4_problem1.tex、5_problem2.tex、6_problem3.tex 等）

• 模型架构图 → 对应模型章节

Step 3.5: 文献预检索（写正文之前必须完成）

⛔ 在写任何 \cite{} 之前，必须先建立已验证的文献池。

PYTHON=$(command -v python3 2>/dev/null || command -v python 2>/dev/null)

mkdir -p _tmp

# 根据论文主题搜索真实论文，建立引用池

# 示例（根据实际选题调整）：

#   $PYTHON "$SCHOLAR_SCRIPT" bibtex "你的核心方法关键词" --max 5

#   $PYTHON "$SCHOLAR_SCRIPT" bibtex "你的研究领域关键词" --max 5

搜索后创建 _tmp/_verified_refs.txt，写正文时只引用池子里的论文。需要新引用时先搜索验证再加入池子。

兜底：如果 scholar_fetch.py 搜不到或 match_label="low"，用 WebSearch 在 Google Scholar / Semantic Scholar 网站上搜索，手动核实标题+作者+年份后再加入池子。

Step 4: Write each chapter

⛔ CRITICAL: ALL numerical results in the paper MUST come from figures/*.json or RESULTS.md. Before writing any result chapter, run:

cat figures/all_results.json

cat RESULTS.md

Copy the exact numbers from JSON into the LaTeX text. Do NOT round differently, do NOT estimate, do NOT make up numbers that "look reasonable". If a number is not in the JSON, it cannot appear in the paper.

⛔ 数模竞赛（cumcm/huawei/mathorcup）必须严格按模板的章节顺序写：

1_restatement.tex  — 问题重述（用自己的话重述，不是抄题目）

2_assumptions.tex  — 模型假设（5-7 条，每条有内容+合理性说明）

3_symbols.tex      — 符号说明（⛔ 见下方格式规则）

4_problem1.tex     — 问题一的建模与求解

5_problem2.tex     — 问题二的建模与求解

6_problem3.tex     — 问题三的建模与求解

7_sensitivity.tex  — 灵敏度分析与模型检验

8_evaluation.tex   — 模型评价与推广

A_code.tex         — 附录：代码

⛔ 模型假设必须在符号说明之前。 不要合并成一个章节，不要调换顺序。文件名必须和模板 \input{sections/...} 一致。

⛔ 模型假设数量控制：4-5 条，不要超过 6 条。每条假设 1-2 句话（假设内容 + 合理性说明），不要写成长段落。假设太多说明问题没简化好。

⛔ 符号说明表格控制：15-20 个变量以内。只列正文中实际使用的核心变量，不要把所有中间变量都列进去。

⛔ 分页规则（所有章节通用）：

• 不要在 section 文件内部加 \newpage、\clearpage 或 \nopagebreak — 让 LaTeX 自动分页，手动干预容易产生空白页

• main.tex 里只在摘要后和目录后用分页，正文章节之间不要加

• compile_utils.sh 会自动移除 section 文件里的 \newpage/\clearpage/\nopagebreak

• 例外：符号说明文件（compile_utils.sh 会自动转成 longtable 支持跨页）

⛔ 符号说明表格格式（必须用 longtable，不要用 table+tabular）：

• 符号说明必须用 longtable 环境，不要用 \begin{table}...\begin{tabular}

• longtable 天然支持跨页，标题永远在表格开头，不会出现标题和表格分离的问题

• 不需要 \centering（longtable 自带居中）

• 不需要引导文字（"本文所用主要符号..."），标题直接紧跟表格

• compile_utils.sh 会自动把 table+tabular 转成 longtable（兜底），但最好从源头写对

• 正确写法：

\section{符号说明}

\begin{longtable}{clc}

\caption{主要符号说明}\label{tab:symbols} \\

\toprule

符号 & 含义 & 单位 \\

\midrule

\endfirsthead

\toprule

符号 & 含义 & 单位 \\

\midrule

\endhead

$N$ & 总数量 & 个 \\

$x_i$ & 第$i$个变量 & --- \\

... \\

\bottomrule

\end{longtable}

符号 & 含义 & 单位 \

\midrule

$T(t)$ & 温度 & $^\circ$C \

$v$ & 速度 & cm/min \

\bottomrule

\end{tabular}

\end{table}

- 符号控制在 15-20 个以内，表格控制在半页以内



**⛔ 写每个章节前，先读 MODELING_REPORT.md 和 RESULTS.md 对应部分的内容。** 不要凭记忆写——数值结果、公式推导、算法步骤都必须从这些文档中提取。



Read RESULTS.md for exact numbers — ensure paper numbers match computation results.



**⛔ 图文数值一致性规则：** 描述图表内容时（如"从图X可以看出，模型A的RMSE为0.023"），数值必须从 `figures/*.json` 或 `RESULTS.md` 中读取，不要凭记忆编写。写完后用 `bash _utils/writing_check.sh paper/` 检查一致性。



**⛔ 长表格处理规则（>12 行的数据表格）：**



正文中**禁止**直接放超过 12 行的表格（调度方案、完整数据列表、逐步迭代结果等）。处理方式：



1. **正文：只放缩略版**（前 3 行 + 后 3 行 + 中间 `$\vdots$` 省略），底部注明"完整结果见附录 X"

2. **附录：放完整表格**（用 longtable 环境，允许跨页）



示例（正文缩略版）：

```latex

\begin{table}[H]

\centering

\caption{问题一调度方案（部分）}

\begin{tabular}{cccc}

\toprule

任务 & 设备 & 开始时间 & 结束时间 \\

\midrule

1 & A & 0 & 5 \\

2 & B & 2 & 8 \\

3 & A & 5 & 12 \\

\multicolumn{4}{c}{$\vdots$} \\

28 & C & 45 & 52 \\

29 & A & 48 & 55 \\

30 & B & 50 & 58 \\

\bottomrule

\end{tabular}

\label{tab:q1_schedule_short}

\footnotesize 注：完整调度方案（30 条记录）见附录 B。

\end{table}

判断标准： 如果 figures/TABLE_*.tex 文件超过 12 行数据行，必须自动生成缩略版 + 附录完整版。不要把 100+ 行的 longtable 直接嵌入正文章节。

Follow the interleaving, embedding, and LaTeX rules from _utils/writing_rules.md.

⛔ 图文并茂硬规则（每个章节都必须遵守）：

• 所有 \begin{figure} 必须用 [H]，不要用 [htbp]

• 每张图/表后面必须有 ≥5 行分析文字（数值解读+对比+结论），然后才能放下一张图

• 绝对禁止两张图连续出现中间没有分析段落

• 图片用 \includegraphics[width=0.85\textwidth,height=0.38\textheight,keepaspectratio]

• ⛔ 禁止使用 subfigure/subcaption 并排图片。 每张图必须独占一个 \begin{figure}[H]...\end{figure} 环境，宽度 ≥ 0.85\textwidth。不要用 \begin{subfigure}{0.48\textwidth} 把两张图缩成半页宽——这会导致图片太小看不清。如果确实需要对比两张图，分成两个独立的 figure 环境，中间加 2-3 句过渡分析文字

• ⛔ 图片宽度下限： width 参数不得小于 0.8\textwidth。任何 width=0.5\textwidth、width=0.48\textwidth、width=0.45\textwidth 都是错误的，必须改为 0.85\textwidth 或 0.9\textwidth

⛔ 超长表格处理规则（>15 行的结果表格必须遵守）：

如果某个结果表格超过 15 行数据（如调度方案、路径规划、逐日预测值等），不要把完整表格放在正文里——会占好几页，挤压正文空间。正确做法：

1. 正文放摘要表：只展示前 5 行 + 后 3 行 + 汇总统计（均值/最优/总计），caption 标注"（部分，完整结果见附录表 X）"

2. 附录放完整表：在 sections/A_code.tex（或新建 sections/A_tables.tex）里放完整表格

% === 正文中的摘要表 ===

\begin{table}[H]

\centering

\caption{问题一调度方案（部分，完整结果见附录表\ref{tab:full_schedule}）}

\begin{tabular}{cccc}

\toprule

车间 & 设备 & 开始时间 & 结束时间 \\

\midrule

1 & A-1 & 0 & 1200 \\

1 & A-2 & 1200 & 2400 \\

\multicolumn{4}{c}{$\vdots$} \\

3 & C-4 & 8400 & 9600 \\

3 & C-5 & 9600 & 10800 \\

\midrule

\multicolumn{2}{c}{总计} & \multicolumn{2}{c}{Makespan = 10800s} \\

\bottomrule

\end{tabular}

\end{table}



% === 附录中的完整表 ===

% 在 A_code.tex 或 A_tables.tex 中：

\section{完整结果表格}

\begin{table}[H]

\centering

\caption{问题一完整调度方案}\label{tab:full_schedule}

\begin{longtable}{cccc}

% ... 完整数据 ...

\end{longtable}

\end{table}

判断标准： 写表格前先数数据行数。≤15 行直接放正文，>15 行用摘要+附录方案。compile_utils.sh 也会自动检测并截断超长表格，但最好在写的时候就处理好。

After each chapter, check character count:

chars=$(wc -c < "paper/sections/当前章节.tex")

echo "当前章节: $chars 字符 (~$(echo "scale=1; $chars/900" | bc) 页)"

# Chinese LaTeX ≈ 800-1000 chars per page

# If chapter expected 5 pages but only 2000 chars (~2.5 pages), expand immediately

<exemplar_depth>

Writing depth reference

国赛一等奖 (25-30 pages, 3 sub-problems):

• 问题重述+分析 (2-3p): restate in own words, extract core problems — not copying the problem statement

• 模型假设+符号说明 (2p): 5-7 assumptions (each with content + justification, not too many not too few), one complete symbol table with 15-25 variables (符号/含义/单位 three columns, do not split into multiple tables)

• Each sub-problem (5-7p): model formulation (2p, with derivation and physical meaning) + solution method (1.5p, with algorithm steps) + results with table+figure (1p) + analysis (0.5-1.5p, numerical meaning + comparison with expectations + reasoning)

• 灵敏度分析 (2-3p): ≥2 key parameters, each with variation curve plot + analysis paragraph

• 模型评价 (2p): 3-4 strengths + 2-3 weaknesses + generalization directions. Pure text discussion, no figures or tables needed

华为杯一等奖 (40-50 pages): deeper derivations, more thorough analysis per sub-problem (8-10p each), 灵敏度分析 (4-5p), 模型评价 (3p, pure text, no figures) with comparison to other methods

统计建模获奖 (35-40 pages) — chapter structure is content-driven, page allocation reference:

• 绪论/前言 (4-8p): research background + literature review (grouped by 3-4 themes, 3-5 papers per theme with detailed discussion) + research objectives

• 数据描述与预处理 (6-8p): data source + variable description table + descriptive statistics table + exploratory analysis (distribution / trend / cross-tabulation plots)

• 模型/方法 chapters (6-10p): theoretical basis + formula derivation + parameter settings + implementation details (adjust by actual number of models)

• Core results analysis (10-16p): the most important part of the paper — every result needs 2-3 paragraphs of interpretation, not just "as shown in Table X"

• 结论与建议 (3-5p): conclusions + recommendations + innovation points + limitations and future work

Common traits of award-winning papers:

• Solid exploratory analysis (cross-tabulation, group comparisons, rich visualization)

• Core analysis chapters occupy 40-50% of total pages, every numerical result has deep interpretation

• Specific chapter titles ("基于集成学习的生育意愿影响因素分析" not "模型构建")

• Include "innovation & limitations" discussion (reviewer bonus points)

| Type | Pages | Characters | References |

|------|-------|-----------|------------|

| 数模国赛 (30p limit) | 25-30 | 18000-25000 | ≥10 |

| 华为杯 | 40-50 | 30000-40000 | ≥15 |

| MathorCup | 35-40 | 25000-35000 | ≥10 |

| 华中杯 | 25-30 | 18000-25000 | ≥10 |

| 五一杯 | 25-30 | 18000-25000 | ≥10 |

| 统计建模 | 35-40 | 25000-35000 | ≥20 |

</exemplar_depth>

<figure_usage_principles>

Figure/table usage by competition type

统计建模: Figures first, tables second. Figure/table selection is driven by the actual analysis methods used — regression uses forest plots + regression tables, prediction uses prediction comparison plots + accuracy tables, classification uses confusion matrices + ROC, evaluation uses radar charts + ranking tables. Every analysis step should have a corresponding figure or table.

数模竞赛: "字不如表，表不如图". Every sub-problem must have independent result display (table + figure). Comprehensive comparison figures are additional supplements. Reviewers value information density and aesthetics.

Do not force figures where they are not needed (pure literature review, theoretical derivation). Claude decides figure count and placement based on content needs.

</figure_usage_principles>

<stats_figure_placement>

Stats modeling paper figure placement rules

Figure/table selection is driven by research content, not fixed type templates. Below are figures mapped to analysis methods:

Data description stage (almost all papers need these):

• Descriptive statistics table (required), data distribution plots / boxplots, correlation heatmap, time series trend plots

Model results stage (select based on actual methods used):

• Regression: regression result table + coefficient forest plot

• Prediction: prediction comparison plot (actual vs predicted) + model accuracy table (MAE/RMSE/R²) + error distribution plot

• Classification: confusion matrix + ROC curve + model comparison table (Accuracy/F1/AUC)

• Clustering: cluster scatter plot / t-SNE + silhouette score table

• Evaluation: comprehensive score ranking table + radar chart

• Causal inference: baseline regression table + robustness check table + heterogeneity comparison plot

Model interpretation / diagnostics stage (select as needed):

• Feature importance / SHAP plot, residual diagnostics, sensitivity analysis plot, ablation table

Principle: Core result figures/tables go in the body text. Appendix only for code and very long auxiliary tables.

</stats_figure_placement>

<chapter_writing_points>

Chapter writing points (universal, adapt to research logic)

⛔ 写作风格铁律（所有章节都必须遵守）：

• 禁止在正文中使用 \begin{itemize} 或 \begin{enumerate} 列表。 黑点/编号列表是最明显的 AI 写作痕迹。必须用连贯的段落叙述。

  • 需要列举时用行内编号："（1）...（2）...（3）..."或"首先...其次...最后..."

  • 例外：模型假设列表（可用编号）、附录

• 每段至少 3-5 句话。 不要写只有 1-2 句的短段落。

• 连续段落不能以相同句式开头。 如连续三段都以"本文..."开头，必须改。

• 图表是论据不是主语。 段落不能以"图X展示了"、"如图X所示"、"由图X可知"开头。图表引用用括号旁注 （图X） 融入论证链条：先论点 → 图表作旁证 → 推论。详见 _utils/writing_rules.md 的"图表是论据"规则。

• 禁止元叙述和内部指令泄露。 正文中不能出现"参赛者"、"参赛队伍"、"RESULTS.md"、"figures/*.json"、"CLAUDE.md"等内部文件名或工作流术语。用"本文"代替"我们团队"，论文是独立学术文档。

绪论/前言 (all papers):

• Research background (why this problem matters) → Literature review / research status (what others did, what gaps remain) → Research objectives / content / contributions → Paper structure overview

• Literature review organized by theme (≥15 citations), not chronologically listed

数据与预处理 (almost all papers need this):

• Data source → Sample description (time range / sample size / variable count) → Variable definition / coding table → Missing value / outlier handling → Exploratory analysis (distribution plots / trend plots / cross-tabulation)

• Exploratory analysis is a key capability demonstration for reviewers — do not skip

模型/方法 chapters (organize by actual research content):

• Each model/method: theoretical basis (1-2 paragraphs) → mathematical formulas → parameter setting rationale → implementation details

• Multiple models: can introduce all models first then show results (Example B style), or give each model its own chapter with results (Example A style)

结果分析 chapters (core, should occupy 40-50% of paper):

• Every result must have: numerical presentation (table/figure) → interpretation (2-3 paragraphs, not just "as shown in Table X") → comparison with expectations / other methods → reasoning

• Multi-model comparison: horizontal comparison table + rationale for selecting the best model

结论与建议:

• Main conclusions (echo research objectives) → Policy / application recommendations (specific and actionable) → Innovation points → Limitations and future work

• "Innovation & limitations" can be inside the conclusion or a standalone chapter (Example B style)

</chapter_writing_points>

Expansion strategies when content is thin (not padding — substantive content):

• Formula listed without derivation → add step-by-step derivation with physical meaning explanation

• Result with only "如表所示" → add 2-3 paragraphs of interpretation (numerical meaning, comparison with expectations, reasoning, comparison with other methods)

• Assumptions as bare list → add justification for each assumption

• Algorithm as pseudocode only → add explanation of key steps, complexity analysis, convergence discussion

• Literature review only lists papers → add method summary for each and connection to this work

Step 5: References

Follow the <references_workflow> in _utils/writing_rules.md.

Stats papers: ≥20 references. The stats template uses natbib with [numbers, square, super] — use \cite{key} in body text (auto-superscript) + \bibliography{references} + \bibliographystyle{plainnat}. The cumcm template uses gbt7714 — only need \bibliography{references}. The huazhong template uses \begin{thebibliography}{99}...\end{thebibliography} — manually add \bibitem entries (do not use \bibliography{}). The wuyi template uses gbt7714-numerical — use \bibliography{references} (same as cumcm).

⛔ 使用 scholar_fetch.py 工具获取所有参考文献的 BibTeX。禁止凭记忆编造 BibTeX。

⛔ 引用编号必须按正文出现顺序排列（1, 2, 3, 4...），不能跳着来。 使用 \begin{thebibliography} 的模板：\bibitem 的排列顺序必须跟正文中 \cite{} 首次出现的顺序一致。使用 \bibliography{references} 的模板：BibTeX 会自动按引用顺序编号（gbt7714-numerical / plainnat 都支持）。写完所有章节后，检查引用编号是否连续递增，如果不是，调整 \bibitem 顺序或 .bib 文件中的条目顺序。

⛔ 引用写法规则：写正文时，citation key 必须包含描述性关键词，格式为 作者姓_年份_主题关键词。

例如：\cite{wang_2023_supply_chain_resilience} 而不是 \cite{wang2023supply}。

不确定作者/年份时用 TODO__ 前缀：\cite{TODO__digital_economy_spatial_spillover}。

# Step 5a: 收集所有引用 key

grep -roh '\\cite[tp]*{[^}]*}' paper/sections/*.tex paper/main.tex 2>/dev/null \

  | grep -oP '\{[^}]+\}' | tr -d '{}' | tr ',' '\n' | sed 's/^ *//;s/ *$//' | sort -u > _tmp/_cited_keys.txt

echo "引用 key 数量: $(wc -l < _tmp/_cited_keys.txt)"



# Step 5b: 逐个搜索并获取 BibTeX（用描述性关键词搜索）

PYTHON=$(command -v python3 2>/dev/null || command -v python 2>/dev/null)

while IFS= read -r key; do

    query=$(echo "$key" | sed 's/^TODO__//; s/_/ /g')

    echo "--- 获取: $key (搜索: $query) ---"

    $PYTHON "$SCHOLAR_SCRIPT" bibtex "$query" --max 3

    sleep 0.5

done < _tmp/_cited_keys.txt

处理每个搜索结果：

1. 检查 match_label："good" → 直接使用。"partial" → 核实标题。"low" → 很可能搜错了，换关键词或用 WebSearch。

2. match_score < 0.3 说明大概率不是目标论文，不要盲目使用。

3. 将 .tex 中的 citation key 替换为 BibTeX 条目中的实际 key。

4. bibtex_source=auto 的条目加 % [VERIFY]。match_label="low" 的加 % [LOW_MATCH]。

Step 5.5: De-AI polish

See <de_ai_polish> in _utils/writing_rules.md.

Step 5.6: Write abstract NOW (after all chapters are complete)

⛔ MANDATORY: NOW write the abstract. Read RESULTS.md and all section .tex files. Extract the actual numerical results from each sub-problem. The abstract must contain ONLY numbers that appear in the body text — do NOT invent or round differently.

⛔ 统计建模必须写中英文两个摘要：先写中文摘要（500-700字），然后将中文摘要忠实翻译为英文摘要（350-500 words），所有数值结果、方法名称、结论必须一一对应。数模竞赛（国赛/五一杯/MathorCup/华中杯等）只写中文摘要。

For math modeling competitions: each sub-problem must have its specific result in the abstract (e.g., "问题一采用XX算法，最优解为YY，空间利用率达ZZ%")。

Read _utils/writing_rules.md for abstract format rules (分段、首行缩进、长度).

Step 6: Final verification

bash _utils/writing_check.sh paper/ 2>/dev/null || bash skills/shared-scripts/writing_check.sh paper/

Also check:

echo "=== 各章节字符数 ==="

total=0

for f in paper/sections/*.tex; do

    chars=$(wc -c < "$f")

    total=$((total + chars))

    echo "  $(basename $f): $chars 字符 (~$(echo "scale=1; $chars/900" | bc) 页)"

done

echo "  总计: $total 字符 (~$(echo "scale=1; $total/900" | bc) 页)"

echo "  目标: ≥ MAX_PAGES × 800 = $(echo "MAX_PAGES * 800" | bc) 字符"

• Total chars ≥ MAX_PAGES × 800 (expand thinnest chapters if not)

• Any sub-problem chapter <3500 chars (~4 pages) needs expansion

• All figures/.pdf and TABLE_.tex referenced in sections

• No template bracket placeholders remaining ([论文标题], [中文摘要内容], etc.)

⛔ Constraint consistency check (MUST do before finishing):

Read PROBLEM_ANALYSIS.md (or the original problem statement in user_data/) and check every numerical result in the paper against the problem's constraints:

=== 论文-题目约束一致性检查 ===

1. 读取题目中的所有约束条件（容量、预算、时间窗、数量限制等）

2. 逐个检查论文正文中的结果是否满足这些约束

3. 例如：题目说"车辆载重上限 6000kg"，论文写"装载 7344kg" → 矛盾，必须修正

4. 例如：题目说"30 个省份"，论文分析了 28 个 → 不完整，必须补充

5. 摘要中的数字是否与正文一致？

6. 不同章节引用同一个结果时数字是否一致？（如问题一的最优解在摘要、正文、结论中出现 3 次，必须完全相同）

如果发现矛盾，修改论文正文（不是修改约束）。如果是代码结果本身违反约束，说明代码有 bug，需要回到 comp-code 修复。

⛔ Page count pre-check (MUST pass before finishing — do NOT leave this to compile step):

echo "=== 页数预检 ==="

total_chars=0

for f in paper/sections/*.tex; do

    [ -f "$f" ] || continue

    chars=$(wc -c < "$f")

    total_chars=$((total_chars + chars))

done

est_pages=$((total_chars / 900))

echo "总字符: $total_chars, 估算页数: ~$est_pages 页, 目标: ≥ MAX_PAGES 页"

if [ "$est_pages" -lt "$((MAX_PAGES * 80 / 100))" ] 2>/dev/null; then

    echo "⛔ CRITICAL: 页数严重不足 ($est_pages < 80% of MAX_PAGES)"

    echo "必须扩充最薄的章节后再结束"

fi

If estimated pages < 80% of MAX_PAGES, you MUST expand the thinnest 2-3 chapters before finishing. Read MODELING_REPORT.md and RESULTS.md for additional content to add (more derivation, more result analysis, more parameter discussion).

Figure embedding verification (must pass before finishing):

echo "=== 图表嵌入检查 ==="

missing=0

for pdf in figures/*.pdf; do

    [ -f "$pdf" ] || continue

    bn=$(basename "$pdf")

    if ! grep -rq "$bn" paper/sections/*.tex paper/main.tex 2>/dev/null; then

        echo "MISSING: $bn 未嵌入任何章节"

        missing=$((missing + 1))

    fi

done

for fig_tex in figures/*.tex; do

    [ -f "$fig_tex" ] || continue

    for lbl in $(grep -oh '\\label{[^}]*}' "$fig_tex" 2>/dev/null); do

        if ! grep -rq "$lbl" paper/sections/*.tex paper/main.tex 2>/dev/null; then

            echo "MISSING: $lbl (from $(basename $fig_tex)) 未嵌入"

            missing=$((missing + 1))

        fi

    done

done

echo "缺失: $missing"

If any figures are missing, go back and embed them into the appropriate sections before finishing. ⛔ Do NOT proceed to Step 7 until missing = 0. Repeat the check after each fix.

⛔ 模板完整性自检（写完所有章节后必须检查 main.tex 没有被破坏）：

echo "=== main.tex 模板完整性检查 ==="

TMPL_OK=0

TMPL_FAIL=0



# 通用检查（所有模板）

grep -q 'documentclass' paper/main.tex && { echo "✅ documentclass"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 缺少 documentclass"; TMPL_FAIL=$((TMPL_FAIL+1)); }

grep -q '\\input{sections/' paper/main.tex && { echo "✅ sections input"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 缺少 sections input"; TMPL_FAIL=$((TMPL_FAIL+1)); }

grep -q 'thebibliography\|bibliography{' paper/main.tex && { echo "✅ 参考文献"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 缺少参考文献"; TMPL_FAIL=$((TMPL_FAIL+1)); }

grep -q 'appendices\|\\\\appendix' paper/main.tex && { echo "✅ 附录"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 缺少附录"; TMPL_FAIL=$((TMPL_FAIL+1)); }

grep -q 'superscript\|\\@cite\|setcitestyle.*super' paper/main.tex && { echo "✅ 上标引用"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 缺少上标引用定义"; TMPL_FAIL=$((TMPL_FAIL+1)); }



# 五一杯特有检查

if grep -qi 'wuyi\|五一杯' CLAUDE.md 2>/dev/null; then

    grep -q '承诺书' paper/main.tex && { echo "✅ 五一杯承诺书页"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 五一杯缺少承诺书页"; TMPL_FAIL=$((TMPL_FAIL+1)); }

    grep -q 'image2' paper/main.tex && { echo "✅ 五一杯封面logo"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 五一杯缺少封面logo"; TMPL_FAIL=$((TMPL_FAIL+1)); }

    grep -q '关键词' paper/main.tex && { echo "✅ 五一杯关键词"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 五一杯缺少关键词"; TMPL_FAIL=$((TMPL_FAIL+1)); }

fi



# MathorCup 特有检查

if grep -qi 'mathorcup' CLAUDE.md 2>/dev/null; then

    grep -q 'MathorCupmodeling' paper/main.tex && { echo "✅ MathorCup cls"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ MathorCup 未使用正确 cls"; TMPL_FAIL=$((TMPL_FAIL+1)); }

    grep -q '\\bianhao\|\\tihao\|\\timu' paper/main.tex && { echo "✅ MathorCup 队伍信息"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ MathorCup 缺少队伍信息"; TMPL_FAIL=$((TMPL_FAIL+1)); }

fi



# 华中杯特有检查

if grep -qi 'huazhong\|华中杯' CLAUDE.md 2>/dev/null; then

    grep -q 'cumcmthesis' paper/main.tex && { echo "✅ 华中杯 cls"; TMPL_OK=$((TMPL_OK+1)); } || { echo "❌ 华中杯未使用 cumcmthesis"; TMPL_FAIL=$((TMPL_FAIL+1)); }

fi



echo ""

echo "模板检查: $TMPL_OK 通过, $TMPL_FAIL 失败"

[ "$TMPL_FAIL" -gt 0 ] && echo "⛔ 必须修复所有失败项！可能是 main.tex 被意外重写了，请从模板重新复制并只替换占位符。"

Step 7: Output

数模竞赛: sections/ numbered by sub-problem (4_problem1.tex, 5_problem2.tex...)

统计建模: sections/ by academic structure (1_introduction.tex, 2_data_method.tex...)

Key Rules

• Use templates from templates/, do not write main.tex from scratch

• Chinese/English abstracts: manual typesetting, not two \begin{abstract} environments

• Citation format: use whichever the template provides (stats: natbib [1], cumcm: gbt7714 superscript, huazhong: thebibliography manual entries, wuyi: gbt7714-numerical)

• No \hypersetup{colorlinks=true} — conflicts with hidelinks

• Body pages ≥ MAX_PAGES (can exceed, must not fall short)

• No team info — use placeholders

• Tables: three-line style (booktabs)

• Primary output: paper/ directory, temp files: _tmp/

• ⛔ 本步骤只写论文 .tex 文件，不要重新生成图表 PDF、不要修改 code/*.py、不要重新运行分析代码。 图表和数据已由前序步骤（paper-figure / comp-code）生成完毕，直接引用即可

• Large files: Bash heredoc