Converting between LaTeX, Typst, Word and Markdown

A short LaTeX note, converted to Typst by one pandoc command and compiled inside a small frame written by hand, still said “Table Table 1”, “Equation [eq:mean]” and “Section Section 1”, and it had lost a drawing without an error. A conversion that compiles is not yet a conversion that is right.

This lesson converts one note between LaTeX, Typst, Word and Markdown with pandoc, and turns it into HTML twice. For each conversion you will see what pandoc reported, what the new file says on the page, and how to prove, by counting and by comparing the text of the PDFs, that nothing went missing. Every session was recorded with pandoc 3.12, TeX Live 2026 and Typst 0.15.1.

pandoc reads a document into a tree, then writes the tree again

pandoc does not translate LaTeX into Typst directly. A reader parses the input into pandoc’s own document treedocument tree: pandoc's own representation of a document: blocks such as headings, paragraphs and tables, holding inlines such as words, emphasis, maths and citations. Every reader builds it and every writer works from it., and a writer turns the tree into the output format, so any reader can feed any writer. Its manual says the cost plainly: the tree is less expressive than many formats, so “one should not expect perfect conversions between every format and every other”. This is the note, and the start of its tree:

% note.tex: a short field note with the things a conversion has to carry.
\documentclass{article}
\usepackage{amsmath}
\usepackage{booktabs}
\usepackage{tikz}
\usepackage[numbers]{natbib}
\newcommand{\mean}[1]{\bar{#1}}
\title{Rain on two roofs}
\author{Meera Pillai}
\date{}
\begin{document}
\maketitle
\section{Method}\label{sec:method}
Two gauges logged the rain for \emph{two weeks}\footnote{Readings were taken at 8 a.m.} \citep{rao2025gauges}.
The daily mean of roof $r$ is
\begin{equation}\label{eq:mean}
  \mean{x}_r = \frac{1}{n}\sum_{i=1}^{n} x_{r,i}
\end{equation}
and Table~\ref{tab:totals} lists the totals.
\begin{itemize}
  \item Block A faces east.
  \item Block B faces west.
\end{itemize}
\section{Result}
\begin{table}[h]
  \centering
  \caption{Totals for two weeks}\label{tab:totals}
  \begin{tabular}{lr}
    \toprule
    Roof & Rain (mm) \\
    \midrule
    Block A & 83.0 \\
    Block B & 76.5 \\
    \bottomrule
  \end{tabular}
\end{table}
Equation~\eqref{eq:mean} gives $\mean{x}_A = 5.93$\,mm a day, as Section~\ref{sec:method} explains.
\begin{tikzpicture}
  \draw (0,0) rectangle (2,1) node[midway] {Block A};
\end{tikzpicture}
\bibliographystyle{plainnat}
\bibliography{refs}
\end{document}

from session import files, needs, run

# What pandoc reads: the start of note.tex as pandoc's own document tree (-t native), which every writer works from.
needs("pandoc")
files("note/note.tex")
run("""
pandoc note.tex -t native | head -24
""")

$ pandoc note.tex -t native | head -24
[ Header 1 ( "sec:method" , [] , [] ) [ Str "Method" ]
, Para
    [ Str "Two"
    , Space
    , Str "gauges"
    , Space
    , Str "logged"
    , Space
    , Str "the"
    , Space
    , Str "rain"
    , Space
    , Str "for"
    , Space
    , Emph [ Str "two" , Space , Str "weeks" ]
    , Note
        [ Para
            [ Str "Readings"
            , Space
            , Str "were"
            , Space
            , Str "taken"
            , Space
            , Str "at"

The first section became Header 1 ( "sec:method" , [] , [] ) [ Str "Method" ]: a level, the label, and the words. A sentence became a Para of Str and Space items, the emphasis an Emph, the footnote a Note. Step through what happened to the whole note on its way to Typst:

Two things follow from the tree. Anything the tree has no node for, such as a TikZ picture, cannot reach any output: the reader skips it, and says so only when asked with --verbose. And a custom macro is applied, not kept: pandoc read \newcommand{\mean}[1]{\bar{#1}} and expanded every use, so the Typst file has macron(x) where note.tex had \mean{x}, and a reader of the Typst file never learns that the macro existed.

From LaTeX to Typst, the text arrives and the references need work

pandoc can write a whole Typst document with its own template (-s), or only the body, which a Typst file of your own then includes. This session tries the first, converts the body with --verbose to see what was skipped, compiles it inside a hand-written main.typ, and compares three sentences of the two PDFs:

// main.typ: the frame of the converted note, written by hand. pandoc converts only the body, into body.typ.
#set document(title: [Rain on two roofs], author: "Meera Pillai")
#set page(paper: "a4")
#set heading(numbering: "1")
#set math.equation(numbering: "(1)")
#show figure.where(kind: table): set figure.caption(position: top)
#align(center)[
  #text(17pt)[Rain on two roofs] \
  Meera Pillai
]

#include "body.typ"

#bibliography("refs.bib", title: [References])

from session import files, needs, run

# note.tex converted to Typst twice: first as a whole document with pandoc's own template, then only the body, which
# main.typ, written by hand, includes. Then three sentences of the LaTeX PDF and of the Typst PDF side by side, and
# how often each PDF's text says Block A.
needs("pandoc", "typst", "pdflatex", "latexmk", "pdftotext", "git")
files("note/note.tex", "note/refs.bib", "note/main.typ")
run("""
!pandoc -s note.tex -o whole.typ && typst compile whole.typ
pandoc --verbose note.tex -t typst -o body.typ 2>&1 | grep 'Skipped'
typst compile main.typ note-typst.pdf
latexmk -pdf -silent note.tex > /dev/null 2>&1
pdftotext note.pdf latex.txt && pdftotext note-typst.pdf typst.txt
grep -E 'daily mean|lists the totals|explains' latex.txt typst.txt
grep -c 'Block A' latex.txt typst.txt
""", env={"SOURCE_DATE_EPOCH": "1767225600"})

$ pandoc -s note.tex -o whole.typ && typst compile whole.typ
error: cannot reference heading without numbering
    ┌─ whole.typ:182:16
    │
182 │ day, as Section~@sec:method explains.
    │                 ^^^^^^^^^^^
    │
    = hint: you can enable heading numbering with `#set heading(numbering: "1.")`

[exit status 1]
$ pandoc --verbose note.tex -t typst -o body.typ 2>&1 | grep 'Skipped'
[INFO] Skipped '\maketitle' at note.tex line 12 column 11
[INFO] Skipped '\centering' at note.tex line 26 column 13
[INFO] Skipped '\begin{tikzpicture}
[INFO] Skipped '\bibliographystyle{plainnat}' at note.tex line 41 column 29
$ typst compile main.typ note-typst.pdf
$ latexmk -pdf -silent note.tex > /dev/null 2>&1
$ pdftotext note.pdf latex.txt && pdftotext note-typst.pdf typst.txt
$ grep -E 'daily mean|lists the totals|explains' latex.txt typst.txt
latex.txt:Two gauges logged the rain for two weeks 1 [1]. The daily mean of roof r is
latex.txt:and Table 1 lists the totals.
latex.txt:Equation (1) gives x̄A = 5.93 mm a day, as Section 1 explains.
typst.txt:Two gauges logged the rain for two weeks1 [1]. The daily mean of roof 𝑟 is
typst.txt:and Table Table 1 lists the totals.
typst.txt:Equation [eq:mean] gives 𝑥̄𝐴 = 5.93 mm a day, as Section Section 1 explains.
$ grep -c 'Block A' latex.txt typst.txt
latex.txt:3
typst.txt:2

The whole document stopped at once: cannot reference heading without numbering, because pandoc’s template leaves headings unnumbered while the text refers to them. The body compiled inside main.typ, which numbers headings and equations, and the three sentences then showed the faults a reader would see:

pandoc’s --verbose messages named four skipped commands, \maketitle, \centering, the tikzpicture and \bibliographystyle, but the doubled words and the wrong reference compiled without a message. The Typst guide for LaTeX users points to the same route, pandoc, and says the Typst web app has the conversion built in; either way, the result has to be checked.

Maths converts formula by formula

Maths is the part pandoc converts best, because LaTeX maths has a node of its own in the tree. Ten formulas, each as pandoc wrote it, and the converted file compiled as a check:

$\frac{a+b}{2}$

$\sqrt[3]{x}$

$x \in \mathbb{R}$

$\hat{\theta} = \operatorname{arg\,max}_\theta L(\theta)$

$|x| = \begin{cases} x & \text{if } x \ge 0 \\ -x & \text{otherwise} \end{cases}$

$\left( \frac{1}{2} \right)^n$

$\int_0^1 x^2 \, \mathrm{d}x = \tfrac{1}{3}$

$A = \begin{pmatrix} 1 & 2 \\ 3 & 4 \end{pmatrix}$

$\vec{v} \cdot \vec{w} \le \lVert v \rVert \lVert w \rVert$

$\mathcal{O}(n \log n)$

from session import files, needs, run

# Ten LaTeX formulas, each as pandoc writes it in Typst, then the converted file compiled as a check.
needs("pandoc", "typst")
files("maths/maths.tex")
run("""
pandoc maths.tex -t typst
pandoc maths.tex -t typst -o maths.typ && typst compile maths.typ
""", env={"SOURCE_DATE_EPOCH": "1767225600"})

$ pandoc maths.tex -t typst
$frac(a + b, 2)$

$root(3, x)$

$x in bb(R)$

$hat(theta) = "arg max"_theta L\(theta\)$

$\|x\|= cases(delim: "{", x & upright("if ") x gt.eq 0, - x & upright("otherwise"))$

$(1 / 2)^n$

$integral_0^1 x^2 thin upright(d) x = 1 / 3$

$A = mat(delim: "(", 1, 2; 3, 4)$

$arrow(v) dot.op arrow(w) lt.eq lr(bar.v.double v bar.v.double) lr(bar.v.double w bar.v.double)$

$cal(O)\(n log n\)$
$ pandoc maths.tex -t typst -o maths.typ && typst compile maths.typ

All ten compiled. Most need nothing more, and a few are worth a look before you keep them:

When a maths-heavy file must keep its LaTeX syntax for a while, MiTeX, a package on Typst Universe described as “LaTeX support for Typst, powered by Rust and WASM”, renders LaTeX maths inside a Typst document with #mi(...) and #mitex(...).

Word and Markdown keep the structure, not the LaTeX

A .docx file is a ZIP package of XML parts, defined by the ECMA-376 standard. pandoc’s manual says that maths in a Word file is written as OMMLOMML: Office Math Markup Language, the XML in which Word files store equations, as m:oMath elements, so that they stay editable in a word processor., Word’s own maths markup. This session writes the note as a Word file with citations formatted by pandoc’s --citeproc, lists what the file holds, reads it back as Markdown, and then turns a Markdown file of lecture notes into a PDF through Typst:

---
title: Rain gauges on flat roofs
author: Meera Pillai
---

# What we measured

Two gauges logged the rain on two roofs for two weeks [@rao2025gauges].
Each daily reading $x_i$ was added up, so the total of a roof is $\sum_{i=1}^{14} x_i$.

| Roof    | Rain (mm) |
|---------|----------:|
| Block A |      83.0 |
| Block B |      76.5 |

# What to do next

- Read both gauges at the same time each day.
- Write *missing* when a reading cannot be taken.

from session import files, needs, run

# The note as a Word file, what the file holds, and the Word file read back as Markdown; then Markdown notes turned
# into a PDF through Typst, with citations formatted by pandoc's citeproc.
needs("pandoc", "typst", "pdftotext")
files("note/note.tex", "note/refs.bib", "md/notes.md", "docx_parts.py")
run("""
pandoc note.tex --citeproc --bibliography=refs.bib --number-sections -o note.docx
python3 docx_parts.py note.docx
pandoc note.docx -t markdown --wrap=none
pandoc notes.md --citeproc --bibliography=refs.bib --pdf-engine=typst -o notes.pdf
pdfinfo notes.pdf | grep -E 'Creator|Tagged|Pages'
pdftotext notes.pdf - | grep 'Rao'
""", env={"SOURCE_DATE_EPOCH": "1767225600"})

$ pandoc note.tex --citeproc --bibliography=refs.bib --number-sections -o note.docx
$ python3 docx_parts.py note.docx
16 parts, among them: word/document.xml, word/numbering.xml, word/styles.xml, word/footnotes.xml, word/comments.xml, word/settings.xml, word/fontTable.xml, word/webSettings.xml
paragraph styles: Author 1, Bibliography 1, BodyText 1, Compact 6, FirstParagraph 5, Heading1 2, TableCaption 1, Title 1
equations (m:oMath): 3
$ pandoc note.docx -t markdown --wrap=none
# 1 Method

Two gauges logged the rain for *two weeks*[^1] (Rao and Kulkarni 2025). The daily mean of roof $r$ is

$${\bar{x}}_{r} = \frac{1}{n}\sum_{i = 1}^{n}x_{r,i}$$

and Table [1](#_8ddf680cae075dc657fb546cd9f3f5f778cab5e) lists the totals.

- Block A faces east.

- Block B faces west.

# 2 Result

  -------------------------------------------------------------
  Roof                           Rain (mm)
  ------------------------------ ------------------------------
  Block A                        83.0

  Block B                        76.5
  -------------------------------------------------------------

  : Totals for two weeks

Equation [\[eq:mean\]](#_908f492889e3ae7c115bf8154b176a12b022e3a) gives ${\bar{x}}_{A} = 5.93$ mm a day, as Section [1](#_09ee8443ab55a3f30fdd88acf82bf7f547c29d1) explains.

Rao, Asha, and Dev Kulkarni. 2025. "Low-Cost Rain Gauges on Flat Roofs." *Journal of Example Hydrology* 12: 40--52. <https://doi.org/10.5555/jeh.2025.040>.

[^1]: Readings were taken at 8 a.m.
$ pandoc notes.md --citeproc --bibliography=refs.bib --pdf-engine=typst -o notes.pdf
$ pdfinfo notes.pdf | grep -E 'Creator|Tagged|Pages'
Creator:         Typst 0.15.1
Tagged:          yes
Pages:           1
$ pdftotext notes.pdf - | grep 'Rao'
Two gauges logged the rain on two roofs for two weeks (Rao and Kulkarni 2025). Each daily
Rao, Asha, and Dev Kulkarni. 2025. “Low-Cost Rain Gauges on Flat Roofs.” Journal of Example

The Word file had 16 parts and Word’s own paragraph styles, Title, Author, Heading1 for the two sections and TableCaption for the caption, and its three formulas were m:oMath equations. Read back as Markdown, the structure had survived, and so had the faults: the reference to the equation still read [eq:mean], and the citation had become (Rao and Kulkarni 2025), because citeproc formats in the Chicago author-date style unless it is given a style. --pdf-engine=typst made a one-page PDF whose pdfinfo said Creator: Typst 0.15.1 and Tagged: yes.

MySmartCoPilot’s Markdown to Word and Markdown to PDF tools do the last two routes in your browser, for notes that need no numbering.

HTML from LaTeX and from Typst keeps different things

A web page needs HTML, and both systems can write it. For LaTeX the tool is TeX4ht; its CTAN entry calls make4ht a simple build system for it, but make4ht stopped here because a module of the LuaXML package was missing from this lesson’s TinyTeX, so the session uses TeX4ht’s older driver, htlatex, which needs nothing more. Typst has an HTML export of its own, which its reference says is experimental, needs --features html and should not be used for production. The session converts the corrected note both ways and counts what each HTML file kept:

"""Counts what an HTML file kept of a document: headings, MathML formulas, images, inline styles and style blocks.

    python3 html_count.py note.html

The counts come from the tags in the file, read with Python's HTML parser.
"""

import sys
from html.parser import HTMLParser


class Counter(HTMLParser):
    def __init__(self):
        super().__init__()
        self.headings, self.maths, self.images, self.styles, self.blocks, self.title = [], 0, [], 0, 0, ""
        self._in = None

    def handle_starttag(self, tag, attrs):
        attrs = dict(attrs)
        if "style" in attrs:
            self.styles += 1
        if tag in ("h1", "h2", "h3", "h4"):
            self._in = tag
            self.headings.append([tag, ""])
        elif tag == "title":
            self._in = "title"
        elif tag == "math":
            self.maths += 1
        elif tag == "img":
            self.images.append(attrs.get("src", ""))
        elif tag == "style":
            self.blocks += 1

    def handle_endtag(self, tag):
        if tag == self._in:
            self._in = None

    def handle_data(self, data):
        if self._in == "title":
            self.title += data
        elif self._in:
            self.headings[-1][1] += data


with open(sys.argv[1], encoding="utf-8", errors="replace") as file:
    counter = Counter()
    counter.feed(file.read())
print(f"title: {counter.title.strip()}")
for tag, text in counter.headings:
    print(f"{tag}: {' '.join(text.split())}")
print(f"MathML formulas: {counter.maths}, images: {', '.join(counter.images) or 'none'}")
print(f"style attributes: {counter.styles}, style blocks: {counter.blocks}")

from session import files, needs, run

# HTML from both sources. For note.tex, after a latexmk build that writes the .bbl it needs: make4ht, which stops
# because a module of LuaXML is not installed in this TinyTeX, then TeX4ht's older driver htlatex. For the converted
# and corrected note: Typst's experimental HTML export. html_count.py then counts what each HTML file kept.
needs("pdflatex", "latexmk", "htlatex", "typst")
files("note/note.tex", "note/refs.bib", "note/main.typ", "note/body-fixed.typ=body.typ", "html_count.py")
run("""
latexmk -pdf -silent note.tex > /dev/null 2>&1
!make4ht note.tex 'mathml' > make4ht.log 2>&1
grep -o "module 'luaxml[a-z-]*' not found" make4ht.log
htlatex note.tex 'xhtml,mathml' > htlatex.log 2>&1
ls note*.html note*.svg note.css
python3 html_count.py note.html
typst compile --features html --format html main.typ note-typst.html 2>&1 | grep '^warning'
python3 html_count.py note-typst.html
""", env={"SOURCE_DATE_EPOCH": "1767225600"})

$ latexmk -pdf -silent note.tex > /dev/null 2>&1
$ make4ht note.tex 'mathml' > make4ht.log 2>&1
[exit status 1]
$ grep -o "module 'luaxml[a-z-]*' not found" make4ht.log
module 'luaxml-domobject' not found
$ htlatex note.tex 'xhtml,mathml' > htlatex.log 2>&1
$ ls note*.html note*.svg note.css
note.css
note.html
note0x.svg
note2.html
$ python3 html_count.py note.html
title: Rain on two roofs
h1: Rain on two roofs
h3: 1 Method
h3: 2 Result
h3: References
MathML formulas: 3, images: note0x.svg
style attributes: 11, style blocks: 0
$ typst compile --features html --format html main.typ note-typst.html 2>&1 | grep '^warning'
warning: html export is under active development and incomplete
warning: page set rule was ignored during HTML export
warning: align was ignored during HTML export
warning: align was ignored during HTML export
warning: rect was ignored during HTML export
$ python3 html_count.py note-typst.html
title: Rain on two roofs
h2: 1 Method
h2: 2 Result
h2: References
MathML formulas: 3, images: none
style attributes: 2, style blocks: 1

Typst listed what it ignored as warnings: the page settings, both align calls and the rect. TeX4ht kept the drawing as an SVG image. Its style attributes matter for any site whose security policy forbids inline styles, which must strip them or move them to classes.

A conversion is checked by its counts and its words

Compiling proves only that the new file is valid. Two checks prove that it says what the old one said. Count the parts that both should have, here with grep on note.tex and, for Typst, with typst eval and query(), which its reference describes as finding elements in the document; then compare the same sentences of both PDFs again after the fixes. This session compares pandoc’s body with the corrected one, then runs both checks:

= Method
<sec:method>
Two gauges logged the rain for #emph[two weeks]#footnote[Readings were
taken at 8 a.m.] @rao2025gauges. The daily mean of roof $r$ is
$ macron(x)_r = 1 / n sum_(i = 1)^n x_(r\,i) $<eq:mean> and
@tab:totals lists the totals.

- Block A faces east.

- Block B faces west.

= Result
<result>
#figure(
  align(center)[#table(
    columns: 2,
    align: (left,right,),
    table.header([Roof], [Rain (mm)],),
    table.hline(),
    [Block A], [83.0],
    [Block B], [76.5],
  )]
  , caption: [Totals for two weeks]
  , kind: table
  )
<tab:totals>

@eq:mean gives $macron(x)_A = 5.93$ mm a
day, as @sec:method explains.

// The TikZ picture of note.tex, which pandoc skipped, drawn again in Typst.
#rect(width: 2cm, height: 1cm, inset: 0pt)[#align(center + horizon)[Block A]]

from session import files, needs, run

# The converted body against the corrected one, then the corrected note checked: the numbered headings, display
# equations and tables of each source and document counted, and the same sentences of the two PDFs compared again.
needs("pandoc", "typst", "pdflatex", "latexmk", "pdftotext", "git")
files("note/note.tex", "note/refs.bib", "note/main.typ", "note/body-fixed.typ")
run("""
pandoc note.tex -t typst -o body.typ
!git diff --no-index -U0 body.typ body-fixed.typ
cp body-fixed.typ body.typ && typst compile main.typ note-typst.pdf
grep -c -E '^.section' note.tex
grep -c -E 'begin.equation' note.tex
grep -c -E 'begin.table' note.tex
typst eval 'query(heading.where(numbering: "1")).len()' --in main.typ
typst eval 'query(math.equation.where(block: true)).len()' --in main.typ
typst eval 'query(figure.where(kind: table)).len()' --in main.typ
latexmk -pdf -silent note.tex > /dev/null 2>&1
pdftotext note.pdf latex.txt && pdftotext note-typst.pdf typst.txt
grep -E 'daily mean|lists the totals|explains' latex.txt typst.txt
grep -c 'Block A' latex.txt typst.txt
""", env={"SOURCE_DATE_EPOCH": "1767225600"})

$ pandoc note.tex -t typst -o body.typ
$ git diff --no-index -U0 body.typ body-fixed.typ
diff --git a/body.typ b/body-fixed.typ
index 4acbeaa..b7d9cfd 100644
--- a/body.typ
+++ b/body-fixed.typ
@@ -6 +6 @@ $ macron(x)_r = 1 / n sum_(i = 1)^n x_(r\,i) $<eq:mean> and
-Table~@tab:totals lists the totals.
+@tab:totals lists the totals.
@@ -28,2 +28,5 @@ Table~@tab:totals lists the totals.
-Equation~#link(<eq:mean>)[\[eq:mean\]] gives $macron(x)_A = 5.93$ mm a
-day, as Section~@sec:method explains.
+@eq:mean gives $macron(x)_A = 5.93$ mm a
+day, as @sec:method explains.
+
+// The TikZ picture of note.tex, which pandoc skipped, drawn again in Typst.
+#rect(width: 2cm, height: 1cm, inset: 0pt)[#align(center + horizon)[Block A]]
[exit status 1]
$ cp body-fixed.typ body.typ && typst compile main.typ note-typst.pdf
$ grep -c -E '^.section' note.tex
2
$ grep -c -E 'begin.equation' note.tex
1
$ grep -c -E 'begin.table' note.tex
1
$ typst eval 'query(heading.where(numbering: "1")).len()' --in main.typ
2
$ typst eval 'query(math.equation.where(block: true)).len()' --in main.typ
1
$ typst eval 'query(figure.where(kind: table)).len()' --in main.typ
1
$ latexmk -pdf -silent note.tex > /dev/null 2>&1
$ pdftotext note.pdf latex.txt && pdftotext note-typst.pdf typst.txt
$ grep -E 'daily mean|lists the totals|explains' latex.txt typst.txt
latex.txt:Two gauges logged the rain for two weeks 1 [1]. The daily mean of roof r is
latex.txt:and Table 1 lists the totals.
latex.txt:Equation (1) gives x̄A = 5.93 mm a day, as Section 1 explains.
typst.txt:Two gauges logged the rain for two weeks1 [1]. The daily mean of roof 𝑟 is
typst.txt:and Table 1 lists the totals.
typst.txt:Equation 1 gives 𝑥̄𝐴 = 5.93 mm a day, as Section 1 explains.
$ grep -c 'Block A' latex.txt typst.txt
latex.txt:3
typst.txt:3

The fixes were two hunks: Table~ and Section~ removed before their references, @eq:mean in place of the link, and the Block A box drawn again with rect. Both sides then counted 2 numbered sections, 1 display equation and 1 table, the sentences read Table 1, Equation 1 and Section 1, and Block A appeared 3 times in each text. Follow the checks in order:

What remains is expected and nameable: LaTeX’s \eqref printed (1) where Typst’s reference printed 1, the maths letters were extracted as r and 𝑟, which Unicode’s NFKCNFKC: A Unicode normalisation form that replaces compatibility variants, such as mathematical italic letters, superscript digits and ligatures, with their plain characters, so that text compares and searches as people type it. More in “Inline and display maths” form turns into the same letter, and the two bibliographies follow different styles.

Practise: list what a conversion lost

The sentence checks above compare by eye. A short function does it for every word of two extracted texts, and with normalising on it ignores the maths letters that differ only in their code points. The sample tests are made from this lesson’s note, and the first hint is open:

def word_changes(before, after, normalise=False):
    """Return (missing, extra): the sorted words that `after` lacks and the sorted words it has in addition."""
    # Replace this line: count the words of each text, then subtract the counts both ways.
    return [], []

from wordcheck import word_changes


def test_doubled_word():
    """finds the word the conversion doubled"""
    assert word_changes("and Table 1 lists the totals.", "and Table Table 1 lists the totals.") == ([], ["Table"])


def test_reference_as_text():
    """finds an equation number that became a label's name"""
    before = "Equation (1) gives 5.93 mm a day."
    after = "Equation [eq:mean] gives 5.93 mm a day."
    assert word_changes(before, after) == (["(1)"], ["[eq:mean]"])


def test_repeats_counted():
    """counts repeated words, so a drawing's lost label shows"""
    before = "Block A faces east.\nBlock A 83.0\nBlock A"
    after = "Block A faces east.\nBlock A 83.0"
    assert word_changes(before, after) == (["A", "Block"], [])


def test_order_and_spaces():
    """ignores the order of words and the kind and amount of whitespace"""
    assert word_changes("mean of roof r", "roof   r\nmean of") == ([], [])


def test_maths_letters():
    """treats a mathematical italic letter as a different word unless asked to normalise"""
    before = "The daily mean of roof r is"
    after = "The daily mean of roof \U0001d45f is"
    assert word_changes(before, after) == (["r"], ["\U0001d45f"])
    assert word_changes(before, after, normalise=True) == ([], [])


def test_sorted_lists():
    """returns both lists sorted"""
    assert word_changes("b a c", "d c e") == (["a", "b"], ["d", "e"])


def test_empty_texts():
    """gives two empty lists for two empty texts"""
    assert word_changes("", "") == ([], [])

The self-check below covers the six ideas, starting with the one you have answered least well, and explains every choice as soon as you make it:

Interview questions

Warm-up (fresher to 2–5 years): What does pandoc keep and lose when it converts a LaTeX file to Typst?

pandoc reads the LaTeX into its own document tree and writes Typst from the tree, so it keeps what the tree can hold: headings with their labels, paragraphs, emphasis, lists, footnotes, tables, maths and citations. It applies custom macros instead of keeping them, so \mean{x} arrived as macron(x). It drops what the tree cannot hold: in this lesson’s note it skipped \maketitle, \centering, a TikZ picture and \bibliographystyle, and said so only with --verbose. It also writes things that compile and read wrongly: Table~\ref{…} became “Table Table 1”, because a Typst reference prints its own word, and \eqref became a link whose text was the label’s name. So a converted file is a draft: compile it, count its headings, equations and tables against the original, and compare the text of both PDFs.

Two harder questions build on this: a thesis chapter that must move to Typst with all its references working, and a lab manual needed as a PDF, a Word file and a web page. Sketch an answer to each, then check it against its first step:

Key takeaways

• pandoc reads a document into its own tree and writes the tree again, so whatever the tree cannot hold, such as a TikZ picture, is dropped, and custom macros arrive expanded.
• Converted to Typst, the note compiled but printed “Table Table 1”, “Section Section 1” and “Equation [eq:mean]”, and had lost its drawing; --verbose named only the skipped commands.
• Maths converts formula by formula into Typst’s own syntax; check each formula on the page once, and use MiTeX while a file must keep LaTeX maths.
• Word files keep headings, tables and maths as Word’s own styles and equations, citeproc formats citations in Chicago author-date unless given a style, and Markdown reaches a tagged PDF through --pdf-engine=typst.
• TeX4ht kept the drawing as SVG and wrote inline styles; Typst’s experimental HTML export dropped the drawing and the title block and said so in its warnings.
• A conversion is checked by counting its parts on both sides and comparing the text of both PDFs until every difference left has a reason.