Law & Economics DOCX Export

What this skill carries — grep references/ for any subject the names below miss: !d=${CLAUDE_SKILL_DIR}; command -v skill-toc >/dev/null 2>&1 && exec skill-toc "$d"; s=$HOME/.claude/skills/plugin-utils/bin/skill-toc; [ -x "$s" ] && exec "$s" "$d"; echo "(skill-toc unavailable: references and scripts are NOT listed here — install the plugin-utils plugin, or start a new session so its bin/ reaches PATH)"

Build a submission-ready Word manuscript from markdown for the Chicago-style law-and-economics journals (JLE, JLS, JLEO, ALER) and for job market papers written for that audience.

These journals publish no official Word template. Their instructions amount to "double spaced throughout, Chicago Manual of Style, author-date with a reference list," and every template circulating online is third-party and mutually inconsistent. This skill is that missing template plus the pipeline that drives it.

Iron Law: never hand-roll the pandoc call

NO L&E .docx WITHOUT build_le_docx.py. A bare pandoc -o out.docx gives Calibri, single spacing, no reference list, numbered headings that collide with the back matter, and pandoc's grid-bordered tables — a document that looks finished and violates four separate JLE requirements. Reaching for raw pandoc to "just get something quickly" is the efficiency drive producing an unusable deliverable; it is slower than the script, not faster.

Agents without the Skill tool (most workflow subagents) can't invoke this skill — run the script directly.

Usage

uv run python3 ${CLAUDE_SKILL_DIR}/scripts/build_le_docx.py PAPER.md [-o OUT.docx] [--pdf]
uv run python3 ${CLAUDE_SKILL_DIR}/scripts/build_le_docx.py PROJECT_DIR      # drafts/*.md in numeric order

Flag	Effect
--acknowledgement "..."	Injects the unnumbered * acknowledgment note. JLE requires one, placed before note 1.
--bibliography FILE	Default, resolved in the document project: references/sources.bib, references.bib, or sources.bib beside the source.
--csl FILE	Default: the vendored assets/chicago-author-date.csl (CMOS 18e author-date).
--spacing onehalf|single	Internal circulation only. Double is the JLE submission requirement and the default.
--pdf	Renders via doc_render.convert after building.

The build: assembles sources (first file's YAML frontmatter becomes the document metadata, later files' footnote labels are namespaced), resolves <!-- include: /abs/path.md --> sentinels, guarantees a ::: {#refs} reference list, runs pandoc with --citeproc + pandoc-crossref + the L&E reference doc, then retags back-matter headings, forces widow control, and restyles tables to booktabs.

Metadata (YAML frontmatter of the first file)

---
title: "Deterrence without Detection"
author:
  - "Isabelle Marchand, University of Chicago Law School, imarchand@uchicago.edu"
  - "A. N. Coauthor, Northwestern University, coauthor@northwestern.edu"
date: "January 2026"
abstract: |
  ... 150 words maximum, strictly enforced by the journal ...
---

JLE requires the title page to carry names, e-mail addresses, and affiliations — put all three in each author entry. The build warns if the abstract exceeds 150 words.

What the template encodes

Full editorial spec — subhead ladder, citation forms, table/figure rules, appendix numbering — is in references/jle-house-style.md. Read it before answering any "how should X be formatted" question; do not answer from memory.

Style	Use	Formatting
Title / Author / Date	Title page	Centered; 14pt bold title
Abstract Title / Abstract	Abstract block	Bold centered head; indented body
Heading 1	1. Section	Bold, auto-numbered
Heading 2	1.1. Subsection	Italic, auto-numbered
Heading 3	1.1.1. Sub-subsection	Roman, auto-numbered
Heading 4	Run-in head	Italic, keepNext
Unnumbered Heading	Abstract / References / Appendix	Bold, no section number
Body Text / First Paragraph	Body	First-line indent; no indent after a heading
Footnote Text	Footnotes	12pt, double spaced
Bibliography	Reference list	Hanging indent, double spaced
Table Note	Note under a table	Double spaced

Everything is Latin Modern Roman at 12pt, double spaced, on US Letter with 1-inch margins and a centered page-number footer. Math is OMML with Latin Modern Math as the math font, so $...$ and $$...$$ render as real Word equations.

Section numbers come from Word list numbering. Write # Introduction, never # 1. Introduction.

Provenance of the typography — and what survived

The typography comes from WordTeX (tomwildenhain.com/wordtex). Its WordTeX Template.dotx turned out to be a pandoc-shaped style set already — Compact, Author, Abstract, BlockText, FootnoteText, TableCaption, CaptionedFigure, SourceCode, VerbatimChar are all pandoc styleIds, and the theme fonts are LM Roman 10 / Latin Modern Math. So the styles extracted cleanly; no hand-rebuild fallback was needed.

What we did not take: the .dot/VSTO plugin, glossary/AutoText, and autocorrect macros. Those are for authoring in Word; pandoc needs a plain .docx that merely carries style definitions.

Where JLE and WordTeX conflict, JLE wins, and the divergence is commented in scripts/make_le_template.py:

WordTeX (LaTeX look)	Template (JLE)	Why
w:line="204" (tight LaTeX leading)	w:line="480" (double)	"double spaced throughout"
11pt body, 8pt footnotes, 10pt abstract	12pt everywhere	"including footnotes, tables, and figure legends"
Justified (jc=both)	Flush left	Chicago manuscript prep; justified + double spaced opens rivers
Numbered Bibliography list	Hanging-indent reference list	Author-date, not numbered references

Regenerating the template

The .docx is built, not hand-edited, so every choice is auditable in one file:

uv run python3 ${CLAUDE_SKILL_DIR}/scripts/make_le_template.py

It starts from pandoc --print-default-data-file reference.docx (guaranteeing every part and styleId pandoc's writer emits is present), transplants the Latin Modern typography, adds the JLE heading numbering and page-number footer, and self-verifies. Output: ${CLAUDE_PLUGIN_ROOT}/references/templates/law_econ_template.docx.

Never open the template in Word and save it. Change make_le_template.py and regenerate.

Round-trip check

examples/sample/ is a small paper exercising headings, a table with a note, a figure, display + inline math, footnotes, and author-date cites:

uv run python3 ${CLAUDE_SKILL_DIR}/scripts/build_le_docx.py \
    ${CLAUDE_SKILL_DIR}/examples/sample/paper.md -o /tmp/sample.docx --pdf

Known limitation: math in the headless PDF path

The build stamps Latin Modern Math in both places it has to go — settings.xml's m:mathPr/m:mathFont (what Word uses; run properties do not override it) and an explicit w:rFonts on every OMML run (for LibreOffice/x2t, which ignore m:mathFont). Both are required: pandoc regenerates settings.xml and drops the template's copy, so without the re-injection Word renders math in Cambria Math even though the body text is correct.

LibreOffice ignores the run font too — it converts OMML to its own formula objects and uses its configured math font, so a LibreOffice render shows Latin Modern body text next to Liberation Serif math. x2t is worse (it misses Latin Modern entirely and falls back to Liberation Sans).

The .docx is correct; only the LibreOffice preview is off. On Linux, render with word-render (below) — verified fully correct. Verified on the sample via LibreOffice: body text is LMRoman10-Regular/Bold/Italic, math is the only fallback.

Do not try to fix this by installing fonts or configuring LibreOffice — both have been tested and neither works.

Attempted fix	Result
Install Latin Modern Math system-wide	No effect. The font is typically already installed (TeX/LaTeX pulls it in) and fc-match "Latin Modern Math" resolves to the real latinmodern-math.otf. Availability was never the problem.
Override LO's Math module default fonts (/org.openoffice.Office.Math/Font → Variables, Functions, Numbers, Text)	No effect. Re-rendered with those forced to Latin Modern Math in a clean LO profile; byte-identical font list. The OOXML import fixes formula fonts at import time and consults neither m:mathFont nor the Math defaults.

To confirm the fallback is math-only rather than a document-wide problem, map glyphs to fonts instead of trusting pdffonts (which ignores -f/-l and reports the whole document for every page) or a vision check (which misjudges double spacing):

uv run --with pdfplumber python3 -c "
import pdfplumber, collections
for i, pg in enumerate(pdfplumber.open('OUT.pdf').pages, 1):
    d = collections.defaultdict(str)
    for ch in pg.chars: d[ch['fontname'].split('+')[-1]] += ch['text']
    print(i, {f: (len(t) if 'LMRoman' in f else t[:60]) for f, t in d.items()})"

Expected: LiberationSerif carries only math glyphs (π ρ ∂ > = + etc.) and appears only on pages with equations. OpenSymbol on page 1 is the * acknowledgment marker, not a fallback. If Liberation shows up on a page with no math, something else is wrong — that is a real bug, not this limitation.

Spacing is likewise verified from the XML, not by eye: every relevant style (BodyText, FirstParagraph, FootnoteText, Bibliography, Abstract, TableNote) must carry w:line="480" (double); 240 would be single.

Getting a correct-math PDF from a Linux box

Use word-render — the QEMU Windows guest driven over SSH (nix module programs.wordRender, see ~/nix/modules/shared/word-render/README.md). It is the only path on Linux that gets body text and math right:

word-render OUT.docx OUT.pdf

Verified on the sample: LMRoman10-{Regular,Bold,Italic} + LatinModernMath-Regular, no Cambria/Calibri.

A fresh guest renders Cambria/Calibri and needs a one-time font install — word-render-install-fonts. Stock lmodern does not work: Word will not render CFF-flavoured OpenType, and it matches families on name ID 1, where Latin Modern stores the optical size (LM Roman 10) rather than the typographic family. Nix builds a converted/renamed set; the README has the full diagnosis.

Other options:

• Ship the .docx. JLE/JLS/JLEO take Word submissions and the file is correct, so the editor's Word renders it properly regardless of your preview.
• A Mac with Word — but not over SSH: macOS TCC blocks it and cmux dispatch can't rescue a remote invocation. See the docx-render skill, "Driving the Mac's Word from another machine over SSH." Prefer word-render.
• LaTeX preview for reading only. tectonic via pandoc gives faithful Latin Modern on Linux, but it is not the submission artifact and diverges from the .docx.

Figures: vector, via svgBlip

Reference the PNG in markdown — ![caption](~/figures/fig1.png) — and keep a same-stem .svg beside it. After pandoc, the build finds each embedded raster, matches it to its SVG by content hash (pandoc rewrites media to rIdN.png, so the filename is gone by then), and attaches the vector:

<a:blip r:embed="rIdPng">
  <a:extLst>
    <a:ext uri="{96DAC541-7B7A-43D3-8B79-37D633B846F1}">
      <asvg:svgBlip xmlns:asvg="http://schemas.microsoft.com/office/drawing/2016/SVG/main"
                    r:embed="rIdSvg"/>
    </a:ext>
  </a:extLst>
</a:blip>

Word 2016+ draws the SVG with its own renderer; everything older falls back to the PNG. No sibling .svg, no change — raster-only projects are unaffected.

The SVG is looked for beside the source files (the L&E layout keeps figure1.png next to paper.md) as well as under the project's figures/ and drafts/, so both the single-file and the PROJECT_DIR invocation find it.

Never reference a bare .svg from markdown. Pandoc embeds it as an image part with no svgBlip, unzip -l shows the media happily, and Word renders nothing — blank space under the caption, no error at any stage.

Never convert SVG→EMF with LibreOffice as a substitute. EMF is a real vector format and Word draws it, so the route looks correct. It is not: LibreOffice's SVG importer silently corrupts complex figures. A five-facet histogram came back missing an entire facet row, every row label, both axis labels, the tick numbers and the zero line — still a plausible-looking chart, so nothing downstream flagged it. Simple one-panel figures convert fine, which is what makes it dangerous: verifying one figure proves nothing about the rest.

Verification — count images in the RENDERED PDF, never in the DOCX:

pdfimages -list manuscript.pdf | tail -n +3 | wc -l   # 0 == every figure is vector

A media part exists for formats Word cannot draw, so a word/media/ count is not evidence the figure reached the page. Check the most structurally complex figure against its source, not the first one.

Verification gate — before you claim the build is done

IDENTIFY the output → RUN the render → READ the pages → VERIFY → CLAIM.

1. Render: python3 ${CLAUDE_PLUGIN_ROOT}/scripts/doc_render.py OUT.docx OUT.pdf (--renderer word for anything a human reads — see the docx-render skill's Iron Law).
2. Rasterize: pdftoppm -r 110 -png OUT.pdf pg.
3. Look at the pages with the visual-verify skill. Check: body font is Computer Modern (not a Times fallback); lines are double spaced; sections read 1., 1.1.; References and Appendix headings carry no number; the table is a real grid with booktabs rules and no word broken mid-token; the figure rendered; the * acknowledgment note is on page 1.
4. Only then report the path.

pandoc exiting 0 is not verification. It exits 0 on a document with a missing figure, a collapsed table, and Calibri fallback.

Red Flags

Action	Why Wrong	Do Instead
About to run pandoc -o out.docx yourself → STOP	No template, no citeproc, no reference list; wrong on four JLE requirements at once	build_le_docx.py
About to use law-review-docx for an L&E paper → STOP	Footnote/Bluebook citations, small caps, TOC — wrong discipline and wrong citation model	This skill
About to write # 1. Introduction → STOP	Word adds its own number: "1. 1. Introduction"	# Introduction
About to put a bare citation in a footnote → STOP	JLE: footnotes must be substantive; simple citations go in the text	[@key] inline
About to add cases to the .bib → STOP	JLE: "Do not include cases in the reference list"	Cite in text: (332 U.S. 392 [1947])
About to build with --spacing single for a submission → STOP	Desk-reject risk; double spacing is explicit and non-negotiable	Default double
About to report success on "pandoc exited 0" → STOP	Exit 0 hides missing figures and collapsed tables	Run the verification gate
About to edit the template in Word → STOP	Next regeneration silently reverts it	Edit make_le_template.py

Related skills

Part of the document skill group:

• law-review-docx — same machinery, law review policy (Bluebook footnote citations, TOC, Times/small caps). build_le_docx.py imports its helpers; see that script's header for why this is a sibling and not a --style flag.
• docx-render — .docx → PDF (--renderer word for deliverables).
• docx-repair — repairs a cloud-editor-damaged .docx.
• docx (generic) — edits docx content (tracked changes, comments).
• visual-verify — required by the verification gate above.