Suda thesis

Suda Thesis

Generate Suzhou University undergraduate academic documents as .docx files with python-docx. For long documents (thesis body, literature review, and literature translation), always generate segmented Python code: content lives in content_part*.py modules, one aggregator imports those modules, and one create_*_docx.py script handles formatting and assembly. For short form documents (task book and midterm-check), use a focused table/form script unless the user's supplied content is long enough to require content parts.

Bundled Resources

• scripts/split_pattern/: the only retained code scaffold. Copy or adapt this folder into the user's output directory when building a new document. Do not edit the bundled scaffold in place for a user paper.
• skills/humanizer/: bundled academic prose humanizer. Use it after thesis body and literature review drafts are generated.
• references/format-standards.md: formatting details for page setup, styles, TOC, bookmarks, figure captions, references, and metadata cleanup.
• references/forms.md: task-book and midterm-check field structure, writing rules, and table/form review checklist.
• references/pitfalls.md: known failure modes and exact code patterns to avoid.
• requirements.txt: Python 3.9+ runtime dependencies for the bundled DOCX scaffold.

Supported Documents

Document	Output Script	Required Split Use	Humanizer
Thesis body	create_thesis_docx.py	Always split	Required after draft
Literature review	create_review_docx.py	Always split	Required after draft
Literature translation	create_translation_docx.py	Always split for long papers	Preserve source meaning; use only for faithful academic fluency checks
Task book	create_task_book_docx.py	Not required; use table/form script	Formal concision check only
Midterm-check form	create_midterm_check_docx.py	Not required; use table/form script	Formal concision check only

Currently supported Suzhou University document types are thesis body, literature review, literature translation, task book, and midterm-check form.

Non-Negotiable Rules

1. Use the split pattern for thesis body, literature review, and long translation tasks. Never put long prose into one giant create_*_docx.py.
2. Keep long-document create_*_docx.py files focused on formatting and assembly. Store prose, references, figure maps, and captions in content_part*.py files imported through content_data.py.
3. Never fabricate references. Verify title, authors, and journal/source before using a reference. A real DOI or article number is not enough if the title/authors/source do not match.
4. Never fabricate experimental details. Ask what the student actually measured or observed before writing Methods, Results, or Discussion text.
5. Introduce each specialized term or abbreviation exactly once across the whole document, including abstracts and body text.
6. Visually inspect every figure before insertion and confirm that the image, caption, surrounding text, and citation match.
7. For thesis body and literature review, run the bundled humanizer after drafting prose and before final DOCX delivery. Use skills/humanizer/SKILL.md in academic mode, preserving all evidence, citations, data, terminology, and claim strength.
8. Erase DOCX privacy metadata for every generated file type. Clear creator/editor fields, comments, title/subject/keywords/category, company/manager fields, and python-docx application fingerprints as described in references/format-standards.md.
9. Never deliver after the first successful generation alone. Every generated DOCX must pass the review step for its document type; thesis body and literature review must repeat the review-regenerate loop until figure-content correspondence, reference authenticity, humanizer quality, and formatting checks pass.
10. For task book and midterm-check forms, do not fabricate student identity, advisor identity, project membership, dates, progress, results, or review opinions. Use placeholders when the user has not provided required form fields.

Standard Workflow

1. Gather the topic, document type, output directory, source materials, reference list expectations, and figure availability. For translation, start from the user's topic direction or thesis title; if the user provides an exact source paper/PDF, use it, and otherwise search for and verify a real English article before translating it. For thesis body, gather the actual experimental protocol and observations. For task book and midterm-check forms, start from the topic direction plus the user's existing results or progress; gather college, major, student, advisor, dates, project type, schedule, and signature fields when available, and use placeholders for missing identity/date fields.

2. Read only the resources needed for the current task:

   • For thesis body, literature review, and long translation tasks, inspect scripts/split_pattern/create_thesis_docx.py and scripts/split_pattern/content_data.py to understand the scaffold.
   • For task book and midterm-check forms, read references/forms.md.
   • Read references/format-standards.md before writing or modifying formatting code.
   • Read references/pitfalls.md before finalizing any generated script.

3. Choose the generation pattern:

   • For thesis body, literature review, and long translation tasks, copy/adapt scripts/split_pattern/ into the output directory. Rename the assembly script to match the document type: create_thesis_docx.py, create_review_docx.py, or create_translation_docx.py.
   • For task book and midterm-check forms, create a focused table/form script named create_task_book_docx.py or create_midterm_check_docx.py. Reuse formatting helpers from references/format-standards.md; keep form data in constants near the top of the script, or in content_data.py if the form text is long.

4. For long documents, generate content modules by section. Use parallel workers when available; otherwise write them sequentially. Keep each module small enough to avoid truncation or broken string literals.

5. For thesis body and literature review, run the bundled humanizer:

   • Read skills/humanizer/SKILL.md.
   • Use Academic mode.
   • Also read skills/humanizer/references/academic-style.md.
   • If matching the user's thesis voice is needed, also read skills/humanizer/references/thesis-style.md.
   • Humanize the prose in content_part*.py section by section while preserving references, figure mentions, numeric values, methods, and cautious academic claims.

6. For long documents, build content_data.py as the only aggregation layer. The assembly script should import from content_data.py, not from each content part directly.

7. Run syntax checks and generate the document.

   Use Python 3.9 or later. On macOS/Linux, create a repository-local .venv instead of installing into the system Python; this is compatible with PEP 668. Set SUDA_THESIS_ROOT to the installed clone and keep using that clone's interpreter when the scaffold is copied to another output directory:

   SUDA_THESIS_ROOT=/path/to/suda-thesis
   python3 -m venv "$SUDA_THESIS_ROOT/.venv"
   "$SUDA_THESIS_ROOT/.venv/bin/python" -m pip install -r "$SUDA_THESIS_ROOT/requirements.txt"
   

   For long split documents:

   "$SUDA_THESIS_ROOT/.venv/bin/python" -m py_compile content_part*.py content_data.py create_*_docx.py
   "$SUDA_THESIS_ROOT/.venv/bin/python" create_*_docx.py
   

   For form documents:

   "$SUDA_THESIS_ROOT/.venv/bin/python" -m py_compile create_task_book_docx.py
   "$SUDA_THESIS_ROOT/.venv/bin/python" create_task_book_docx.py
   # or
   "$SUDA_THESIS_ROOT/.venv/bin/python" -m py_compile create_midterm_check_docx.py
   "$SUDA_THESIS_ROOT/.venv/bin/python" create_midterm_check_docx.py
   

8. Run the review-regenerate loop below. If any issue is found, fix the relevant content part, form data constants, figure map, reference list, or assembly script; regenerate the DOCX; then review again.

9. Refresh and verify fields after the final pagination-affecting change. Thesis front matter uses lowercase Roman page numbers; the body starts a new section at -1-. Mark TOC/PAGEREF fields dirty and set w:updateFields, but do not assume every renderer honors those flags. Update the TOC in Word or another field-aware editor, or compare the cached page numbers with the latest rendered pages before delivery.

Split Pattern Contract

A typical output directory should look like this:

content_part1_abstract.py
content_part2_ch1.py
content_part3_ch2.py
content_part4_ch3.py
content_part5_ch4_ch5.py
content_part6_refs.py
content_data.py
create_<document_type>_docx.py
figures/

This layout is a starting point, not a hard chapter count. Add more content_partN_*.py files when a document has more sections, or rename the suffixes to match a translation's original paper structure. Preserve the core contract:

• content_part*.py contains prose/data constants only.
• content_data.py imports and re-exports the constants used by the assembly script.
• create_*_docx.py imports from content_data.py, defines formatting helpers, inserts content, cleans metadata, saves the DOCX, and prints the output path.

Document Rules

Thesis Body

• Include Chinese abstract, English abstract, TOC, chapters, references, and acknowledgement.
• Use Suzhou University page setup, heading styles, TOC field codes, and -N- page numbers.
• References and acknowledgement are not numbered chapters. Style them visually as first-level headings and add them to the TOC without putting them in Word's heading outline.
• Use the student's actual data and observations only.
• After the first complete draft, run the bundled humanizer before regenerating the final DOCX.

Literature Review

• Include abstract, TOC, foreword/introduction, thematic chapters, conclusion, and references.
• Let chapter count follow the topic. Do not force a fixed number of chapters.
• Use real, verified references only. Put citation numbers at the end of clauses or sentences rather than directly after author names.
• Use figures only from cited sources, and include the source in each caption.
• After the first complete draft, run the bundled humanizer before regenerating the final DOCX.
• Verify every reference used in the review before delivery. Literature review citations are high-risk because plausible but fake bibliographic entries are easy to generate.

Literature Translation

• Mirror the source paper's structure exactly. Do not add sections that are absent from the original paper.
• Translate every paragraph without summarizing or condensing.
• Preserve original figure and panel references such as Figure 1B-C.
• Include the source paper as the only reference unless the user explicitly asks for additional context.
• Split by original paper section when the translation is long.
• Keep or download the source PDF/source URL alongside the generated DOCX when available, so translation structure and figure references can be checked.

Task Book

• Generate the Suzhou University undergraduate thesis/design task book as a structured table/form DOCX.
• Include the document title, college/department line, topic line, student/advisor/basic information table, major task and goal, main content, basic requirements, main references, and schedule.
• Use concise formal prose. The task book should describe what the project will do, what the student must complete, and the expected schedule; it should not read like a full thesis chapter.
• Use real, verified references for the reference section. If the user has not provided references, search or ask before filling them; do not invent plausible entries.
• Use placeholders such as XXX or 待填写 for missing student/advisor/date fields rather than inventing personal information.
• Follow references/forms.md for field order and table/form review.
• The split pattern is not required unless the user provides unusually long content.

Midterm-Check Form

• Generate the Suzhou University undergraduate thesis/design midterm progress check form as a table/form DOCX.
• Include the title, college line, student/year/major/date fields, topic, completed tasks, whether progress matches the task book, remaining tasks, whether the project can finish on time, problems, proposed solutions, and opinion/signature placeholders.
• Describe completed work at the work-progress level. Do not insert specific numerical results, named variables, private data, or detailed conclusions unless the user explicitly provides them and asks to keep them.
• Problems and solutions should be practical and concise: workload, data organization, figure preparation, writing schedule, format checking, and revision plan. Avoid turning the form into a literature discussion.
• Do not fabricate progress or approval opinions. Use neutral placeholders for signatures and review comments unless the user provides exact wording.
• Follow references/forms.md for field order and table/form review.
• The split pattern is not required unless the user provides unusually long content.

Review-Regenerate Loop

Run this loop after every generated DOCX for thesis body and literature review. Do not stop at "the script ran" or "the file exists". For literature translation, task book, and midterm-check forms, run the shorter document-specific reviews at the end of this section.

1. Figure-content correspondence (highest priority):

   • Inspect every figure file visually.
   • Locate the exact section where the figure is inserted.
   • Confirm the surrounding paragraph discusses the same object, method, result, or mechanism shown in the image.
   • Confirm the caption describes the actual image and, for literature review figures, cites the same verified source as the image.
   • If any figure is missing, mislabeled, from the wrong paper, or placed in the wrong section, fix FIGURE_MAP, FIGURE_CAPTIONS, or the relevant prose and regenerate.

2. Reference authenticity (highest priority):

   • For thesis body and literature review, verify each reference title, authors, journal or source, year, volume/issue, pages/article number, and DOI/URL when present.
   • A real DOI or article number alone is not enough; title, authors, and source must match.
   • Check every in-text citation is within range and every cited source appears in the reference list.
   • If any reference cannot be verified, replace it with a real source or remove the unsupported claim, then regenerate.

3. Humanizer quality for thesis/review:

   • Confirm the bundled humanizer has been applied to all prose content parts.
   • Re-scan for AI-style artifacts such as empty significance claims, overlong English-syntax Chinese, vague author-name narrative citations, filler 等, metaphors, colloquialisms, first-person thesis prose, and exaggerated causal claims.
   • If the humanizer introduced new issues, fix them and regenerate.

4. Format and document integrity:

   • Check page setup, header/footer, heading styles, TOC, bookmarks, figure captions, references, acknowledgement, page numbers, and metadata erasure.
   • Re-run syntax checks before each regeneration.

Repeat the loop until all four categories pass. Keep the fixes in the segmented content files so the final DOCX can be regenerated reproducibly.

For task book and midterm-check forms, run this shorter review before delivery:

1. Confirm every required form field is filled from user-provided information or marked with an explicit placeholder.
2. Confirm references in a task book are real and formatted consistently.
3. Confirm midterm progress statements do not fabricate results, measurements, or approval opinions.
4. Confirm table borders, merged cells, fonts, spacing, margins, and metadata cleanup.
5. If any issue is found, revise the form script and regenerate the DOCX.

For literature translation, run this review before delivery:

1. Confirm the generated document mirrors the source paper's section order and headings.
2. Confirm every source paragraph is translated without summarizing, omission, or added sections.
3. Confirm original figure, table, supplement, and panel references are preserved.
4. Confirm the source paper is the only reference unless the user requested added context.
5. Confirm formatting and metadata cleanup. If any issue is found, revise the relevant content part or assembly script and regenerate the DOCX.

Post-Generation Review

Before delivering the document, confirm the final loop pass:

• Long-document formatting: CJK fonts set through XML, first-line indent uses firstLineChars='200', heading spacing uses beforeLines='100' and afterLines='100', front matter uses lowercase Roman page numbers, the thesis body restarts at -1-, TOC field code and cached pages are correct, captions are 12 pt, and metadata is erased.
• Content: abstract matches the chapters, transitions are natural, conclusion answers the document's opening questions, and thesis/review prose has passed the bundled humanizer.
• References: every entry is real, citation numbers are in range, GB/T 7714 formatting is consistent, and title/authors/source match verification results.
• Figures: every image has been inspected, is placed in the correct section, and has a matching caption and source.
• Translation: source structure, paragraph coverage, figure/table references, and source reference policy have been checked against the original paper.
• Forms: task book and midterm-check fields are complete, table layout is intact, missing personal information remains explicit placeholders, and no progress or review opinion has been invented.

If any check fails, fix the relevant content part or assembly script, regenerate the DOCX, and repeat verification.