6verity

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

验证和验收（Typst / LaTeX）

Verification and Acceptance (Typst / LaTeX)

本 skill 是完整工作流的最后一关。它不重新建模、不生成新结果、不代替写作阶段重写论文；它负责发现硬错误、修复可直接修复的问题，并输出

reports/VERIFY_REPORT.md

。

This skill is the final gate of the complete workflow. It does not remodel the problem, generate new results, or rewrite the paper in place of the writing stage; it detects critical errors, fixes issues that can be fixed directly, and outputs

reports/VERIFY_REPORT.md

数学建模规范参考

Mathematical Modeling Standards Reference

如需领域判断，读取

../_references/math_modeling_norms.md

中的“论文验收与一致性”小节。CUMCM 2026 模式还必须读取

../_references/CUMCM_2026_RULES.md

；具体目录、入口文件、结果文件和图表目录由当前项目结构决定。

When domain-specific judgment is needed, read the “Paper Acceptance and Consistency” section in

../_references/math_modeling_norms.md

. CUMCM 2026 mode must also read

../_references/CUMCM_2026_RULES.md

; the specific directories, entry files, result files, and figure directories are determined by the current project structure.

阶段边界

Stage Boundaries

本阶段负责：结构验收、文本质量门禁、图表引用检查、结果一致性检查、Typst/LaTeX 编译检查、PDF 视觉检查、提交清单。
本阶段不负责：重新设计模型、重新跑大规模实验、重新组织整篇论文。
发现硬错误时，优先做不改变模型含义的小范围修复；如果需要回到前序阶段，写入
```
reports/VERIFY_REPORT.md
```
，标记未通过并给出结构化路由。

This stage is responsible for: structural acceptance, text quality gates, figure and table reference checks, result consistency checks, Typst/LaTeX compilation checks, PDF visual inspection, and the submission checklist.
This stage is not responsible for: redesigning the model, rerunning large-scale experiments, or reorganizing the entire paper.
When a critical error is found, prioritize small-scale fixes that do not change the meaning of the model; if it is necessary to return to an earlier stage, record the issue in
```
reports/VERIFY_REPORT.md
```
, mark it as failed, and provide structured routing.

CUMCM 2026 提交门禁

CUMCM 2026 Submission Gates

识别到

competition_mode: CUMCM_2026

时，下列任一硬门禁失败都必须令

SUBMISSION_READY = false

；普通

PASS

不得覆盖它：

PDF 第 1 页为摘要专用页，含标题、摘要、关键词；摘要原则上不超过 1 页；摘要页页码为 1，页脚中部连续编号。
不存在目录；A4；四边距至少 2.5 cm；正文不超过 30 页；论文文件不超过 20 MB。
摘要、正文、附录和支撑材料无姓名、学校、赛区及其他身份信息。
论文含官方“AI工具使用声明”且位于参考文献之前；若使用 AI，有
```
AI工具使用详情.pdf
```
，详情覆盖工具名称/版本或模型、用途与阶段、主要提示方式与过程、采纳/人工修改/人工核验。
```
AI_USAGE_LOG.md
```
、
```
HUMAN_REVIEW_LOG.md
```
、论文声明和 AI 详情一致；H1–H4 均为真实人工确认。
附录有支撑材料文件列表和全部完整可运行源码；有统一运行入口、环境清单，核心结果可复现且与论文一致。
支撑材料 ZIP/RAR 存在且不超过 20 MB，内容、文件名和论文附录清单一致，无承诺书、编号页或身份信息。
正式竞赛模式的
```
SOURCE_LOG.md
```
不包含从交流平台取得的当前赛题答案、思路、代码或论文。

CUMCM 专项的最终文件扫描、AI 详情和打包由

7cumcm-submit

完成；本阶段仍需报告上述科学与格式门禁，不能把它们全部推迟到下一阶段。

When

competition_mode: CUMCM_2026

is detected, failure of any mandatory gate below must set

SUBMISSION_READY = false

; an ordinary

PASS

must not override it:

Page 1 of the PDF must be a dedicated abstract page containing the title, abstract, and keywords. The abstract should generally not exceed one page. The abstract page must be numbered as page 1, with continuous numbering centered in the footer.
There must be no table of contents; the document must use A4 paper, with margins of at least 2.5 cm on all sides; the main text must not exceed 30 pages; and the paper file must not exceed 20 MB.
The abstract, main text, appendices, and supporting materials must not contain names, schools, competition regions, or other identifying information.
The paper must contain the official “AI Tool Use Statement” before the references. If AI was used,
```
AI工具使用详情.pdf
```
must be provided, covering tool names, versions or models, purposes and stages, primary prompting methods and processes, adoption, manual modifications, and manual verification.
```
AI_USAGE_LOG.md
```
,
```
HUMAN_REVIEW_LOG.md
```
, the paper declaration, and the AI details must be consistent; H1–H4 must all represent genuine human confirmations.
The appendices must contain a list of supporting material files and all complete, runnable source code; there must be a unified execution entry point and environment manifest, and the core results must be reproducible and consistent with the paper.
The supporting-material ZIP/RAR archive must exist, must not exceed 20 MB, and its contents, filenames, and the appendix list in the paper must be consistent. It must not contain pledge forms, numbered pages, or identifying information.
In formal competition mode,
```
SOURCE_LOG.md
```
must not contain current competition problem answers, ideas, code, or papers obtained from communication platforms.

The CUMCM-specific final file scan, AI details, and packaging are handled by

7cumcm-submit

; this stage must still report the scientific and formatting gates above and must not defer all of them to the next stage.

输入

Inputs

由模型先根据当前工作区判断项目布局，再把实际路径传给检查脚本。常见输入包括但不限于：

论文入口文件：
```
main.typ
```
（Typst）或
```
main.tex
```
（LaTeX）。
正文章节目录或若干正文文件（
```
.typ
```
或
```
.tex
```
）。
参考文献文件（
```
references.typ
```
或
```
references.tex
```
）。
前序阶段的分析、建模、结果、图示报告。
图表目录
可复现代码目录。
编译后的 PDF，或可由入口文件编译得到的输出 PDF。

不要假设论文目录一定叫

paper/

，也不要假设结果文件一定在项目根。若项目使用不同命名，按实际结构传参并在

reports/VERIFY_REPORT.md

中说明。

The model should first determine the project layout from the current workspace, then pass the actual paths to the checking script. Common inputs include, but are not limited to:

Paper entry file:
```
main.typ
```
(Typst) or
```
main.tex
```
(LaTeX).
Main-text section directory or several main-text files (
```
.typ
```
or
```
.tex
```
).
References file (
```
references.typ
```
or
```
references.tex
```
).
Analysis, modeling, result, and visualization reports from previous stages.
Figure and table directory.
Reproducible code directory.
Compiled PDF, or an output PDF that can be compiled from the entry file.

Do not assume that the paper directory is always named

paper/

, or that result files are always located in the project root. If the project uses different names, pass the actual paths and explain them in

reports/VERIFY_REPORT.md

工作流程

Workflow

Step 1: 运行文本质量门禁

Step 1: Run the Text Quality Gate

优先运行本 skill 的脚本。脚本按入口文件扩展名自动选择检查逻辑（

.typ

→ Typst 检查，

.tex

→ LaTeX 检查）：

bash

set -o pipefail
mkdir -p _tmp
SCRIPT_PATH="<按当前 skill 实际位置确定>/scripts/writing_check.sh"
bash "$SCRIPT_PATH" \
  --paper-dir "$PAPER_DIR" \
  --main "$MAIN_FILE" \
  --sections-dir "$SECTIONS_DIR" \
  --references "$REFERENCES_FILE" \
  --figures-dir "$FIGURES_DIR" \
  --results-file "$RESULTS_FILE" \
  --problem-analysis "$PROBLEM_ANALYSIS_FILE" \
  --all-results "$ALL_RESULTS_FILE" \
  | tee _tmp/writing_check.log

CUMCM 2026 模式在上述参数中再加入

--cumcm-2026

，启用无目录、页码、AI 声明顺序及附录要求的源文件硬门禁。

如果本 skill 被复制到其他目录，使用实际脚本路径。可以先运行

bash <script> --help

查看参数。不要把脚本路径、论文目录或文件名写死在验收逻辑中。

脚本只扫描文本，不生成论文，也不编译 PDF。它的

FAIL

属于硬错误，必须修复后重跑。

Prioritize running this skill's script. The script automatically selects the checking logic based on the entry file extension (

.typ

→ Typst checks,

.tex

→ LaTeX checks):

bash

set -o pipefail
mkdir -p _tmp
SCRIPT_PATH="<按当前 skill 实际位置确定>/scripts/writing_check.sh"
bash "$SCRIPT_PATH" \
  --paper-dir "$PAPER_DIR" \
  --main "$MAIN_FILE" \
  --sections-dir "$SECTIONS_DIR" \
  --references "$REFERENCES_FILE" \
  --figures-dir "$FIGURES_DIR" \
  --results-file "$RESULTS_FILE" \
  --problem-analysis "$PROBLEM_ANALYSIS_FILE" \
  --all-results "$ALL_RESULTS_FILE" \
  | tee _tmp/writing_check.log

In CUMCM 2026 mode, add

--cumcm-2026

to the arguments above to enable source-level mandatory gates for the absence of a table of contents, pagination, AI statement ordering, and appendix requirements.

If this skill has been copied to another directory, use the actual script path. You can run

bash <script> --help

to view the arguments. Do not hard-code the script path, paper directory, or filenames into the acceptance logic.

The script only scans text; it does not generate the paper or compile the PDF. A

FAIL

from the script is a critical error and must be fixed before rerunning it.

Step 2: 章节数量和标题顺序

Step 2: Section Count and Heading Order

Typst 引擎检查：

入口
```
.typ
```
文件中
```
#include("...")
```
的数量是否与实际正文结构匹配。
include 顺序是否符合文件名前缀顺序，例如
```
1_...
```
,
```
2_...
```
,
```
3_...
```
。
每个 section 是否有明确一级标题（
```
= 标题
```
，等号后有空格）。
标题顺序是否符合所选论文类型。

LaTeX 引擎检查：

入口
```
.tex
```
文件中
```
\input{...}
```
或
```
\include{...}
```
的数量是否与实际正文结构匹配。
章节顺序是否符合文件名前缀顺序。
每个 section 是否有
```
\section{}
```
或对应级别标题。

通用检查（两种引擎）：

章节文件是否缺失、重复引用、未被引用。
如果题目不是三问，不强行要求三段问题章节；按
```
ANALYSIS_MODELING_REPORT.md
```
的子问题数量核对。

Typst engine checks:

Whether the number of
```
#include("...")
```
calls in the entry
```
.typ
```
file matches the actual main-text structure.
Whether the include order follows the filename prefix order, such as
```
1_...
```
,
```
2_...
```
,
```
3_...
```
.
Whether every section has an explicit first-level heading (
```
= Title
```
, with a space after the equals sign).
Whether the heading order conforms to the selected paper type.

LaTeX engine checks:

Whether the number of
```
\input{...}
```
or
```
\include{...}
```
calls in the entry
```
.tex
```
file matches the actual main-text structure.
Whether the section order follows the filename prefix order.
Whether every section has a
```
\section{}
```
or a corresponding level heading.

General checks for both engines:

Whether section files are missing, referenced more than once, or left unreferenced.
If the problem does not contain three questions, do not force three problem-analysis sections; verify the count against the number of subproblems in
```
ANALYSIS_MODELING_REPORT.md
```
.

Step 3: 图表和章节匹配

Step 3: Figure/Table and Section Matching

Typst 引擎检查：

图表目录中的 PDF 是否在正文中被引用。
```
#figure(image(...), caption: [...])
```
的图片是否真实存在。图片路径必须相对于
```
.typ
```
文件。
数据图是否放在对应结果/分析章节，非数据流程图是否放在方法/总体思路章节。

LaTeX 引擎检查：

```
\includegraphics{}
```
引用的图片文件是否真实存在。路径相对于
```
.tex
```
文件。
```
\caption{}
```
是否存在。
数据图是否放在对应结果/分析章节。

通用检查（两种引擎）：

连续图表之间是否有足够解释文字。
caption 是否过长、过泛或与图意不一致。
图表编号、正文引用和章节语义是否一致。

不要生成

*_typst_includes.typ

或

*_latex_includes.tex

；图表必须直接嵌在对应 section 中。

Typst engine checks:

Whether PDFs in the figure directory are referenced in the main text.
Whether images used by
```
#figure(image(...), caption: [...])
```
actually exist. Image paths must be relative to the
```
.typ
```
file.
Whether data figures are placed in the corresponding result or analysis sections, and non-data flowcharts are placed in the methodology or overall-approach sections.

LaTeX engine checks:

Whether image files referenced by
```
\includegraphics{}
```
actually exist. Paths must be relative to the
```
.tex
```
file.
Whether
```
\caption{}
```
is present.
Whether data figures are placed in the corresponding result or analysis sections.

General checks for both engines:

Whether there is sufficient explanatory text between consecutive figures or tables.
Whether captions are overly long, too vague, or inconsistent with the figure or table.
Whether figure and table numbering, in-text references, and section semantics are consistent.

Do not generate

*_typst_includes.typ

*_latex_includes.tex

; figures and tables must be embedded directly in their corresponding sections.

Step 4: 写作质量和泄露检查

Step 4: Writing Quality and Leakage Checks

检查并修复：

TODO

、

PLACEHOLDER

、

待补充

、

待续写

、

示例数据

等占位符。

论文正文出现内部工作流文件名、临时目录名、代码目录名或结果 JSON 路径。
过多列表式写作（Typst 中大量
```
#list
```
、
```
enum
```
，LaTeX 中大量
```
\begin{itemize}
```
、
```
\begin{enumerate}
```
）。
段落反复以"如图""由图""图 X 展示了"开头。
图表后没有解释、公式后没有变量含义、结论只报数不解释。

Check and fix:

Placeholders such as

TODO

PLACEHOLDER

待补充

待续写

, and

示例数据

Internal workflow filenames, temporary directory names, code directory names, or result JSON paths appearing in the main text.
Excessive list-based writing, such as many
```
#list
```
or
```
enum
```
elements in Typst, or many
```
\begin{itemize}
```
or
```
\begin{enumerate}
```
environments in LaTeX.
Paragraphs repeatedly beginning with phrases such as “As shown in the figure,” “From the figure,” or “Figure X shows.”
Missing explanations after figures or tables, missing explanations of variable meanings after formulas, and conclusions that only report numbers without interpreting them.

Step 5: 数值和结果一致性

Step 5: Numerical and Result Consistency

检查：

论文中的关键数值必须来自当前工作流声明的结果记录或结果 JSON。
目标函数值、误差指标、排名、权重、阈值、灵敏度结果不得与结果记录冲突。
如果存在汇总结果 JSON，抽取关键指标并确认论文正文中有对应结果。
公式中的符号应在符号说明或正文首次出现处解释。

发现数值冲突时，不要自行发明新结果；应回到结果记录或代码输出修正论文。

Check:

Key numerical values in the paper must come from the result records or result JSON declared by the current workflow.
Objective function values, error metrics, rankings, weights, thresholds, and sensitivity results must not conflict with the result records.
If a summary result JSON file exists, extract the key metrics and confirm that the corresponding results appear in the main text.
Symbols in formulas should be explained in the symbol description or where they first appear in the main text.

When a numerical conflict is found, do not invent new results; return to the result records or code output and correct the paper accordingly.

Step 6: 引用和模板规范

Step 6: Citations and Template Standards

检查：

参考文献文件是否存在，或模板是否采用了其他真实参考文献机制。
正文引用标记（Typst 的
```
@label
```
/
```
#super
```
，LaTeX 的
```
\cite{}
```
）是否能对应到真实参考文献。
中文论文 caption、表题、摘要语言保持中文；英文论文保持英文。
选定的模板入口是否保留所选比赛模板的必要封面、摘要、编号、页眉页脚或提交格式。
不要把模板结构误删成普通空白文档。

Check:

Whether the references file exists, or whether the template uses another genuine reference mechanism.
Whether in-text citation markers, such as Typst's
```
@label
```
/
```
#super
```
or LaTeX's
```
\cite{}
```
, correspond to real references.
Captions, table titles, and the abstract language must remain Chinese for Chinese papers and English for English papers.
Whether the selected template entry file retains the required cover, abstract, numbering, header and footer, or submission format for the selected competition template.
Do not accidentally delete the template structure and turn the document into an ordinary blank document.

Step 7: 编译

Step 7: Compilation

Typst 编译：

bash

command -v typst >/dev/null 2>&1 && typst compile "$MAIN_FILE" "$OUTPUT_PDF"

LaTeX 编译：

bash

command -v xelatex >/dev/null 2>&1 && xelatex -interaction=nonstopmode "$MAIN_FILE" && xelatex -interaction=nonstopmode "$MAIN_FILE"

xelatex 需跑两遍解决交叉引用。CUMCM 模式不得因编译流程而生成目录。

编译失败必须修复语法、路径、图片引用或模板问题后重跑。编译通过后确认输出 PDF 非空。

Typst compilation:

bash

command -v typst >/dev/null 2>&1 && typst compile "$MAIN_FILE" "$OUTPUT_PDF"

LaTeX compilation:

bash

command -v xelatex >/dev/null 2>&1 && xelatex -interaction=nonstopmode "$MAIN_FILE" && xelatex -interaction=nonstopmode "$MAIN_FILE"

xelatex must be run twice to resolve cross-references. CUMCM mode must not generate a table of contents as part of the compilation process.

Compilation failures must be fixed by correcting syntax, paths, image references, or template issues, and then rerun. After successful compilation, confirm that the output PDF is not empty.

Step 8: PDF 视觉检查

Step 8: PDF Visual Inspection

如果模型有视觉能力，必须把编译后的 PDF 每页导出为 PNG 并逐页查看。这个步骤用于发现纯文本扫描和编译器无法发现的版式错误。

优先使用系统已有工具导出页面 PNG；不要为了视觉检查引入沉重依赖。可选命令示例：

bash

mkdir -p _tmp/pdf-pages
if command -v pdftoppm >/dev/null 2>&1; then
  pdftoppm -png -r 160 "$OUTPUT_PDF" _tmp/pdf-pages/page
elif command -v mutool >/dev/null 2>&1; then
  mutool draw -r 160 -o _tmp/pdf-pages/page-%03d.png "$OUTPUT_PDF"
elif command -v magick >/dev/null 2>&1; then
  magick -density 160 "$OUTPUT_PDF" _tmp/pdf-pages/page-%03d.png
else
  echo "No PDF rasterizer found; record visual check as not run."
fi

导出后逐页检查：

页面是否空白、缺页、页数异常或页面尺寸异常。
标题、摘要、正文、页眉页脚、页码是否被裁切或位置明显错误。
表格是否超出页边距，单元格文字是否重叠、溢出、被截断。
图片、图题、表题、公式、编号是否与正文重叠。
公式是否越界，长公式是否压到页边距或下一段文字。
列表、段落、脚注、参考文献是否出现异常大空白、重叠或孤立残行。
中文/英文/数学符号字体是否明显缺字、乱码或 fallback 异常。
封面、摘要页、附录等模板关键页面是否保留比赛要求的视觉结构；CUMCM 若出现目录页直接判定失败。

如果是模板转换或已有参考 PDF 的项目，还应将不同引擎的 PDF 都逐页导出 PNG，按页对比版式差异；页数或页面尺寸不一致必须记录为硬错误或明确说明原因。

如果模型没有视觉能力，必须在

reports/VERIFY_REPORT.md

中明确写出“未执行视觉检查”的原因，并至少完成 PDF 非空、页数、页面尺寸等可程序化检查。

If the model has visual capabilities, it must export every page of the compiled PDF to PNG and inspect each page. This step is intended to detect layout errors that cannot be found through text-only scanning or by the compiler.

Prefer rasterization tools already available on the system; do not introduce heavy dependencies solely for visual inspection. Example commands:

bash

mkdir -p _tmp/pdf-pages
if command -v pdftoppm >/dev/null 2>&1; then
  pdftoppm -png -r 160 "$OUTPUT_PDF" _tmp/pdf-pages/page
elif command -v mutool >/dev/null 2>&1; then
  mutool draw -r 160 -o _tmp/pdf-pages/page-%03d.png "$OUTPUT_PDF"
elif command -v magick >/dev/null 2>&1; then
  magick -density 160 "$OUTPUT_PDF" _tmp/pdf-pages/page-%03d.png
else
  echo "No PDF rasterizer found; record visual check as not run."
fi

After exporting, inspect every page for:

Blank pages, missing pages, abnormal page counts, or abnormal page dimensions.
Clipped titles, abstracts, main text, headers, footers, or page numbers, or elements placed in clearly incorrect positions.
Tables extending beyond the margins, or cell text that overlaps, overflows, or is truncated.
Images, figure and table captions, formulas, or numbering overlapping the main text.
Formulas extending beyond their boundaries, or long formulas intruding into the margins or subsequent paragraphs.
Abnormally large whitespace, overlapping elements, or orphaned lines in lists, paragraphs, footnotes, and references.
Obvious missing glyphs, garbled characters, or fallback issues in Chinese, English, or mathematical symbol fonts.
Whether critical template pages, such as the cover, abstract page, and appendices, retain the visual structure required by the competition. Any table of contents in CUMCM mode must be treated as an immediate failure.

For template-conversion projects or projects with an existing reference PDF, PDFs from both engines should also be exported to PNG and compared page by page. Differences in page count or page dimensions must be recorded as a critical error or accompanied by a clear explanation.

If the model does not have visual capabilities, it must explicitly state the reason for “Visual inspection was not performed” in

reports/VERIFY_REPORT.md

and must at least complete programmatic checks of PDF non-emptiness, page count, page dimensions, and other basic properties.

Step 9: 写验收报告

Step 9: Write the Acceptance Report

创建

reports/VERIFY_REPORT.md

：

markdown

undefined

Create

reports/VERIFY_REPORT.md

markdown

undefined

验证和验收报告

结论

PASS / FAIL

检查项

检查项	结果	说明

检查项	结果	说明

章节结构

图表引用

数值一致性

文本质量门禁

编译

PDF 视觉检查

CUMCM 2026 合规门禁

问题路由

仍需处理的问题


CUMCM 2026 模式下，还要把本阶段由 AI 执行的文本扫描、数值核对、编译诊断和版式检查追加到 `compliance/AI_USAGE_LOG.md`，并区分自动检查结果与参赛队人工确认。

只有当硬错误都修复、文本门禁通过、核心图表都引用、数值一致、编译通过或明确说明不可编译原因、视觉检查通过或明确说明无法执行原因时，才写 `PASS`。


In CUMCM 2026 mode, the text scans, numerical checks, compilation diagnostics, and layout inspections performed by AI at this stage must also be appended to `compliance/AI_USAGE_LOG.md`, with automated check results distinguished from the participating team's human confirmations.

Only write `PASS` when all critical errors have been fixed, the text gate has passed, all core figures and tables are referenced, numerical values are consistent, compilation has succeeded or the reason it cannot be performed has been clearly stated, and visual inspection has passed or the reason it cannot be performed has been clearly stated.

硬错误标准

Critical Error Criteria

以下问题必须判定

FAIL

：

缺少选定的论文入口文件（
```
main.typ
```
或
```
main.tex
```
）或核心正文。
论文入口引用的章节文件不存在。
Typst 入口缺少
```
#include
```
；LaTeX 入口缺少
```
\input
```
/
```
\include
```
。
正文章节缺少一级标题（Typst
```
= 
```
后缺空格，LaTeX
```
\section{}
```
缺失）。
章节顺序明显错误或重复。
正文仍有占位符。
正文泄露内部工作流文件名。
引用的图片不存在。
关键数值与结果记录冲突。
编译器可用但论文编译失败。
编译后的 PDF 为空、缺页、页数异常或页面尺寸异常且无法解释。
视觉检查发现正文、表格、图片、公式、页眉页脚、页码等关键元素重叠、裁切、越界或乱码。
CUMCM 2026 的任一提交硬门禁失败。

The following issues must result in

FAIL

The selected paper entry file (
```
main.typ
```
or
```
main.tex
```
) or the core main text is missing.
A section file referenced by the paper entry file does not exist.
The Typst entry file has no
```
#include
```
, or the LaTeX entry file has no
```
\input
```
/
```
\include
```
.
A main-text section lacks a first-level heading, such as a missing space after
```
= 
```
in Typst or a missing
```
\section{}
```
in LaTeX.
The section order is clearly incorrect or contains duplicates.
Placeholders remain in the main text.
Internal workflow filenames are exposed in the main text.
A referenced image does not exist.
A key numerical value conflicts with the result records.
The compiler is available but the paper fails to compile.
The compiled PDF is empty, has missing pages, has an abnormal page count, or has abnormal page dimensions that cannot be explained.
Visual inspection finds overlap, clipping, overflow, or garbled text involving key elements such as the main text, tables, images, formulas, headers, footers, or page numbers.
Any CUMCM 2026 mandatory submission gate fails.

错误路由

Error Routing

在

VERIFY_REPORT.md

对每个未解决问题记录：

yaml

issue_type: <assumption_error | numeric_inconsistency | figure_error | writing_error | submission_error>
route_to: <2analysis-modeling | 3coding-visual | 4drawio | 5writing | 7cumcm-submit>
severity: <hard | warning>
evidence: <file, page, command or value>

题意、假设、模型或约束错误 →
```
2analysis-modeling
```
程序、结果、baseline、可行性或复现错误 →
```
3coding-visual
```
非数据图结构与已确认模型不符 →
```
4drawio
```
论文表达、引文、数值抄写或模板源错误 →
```
5writing
```
匿名、AI 详情、文件大小、支撑材料或打包错误 →
```
7cumcm-submit
```

修复后重新执行受影响的下游阶段和本验收，不得只把 FAIL 文本改成 PASS。

Record every unresolved issue in

VERIFY_REPORT.md

yaml

issue_type: <assumption_error | numeric_inconsistency | figure_error | writing_error | submission_error>
route_to: <2analysis-modeling | 3coding-visual | 4drawio | 5writing | 7cumcm-submit>
severity: <hard | warning>
evidence: <file, page, command or value>

Errors in problem interpretation, assumptions, the model, or constraints →
```
2analysis-modeling
```
.
Errors in the program, results, baseline, feasibility, or reproducibility →
```
3coding-visual
```
.
A non-data diagram structure that does not match the confirmed model →
```
4drawio
```
.
Errors in the paper's expression, citations, numerical transcription, or template source →
```
5writing
```
.
Errors involving anonymization, AI details, file size, supporting materials, or packaging →
```
7cumcm-submit
```
.

After fixing the issues, rerun both the affected downstream stages and this acceptance stage; do not merely change the text from FAIL to PASS.

警告标准

Warning Criteria

以下问题可判定为

WARN

，但应尽量修复：

未引用的备用图片。
某章节过短或明显不均衡。
caption 偏长。
参考文献偏少。
图表后解释文字不足。
视觉检查工具不可用，但已经记录原因并完成基础 PDF 元数据检查。
代码完整复现耗时过长，只做了轻量检查。

The following issues may be classified as

WARN

, but should be fixed whenever possible:

Unreferenced backup images.
A section that is too short or clearly unbalanced.
An overly long caption.
Too few references.
Insufficient explanatory text after a figure or table.
The visual inspection tool is unavailable, but the reason has been recorded and basic PDF metadata checks have been completed.
Full code reproduction takes too long, so only a lightweight check was performed.