mm-writing

name	mm-writing
description	Stage 4 of the mathematical modeling pipeline. Generates LaTeX paper source from all workspace outputs, compiles to PDF. Supports multiple competition templates. Invoked by the math-modeling skill during Stage 4. Do not invoke directly.
version	0.4.4

Stage 4: Paper Generation (LaTeX → PDF)

Purpose

Generate a complete competition paper in LaTeX from all workspace outputs, then compile to PDF.

Pre-conditions (v0.4.4 硬性检查)

在执行任何论文生成步骤之前，你必须验证以下条件：

1. 读取 mm-workspace/pipeline_state.json
2. 验证 stage_3_5_review 状态为 "complete"
3. 验证 stage_3_5_passed 为 true
4. 验证 mm-workspace/03.5_review.json 文件物理存在（用 Read 工具尝试读取）

如果以上任一条件不满足 → STOP。报告缺少什么条件，建议用户先运行 Stage 3.5 全局审查。不要跳过此检查直接生成论文。

Input

• mm-workspace/01_analysis.json — problem analysis
• mm-workspace/02_modeling.json — modeling solution and tasks
• mm-workspace/03_task_{id}.json — all task outputs
• mm-workspace/03.5_review.json — global review results
• mm-workspace/charts/ — all figures
• mm-workspace/data/ — result data files

Competition Template System

Templates are stored in references/templates/. Each template is a directory containing:

• main.tex — main document file with placeholder markers
• refs.bib — BibTeX reference file (if applicable)
• Any style files (.cls, .sty, .bst)

Available Templates

Template ID	Competition	Language	Description
mcm	MCM/ICM	English	美赛官方格式
cumcm	CUMCM	Chinese	国赛官方格式
generic	Any	Configurable	通用格式

If the user's competition type is detected, use the corresponding template. Otherwise use generic.

Template extensibility: Users can add custom templates by creating a new directory under references/templates/ with a main.tex file.

Process

Step 1: Generate Abstract

Load references/abstract_guide.md and follow its structure guidelines.

Generate the abstract directly as a LaTeX section following the abstract_guide structure. The abstract MUST follow the guide's standards even if not saved as a separate JSON file.

Abstract 生成方式 (v0.4.4 简化):

• 直接生成 LaTeX 格式的 abstract 文本，嵌入到论文中
• 必须遵循 abstract_guide.md 定义的结构标准（问题重述、方法摘要、关键结果、结论）
• 保存到 mm-workspace/04_abstract.json 作为结构化记录（可选但推荐）：

  {
    "abstract_text": "LaTeX 格式的摘要全文",
    "keywords": ["关键词1", "关键词2"],
    "stage": "abstract_complete"
  }
  

Step 2: Load Template

Read the appropriate template from references/templates/{template_id}/main.tex. If no template exists for the detected competition, use references/templates/generic/main.tex.

Step 3: Generate Paper Sections

For each section of the paper, generate LaTeX content based on workspace outputs:

Abstract Section

• Use the abstract generated in Step 1
• Insert into the abstract placeholder

Problem Restatement

• Based on mm-workspace/raw_problem.txt
• Rephrase in own words, 1-2 paragraphs

Assumptions

• Extract from 02_modeling.json modeling_solution
• Format as numbered list with justifications

Notation / Symbol Table

• Extract all variables from task formulas
• Generate a LaTeX table: Symbol | Description | Unit

Model Establishment (per task)

For each task in dag_order:

• Read 03_task_{id}.json
• Extract: formulas, modeling_process
• Generate LaTeX section with:
  • Method description
  • Mathematical formulations (already in LaTeX)
  • Model assumptions specific to this sub-problem

Model Solution

For each task:

• Describe the solution approach
• Reference the code implementation
• Present key numerical results in tables/figures

Results and Analysis

For each task:

• Read result_interpretation and answer
• Present results with figures (reference charts)
• Discuss findings and implications

Sensitivity Analysis

• Read the sensitivity analysis task output (if exists)
• Present parameter perturbation results
• Discuss model robustness

Model Evaluation

• Strengths of the model
• Weaknesses and limitations
• Possible improvements

References

• Collect method references from HMML retrieval
• Format as BibTeX entries
• Generate bibliography

Step 4: Integrate Figures

For each figure in mm-workspace/charts/:

1. Copy or symlink to mm-workspace/05_paper/figures/
2. Generate LaTeX \includegraphics command
3. Add appropriate caption and label

Step 5: Assemble and Save

Write the complete paper to mm-workspace/05_paper/main.tex:

• Replace template placeholders with generated content
• Ensure all \ref{} and \label{} are consistent
• Ensure all figures are referenced
• Write BibTeX file to mm-workspace/05_paper/refs.bib (if applicable)

Step 5.5: Reference Richness Check

After assembling the paper, count the number of references:

• MCM/ICM papers: minimum 15 references recommended
• CUMCM papers: minimum 10 references recommended

If insufficient, supplement from these sources:

1. Standard references for methods in the HMML method index
2. Original papers for methods used (e.g., Ridge regression → Hoerl & Kennard 1970)
3. Survey papers for the data domain
4. Textbook references for modeling approaches

After supplementing, update refs.bib (if applicable) or the thebibliography environment.

Then commit: cd mm-workspace && git add -A && git commit -m "feat(s4): paper generated"

Step 6: Compile to PDF

cd mm-workspace/05_paper && pdflatex main.tex && bibtex main && pdflatex main.tex && pdflatex main.tex

If compilation fails:

1. Read the error log
2. Fix LaTeX syntax issues (common: unescaped special characters, missing packages)
3. Retry up to 3 times

Step 7: Present to User

Display:

• Paper structure overview (sections and page estimate)
• Abstract text
• All figures included
• PDF file path
• Any compilation warnings

Then ask: "论文已生成。是否需要修改？"

Output Structure

mm-workspace/05_paper/
├── main.tex          # 完整 LaTeX 源码
├── main.pdf          # 编译后的 PDF
├── main.aux          # LaTeX 辅助文件
├── refs.bib          # 参考文献（如适用）
├── figures/          # 论文图片
│   ├── task_1_fig1.png
│   └── ...
└── sections/         # 各章节源码（可选，用于调试）
    ├── abstract.tex
    ├── problem.tex
    └── ...