7cumcm-submit

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

CUMCM 2026 提交合规与打包

CUMCM 2026 Submission Compliance and Packaging

本 skill 只在
competition_mode: CUMCM_2026
时使用。先完整读取
../_references/CUMCM_2026_RULES.md
,并以全国组委会最新官方通知为最高优先级。
本阶段收口提交文件,不重新建模、改变结果或替参赛队通过人工审核。若发现实质错误,按
6verity
的问题路由返回上游,修复后重新执行受影响阶段。
This skill is used only when
competition_mode: CUMCM_2026
. First read
../_references/CUMCM_2026_RULES.md
in full, with the latest official notice from the National Organizing Committee taking the highest priority.
This stage finalizes the submission files. It does not remodel the problem, alter the results, or pass manual review on behalf of the participating team. If substantive errors are found, return them upstream through the
6verity
issue-routing process and rerun the affected stages after the issues are fixed.

前置条件

Prerequisites

开始前确认存在:
  • 已编译的单一论文 PDF,以及对应 Typst/LaTeX 源文件;
  • reports/VERIFY_REPORT.md
    ;
  • compliance/AI_USAGE_LOG.md
    、
    HUMAN_REVIEW_LOG.md
    、
    SOURCE_LOG.md
    ;
  • code/run_all.py
    、环境清单、全部必要源码;
  • results/final_results.json
    和约束/有效性检查记录;
  • compliance/private_identity_terms.txt
    ,仅供本地扫描。
H1–H4 任一不是参赛队真实确认的
HUMAN_APPROVED
,或普通验收仍有 hard failure 时,停止打包并写
SUBMISSION_READY = false
。不得代填通过状态。
Before starting, confirm the existence of:
  • A compiled single-paper PDF and the corresponding Typst/LaTeX source files;
  • reports/VERIFY_REPORT.md
    ;
  • compliance/AI_USAGE_LOG.md
    ,
    HUMAN_REVIEW_LOG.md
    , and
    SOURCE_LOG.md
    ;
  • code/run_all.py
    , the environment manifest, and all necessary source code;
  • results/final_results.json
    and constraint/validity check records;
  • compliance/private_identity_terms.txt
    , for local scanning only.
If any of H1–H4 has not been confirmed as
HUMAN_APPROVED
by the participating team, or if the standard acceptance checks still contain a hard failure, stop packaging and write
SUBMISSION_READY = false
. Approval status must not be filled in on the team's behalf.

1. 论文硬门禁

1. Paper Hard Gates

对源文件、PDF 文本、PDF 元数据和逐页渲染结果交叉检查:
  1. 第 1 页是包含标题、摘要、关键词的摘要专用页;无承诺书和编号专用页。
  2. 摘要原则上只占第 1 页;页码从摘要页的 1 开始,在页脚中部连续编号。
  3. 不存在目录或目录命令。
  4. A4,四边距至少 2.5 cm;正文在摘要之后、附录之前且不超过 30 页;附录不计入正文上限。
  5. PDF 不超过 20 MiB。
  6. “AI工具使用声明”使用官方原句并位于参考文献之前。
  7. 附录文件列表与实际支撑材料一致,并包含全部完整、可运行源程序代码或官方规定的无程序说明。
源文件中的
CUMCM_BODY_START
/
CUMCM_BODY_END
只帮助定位正文边界;最终页数必须结合编译后的 PDF 核对,不能用 PDF 总页数代替正文页数。
Cross-check the source files, PDF text, PDF metadata, and rendered output from every page:
  1. Page 1 must be a dedicated abstract page containing the title, abstract, and keywords; there must be no commitment letter or dedicated identification page.
  2. In principle, the abstract must occupy only page 1. Page numbering must start at 1 on the abstract page and continue consecutively at the center of the footer.
  3. There must be no table of contents or table-of-contents command.
  4. The document must use A4 paper, with margins of at least 2.5 cm on all sides. The main text must appear after the abstract and before the appendix and must not exceed 30 pages. Appendices do not count toward the main-text limit.
  5. The PDF must not exceed 20 MiB.
  6. The “AI工具使用声明” must use the official wording verbatim and appear before the references.
  7. The appendix file list must match the actual supporting materials and include all complete, runnable source code or the official no-program statement.
The
CUMCM_BODY_START
/
CUMCM_BODY_END
markers in the source files only help locate the main-text boundaries. The final page count must be verified against the compiled PDF; the total number of PDF pages cannot be used as a substitute for the main-text page count.

2. 生成并核对 AI 使用详情

2. Generate and Reconcile AI Usage Details

先把本阶段由 AI 辅助完成的合规扫描、匿名检查、详情汇总和打包操作追加到
AI_USAGE_LOG.md
,再生成详情 PDF,保证日志覆盖完整工作流。生成详情之后若 AI 又参与修复或重打包,应更新日志并重新生成详情。
从真实的
AI_USAGE_LOG.md
和
HUMAN_REVIEW_LOG.md
汇总
support_materials/AI工具使用详情.pdf
。至少覆盖:
  • 每个 AI 工具的名称、版本或模型;
  • 具体用途和使用阶段;
  • 主要提示方式与使用过程,可附典型交互;
  • 输出的采纳、未采纳、人工修改和人工核验;语言润色除外。
可运行:
bash
python3 <skill-dir>/scripts/build_ai_details.py \
  --ai-log compliance/AI_USAGE_LOG.md \
  --human-log compliance/HUMAN_REVIEW_LOG.md \
  --output support_materials/AI工具使用详情.pdf
脚本使用 XeLaTeX 生成 PDF;若环境没有
xelatex
,它会保留可编译
.tex
并停止,随后应使用可用的文档/PDF 工具生成同名 PDF。不要为“通过检查”补造工具版本、提示或人工核验。
对照日志、论文声明和详情:工具清单、用途和人工审核必须一致。使用本 Agent 即属于使用 AI,论文不得声明未使用。
First append the AI-assisted compliance scanning, anonymity checks, detail aggregation, and packaging operations performed in this stage to
AI_USAGE_LOG.md
, and then generate the details PDF, ensuring that the log covers the complete workflow. If AI participates in repairs or repackaging after the details PDF has been generated, update the log and regenerate the details PDF.
Use the actual
AI_USAGE_LOG.md
and
HUMAN_REVIEW_LOG.md
to generate
support_materials/AI工具使用详情.pdf
. It must cover at least:
  • The name and version or model of every AI tool;
  • The specific purpose and stage of use;
  • The primary prompting method and usage process, with typical interactions optionally included;
  • Whether outputs were adopted or rejected, and any human modifications and human verification; language polishing is exempt.
Runnable:
bash
python3 <skill-dir>/scripts/build_ai_details.py \
  --ai-log compliance/AI_USAGE_LOG.md \
  --human-log compliance/HUMAN_REVIEW_LOG.md \
  --output support_materials/AI工具使用详情.pdf
The script uses XeLaTeX to generate the PDF. If
xelatex
is unavailable in the environment, it retains the compilable
.tex
file and stops. In that case, use an available document/PDF tool to generate the PDF with the same filename. Do not fabricate tool versions, prompts, or human verification merely to pass the checks.
Compare the logs, the paper declaration, and the details. The tool lists, purposes, and human reviews must be consistent. Use of this Agent constitutes AI use, and the paper must not claim that no AI was used.

3. 匿名扫描

3. Anonymity Scan

使用私有词表扫描论文、源码、数据文件名、支撑材料、压缩包成员名、PDF 文本与可读取元数据:
bash
python3 <skill-dir>/scripts/check_anonymity.py \
  --terms compliance/private_identity_terms.txt \
  paper support_materials code results
检查姓名、学校全称与简称、学院、导师、赛区、学号、邮箱、手机号、用户名、绝对路径以及 PDF/图片/Office 元数据。词表和扫描报告不得进入提交包。任何明确身份命中都是 hard failure;不得删除日志来隐藏命中。
Use the private term list to scan the paper, source code, data filenames, supporting materials, archive member names, PDF text, and readable metadata:
bash
python3 <skill-dir>/scripts/check_anonymity.py \
  --terms compliance/private_identity_terms.txt \
  paper support_materials code results
Check full names and abbreviations of individuals, schools, colleges, advisors, regional divisions, student IDs, email addresses, phone numbers, usernames, absolute paths, and PDF/image/Office metadata. Neither the term list nor the scan report may be included in the submission package. Any confirmed identity match is a hard failure; do not delete logs to conceal matches.

4. 来源纪律检查

4. Source Discipline Check

读取
SOURCE_LOG.md
。正式竞赛模式如发现交流平台上的当前赛题答案、思路、代码、论文或讨论,记录
POTENTIAL_COMPETITION_RULE_VIOLATION
,停止提交就绪判定并交由参赛队处理。不要自动清洗、删除或改写来源证据。
Read
SOURCE_LOG.md
. In formal competition mode, if current competition answers, ideas, code, papers, or discussions are found on communication platforms, record
POTENTIAL_COMPETITION_RULE_VIOLATION
, suspend the submission-ready determination, and refer the matter to the participating team. Do not automatically clean up, delete, or rewrite source evidence.

5. 组织支撑材料

5. Organize Supporting Materials

support_materials/
至少包含实际使用的:
text
AI工具使用详情.pdf
code/                  # 全部可运行源码与统一入口
source_data/           # 自主查阅使用且允许提交的数据
result_tables/         # 必要的大篇幅中间结果
README.txt             # 最小复现说明
从论文附录文件列表生成
compliance/SUBMISSION_MANIFEST.md
,逐项核对文件路径、用途、大小和校验值。不要把赛题原始附件、承诺书、编号专用页、身份词表、私有日志、缓存、临时文件、绝对路径或无关环境文件误放进包。
生成 ZIP:
bash
python3 <skill-dir>/scripts/build_support_package.py \
  --support-dir support_materials \
  --output submission/support_materials.zip \
  --manifest compliance/SUBMISSION_MANIFEST.md
若压缩包超过 20 MiB,只能通过移除无关文件、压缩可替代的大型图表或改进文件格式来减小;不得删除官方要求的必要源码、AI 详情或支撑证据。
support_materials/
must contain at least the following items that were actually used:
text
AI工具使用详情.pdf
code/                  # 全部可运行源码与统一入口
source_data/           # 自主查阅使用且允许提交的数据
result_tables/         # 必要的大篇幅中间结果
README.txt             # 最小复现说明
Generate
compliance/SUBMISSION_MANIFEST.md
from the appendix file list in the paper, and verify each file path, purpose, size, and checksum. Ensure that the package does not mistakenly include the original competition attachments, commitment letter, dedicated identification page, identity term list, private logs, caches, temporary files, absolute paths, or irrelevant environment files.
Generate the ZIP archive:
bash
python3 <skill-dir>/scripts/build_support_package.py \
  --support-dir support_materials \
  --output submission/support_materials.zip \
  --manifest compliance/SUBMISSION_MANIFEST.md
If the archive exceeds 20 MiB, it may only be reduced by removing irrelevant files, compressing substitutable large charts, or improving file formats. Required source code, AI details, or supporting evidence must not be deleted.

6. 自动预检与人工视觉确认

6. Automated Preflight and Manual Visual Confirmation

运行:
bash
python3 <skill-dir>/scripts/check_submission.py \
  --root . \
  --paper paper/paper.pdf \
  --archive submission/support_materials.zip \
  --identity-terms compliance/private_identity_terms.txt
自动检查不能可靠判断摘要是否视觉上恰好单页、页码是否位于页脚中部、正文边界页数和全部版式细节。必须把 PDF 逐页渲染并由参赛队完成最终视觉确认,将证据写入
CUMCM_2026_CHECKLIST.md
。
Run:
bash
python3 <skill-dir>/scripts/check_submission.py \
  --root . \
  --paper paper/paper.pdf \
  --archive submission/support_materials.zip \
  --identity-terms compliance/private_identity_terms.txt
Automated checks cannot reliably determine whether the abstract is visually confined to exactly one page, whether page numbers are positioned at the center of the footer, the main-text boundary page count, or every layout detail. The PDF must be rendered page by page, and the participating team must complete the final visual confirmation. Record the evidence in
CUMCM_2026_CHECKLIST.md
.

7. 最终结论

7. Final Conclusion

更新:
  • compliance/CUMCM_2026_CHECKLIST.md
  • compliance/SUBMISSION_MANIFEST.md
  • reports/VERIFY_REPORT.md
只有规则参考中的全部条件、自动检查和参赛队视觉确认都通过,才能写:
text
SUBMISSION_READY = true
任何失败、待人工确认、未执行检查或纪律风险都必须写
SUBMISSION_READY = false
,并附证据与返回阶段。
Update:
  • compliance/CUMCM_2026_CHECKLIST.md
  • compliance/SUBMISSION_MANIFEST.md
  • reports/VERIFY_REPORT.md
Only when all conditions in the rules reference, the automated checks, and the participating team's visual confirmation have passed may the following be written:
text
SUBMISSION_READY = true
Any failure, pending manual confirmation, unexecuted check, or disciplinary risk must result in
SUBMISSION_READY = false
, accompanied by evidence and the stage to which the issue should be returned.