LaTeX Writing Best Practices

This skill guides the creation of well-structured, semantically correct LaTeX documents following established best practices.

Core Principle: Semantic Markup

Use LaTeX environments that match the semantic meaning of the content, not just the visual appearance.

List Environments: When to Use What

Use description for Term-Definition Pairs

When you have labels followed by explanations, definitions, or descriptions, use the description environment:

\begin{description}
\item[Term] Definition or explanation of the term
\item[Label] Content associated with the label
\item[Property] Description of the property
\end{description}

NEVER do this:

\begin{itemize}
\item \textbf{Term:} Definition or explanation
\item \textbf{Label:} Content associated with label
\end{itemize}

Common Use Cases for description

• API parameters: \item[username] The user's login name
• Configuration options: \item[timeout] Maximum wait time in seconds
• Glossary entries: \item[LaTeX] A document preparation system
• Passes/Fails examples: \item[Passes] Correct implementation...
• Feature descriptions: \item[Auto-save] Automatically saves every 5 minutes

Exception: Pedagogical Meta-Commentary

Do not use a visible description list for instructor-facing pedagogical annotations such as \enquote{What varies}, \enquote{What stays invariant}, or sequencing rationale in educational materials. Those belong in \ltnote{...} via the didactic-notes skill. Use description only when the labeled content is part of the student-facing document itself.

Use itemize for Simple Lists

Use itemize when items are uniform list elements without labels:

\begin{itemize}
\item First uniform item
\item Second uniform item
\item Third uniform item
\end{itemize}

Use enumerate for Numbered Steps or Rankings

Use enumerate when order matters:

\begin{enumerate}
\item First step in the process
\item Second step in the process
\item Third step in the process
\end{enumerate}

Interleave Commentary Between Numbered Items (enumitem resume)

When enumerated items (survey/quiz questions, requirements, exam tasks) each need explanatory commentary, keep the \item text bare and put the commentary as prose between single-item environments, resuming the numbering with enumitem:

\begin{enumerate}
  \item \enquote{Have you programmed before this course?}
\end{enumerate}
Commentary motivating the question, referencing what it serves.
\begin{enumerate}[resume]
  \item \enquote{How do you plan to study in this course?}
\end{enumerate}
More commentary...

• Items sharing the same commentary share one environment.
• Requires \usepackage{enumitem}; in dual beamer/article builds load it in the article driver only (enumitem conflicts with beamer's list internals) and keep such lists in article-mode prose, which \mode* skips in the slides job.
• Anti-pattern: appending the commentary inside the \item after the question text — the question and its rationale then blur together.

Recognition Patterns

When reviewing or writing LaTeX, look for these patterns that indicate description should be used:

• \item \textbf{SomeLabel:} → Should use \item[SomeLabel]
• \item \emph{SomeLabel:} → Should use \item[SomeLabel]
• \item SomeLabel --- → Should use \item[SomeLabel]
• Lists where every item starts with bold/emphasized text

Fixing Common Anti-Patterns

Anti-Pattern: Bold Labels in Itemize

% INCORRECT
\begin{itemize}
\item \textbf{Passes:} \verb|\documentclass{article}|
\item \textbf{Fails:} No documentclass declaration
\end{itemize}

Correct: Description Environment

% CORRECT
\begin{description}
\item[Passes] \verb|\documentclass{article}|
\item[Fails] No documentclass declaration
\end{description}

Anti-Pattern: Manual Formatting Instead of Semantic Structure

% INCORRECT
\noindent\textbf{Configuration:} Set timeout to 30 seconds.\\
\textbf{Performance:} Optimized for large datasets.

Correct: Semantic Description

% CORRECT
\begin{description}
\item[Configuration] Set timeout to 30 seconds
\item[Performance] Optimized for large datasets
\end{description}

Literate Programming (.nw files)

CRITICAL: When writing LaTeX in literate programming files (.nw), use noweb's [[code]] notation for quoting code, not \texttt with manual escaping.

Use [[code]] Notation, Not \texttt{...\_...}

The noweb literate programming system provides special notation for code references that automatically handles special characters like underscores.

Anti-pattern: Manual underscore escaping with \texttt

% INCORRECT - in .nw files
The \texttt{get\_submission()} method calls \texttt{\_\_getattribute\_\_}.
We store \texttt{\_original\_get\_submission} in the closure.
The \texttt{NOREFRESH\_GRADES} constant defines final grades.

Correct: Use [[code]] notation

% CORRECT - in .nw files
The [[get_submission()]] method calls [[__getattribute__]].
We store [[_original_get_submission]] in the closure.
The [[NOREFRESH_GRADES]] constant defines final grades.

Why this matters:

• Automatically escapes special characters (underscores, backslashes, etc.)
• Makes source code more readable (no manual escaping)
• Follows literate programming conventions
• Prevents LaTeX errors from forgotten escapes
• Clearly distinguishes code from prose

When to Use [[code]] vs \texttt

Use [[code]] in .nw files for:

• Function and method names: [[get_submissions()]], [[__init__]]
• Variable names: [[_includes]], [[user_id]]
• Class names: [[LazySubmission]], [[Assignment]]
• Constants: [[NOREFRESH_GRADES]], [[MAX_RETRIES]]
• Module names: [[pickle]], [[Canvas]]
• Any identifier with underscores or special characters

Use \texttt in .nw files for:

• Short non-code technical terms without special characters
• File extensions: \texttt{.py}, \texttt{.nw}
• Simple commands without underscores

In regular .tex files (not literate programs):

• Use \texttt with proper escaping as [[...]] is not available
• Or use packages like minted or listings for code

Recognition Pattern for Review

When reviewing .nw files, look for these anti-patterns:

• \texttt{..._...} → Should use [[...]]
• \texttt{...__...} → Should use [[...]]
• \item[SOME\_CONSTANT behavior] → Either rephrase the label and put [[SOME_CONSTANT]] in the body, or wrap the constant in [[...]] inside the label — \item[ [[SOME_CONSTANT]] behavior]. If you take the second option, separate the ]] from the closing ] with at least one character (usually a space) so you do not produce three brackets in a row, which triggers a runaway-argument error.

Examples from Real Code

Documenting methods:

% INCORRECT
The \texttt{\_\_getstate\_\_} method excludes \texttt{\_original\_get\_submission}.

% CORRECT
The [[__getstate__]] method excludes [[_original_get_submission]].

Documenting constants:

% INCORRECT
\item[NOREFRESH\_GRADES behavior] Submissions with final grades...

% CORRECT
\item[Final grade policy] Submissions with final grades (A, P, P+, complete)
  are never refreshed, maintaining the [[NOREFRESH_GRADES]] policy.

Documenting attributes:

% INCORRECT
The decorator adds a \texttt{\_\_cache} dictionary and \texttt{\_\_all\_fetched} flag.

% CORRECT
The decorator adds a [[__cache]] dictionary and [[__all_fetched]] flag.

Additional Best Practices

Cross-References

• Always use \cref{...} (cleveref package) for all cross-references
• Never use \S\ref{...} or manually type section/figure prefixes
• Use descriptive labels: \label{sec:introduction} not \label{s1}
• Examples:
  • Sections: \cref{sec:background} → "Section 2.1"
  • Figures: \cref{fig:diagram} → "Figure 3"
  • Tables: \cref{tab:results} → "Table 1"
  • Multiple: \cref{sec:intro,sec:conclusion} → "Sections 1 and 4"

Anti-pattern: Manual prefixes

% INCORRECT
Section~\ref{sec:intro} shows...
\S\ref{sec:background} discusses...
Figure~\ref{fig:plot} demonstrates...

% CORRECT
\cref{sec:intro} shows...
\cref{sec:background} discusses...
\cref{fig:plot} demonstrates...

Why: The cleveref package automatically adds the correct prefix (Section, Figure, etc.) and handles pluralization, ranges, and language-specific formatting.

Citations

• Use proper citation commands (\cite, \citep, \citet) not manual references
• Never write [1] or (Smith 2020) manually

Quotations (csquotes package)

• Always use \enquote{...} for quotes, never manual quote marks
• Handles nested quotes automatically: \enquote{outer \enquote{inner} quote}
• Language-aware: Swedish uses »...« or "...", English uses "..." or '...'
• For block quotes, use \begin{displayquote}...\end{displayquote}

Anti-pattern: Manual quotes

% INCORRECT
"This is a quote"
``This is a quote''
'single quotes'

Correct: Use csquotes

% CORRECT
\enquote{This is a quote}
\enquote{outer \enquote{inner} quote}

Why: Manual quote marks don't adapt to language settings and can cause typographical inconsistencies. The csquotes package handles all quote styling correctly based on document language.

Cited quotation vs. attributed paraphrase

The command depends on whether the text is verbatim from the source:

• Verbatim text reproduced from a source → use the integrated cited-quotation commands so the quotation carries its own reference; do not write \enquote{...} plus a detached \cite/\autocite:
  • \textcquote[⟨prenote⟩][⟨postnote⟩]{key}{text} — inline
  • \blockcquote[⟨postnote⟩]{key}{text} — block (long quotes)
• Paraphrase, reconstructed dialogue, or a recounted example → it is not a quotation. Attribute it integrally with \textcite/\autocite, and keep any reported words in plain \enquote.
• Plain \enquote{...} otherwise: scare quotes, words-as-mention, and back-references to an already-cited quotation.

% INCORRECT — verbatim source text as \enquote + detached citation
The study reports \enquote{much higher (63\%)} results \autocite[p.~167]{NCOL}.
% CORRECT — verbatim quotation carries its own citation
The study reports \textcquote[p.~167]{NCOL}{much higher (63\%)} results.

% CORRECT — a paraphrased/recounted example is attributed, NOT \textcquote'd
\Textcite[pp.~24--25]{NCOL} reports a study in which a child answers
\enquote{five} for one hand and \enquote{ten} for the other.

Crucial caveat: \textcquote/\blockcquote assert the braced text is verbatim from the source. Verify the words actually appear there before wrapping them; do not use these commands for paraphrase, reconstructed dialogue, or words-as-mention — attribute those with \textcite/\autocite instead. (See the backing-claims skill — read the source and verify before you cite.)

Emphasis

• Never use ALL CAPITALS for emphasis in running text
• Use \emph{...} to emphasize words or phrases
• For strong emphasis, use \textbf{...} or nested \emph{\emph{...}}
• Let LaTeX handle the typographic styling

Anti-pattern: ALL CAPITALS for emphasis

% INCORRECT
This is VERY important to understand.
We must do this NOW before moving forward.
The BENEFITS of classes are clear.

Correct: Semantic emphasis

% CORRECT
This is \emph{very} important to understand.
We must do this \emph{now} before moving forward.
The \emph{benefits} of classes are clear.

Why: ALL CAPITALS in running text is considered shouting and poor typography. It's harder to read and looks unprofessional. Use \emph{...} to provide semantic emphasis, and LaTeX will render it appropriately (typically as italics, but this can be configured based on context and document style).

Exception: Acronyms and proper names that are conventionally written in capitals (e.g., NASA, USA, PDF) are fine and should not be emphasized.

Floats: Figures and Tables

Core principle: An image is not a figure, but a figure can contain an image. Use proper figure and table environments with captions and labels.

Using sidecaption (memoir class)

When using the memoir document class, prefer sidecaption over traditional \caption for better layout and accessibility:

For figures:

\begin{figure}
  \begin{sidecaption}{Clear description of image content}[fig:label]
    \includegraphics[width=0.7\textwidth]{path/to/image}
  \end{sidecaption}
\end{figure}

For tables:

\begin{table}
  \begin{sidecaption}{Description of table content}[tab:label]
    \begin{tabular}{lcc}
      \toprule
      Header1 & Header2 & Header3 \\
      \midrule
      Data1 & Data2 & Data3 \\
      \bottomrule
    \end{tabular}
  \end{sidecaption}
\end{table}

Key features of sidecaption:

• Caption placement: Caption appears alongside the content (figure/table), not above or below
• Space efficiency: Better use of page width, especially for narrow images/tables
• Improved readability: Caption and content are visually connected
• Accessibility: Clear semantic separation between caption text and content
• Consistent labeling: Label goes in optional second argument [fig:label]

When to use sidecaption:

• Medium-sized figures/tables that don't need full text width
• Educational materials where caption-content proximity aids understanding
• Documents using memoir class (Beamer presentations with memoir features)

Traditional caption alternative:

\begin{figure}
  \centering
  \includegraphics[width=0.7\textwidth]{path/to/image}
  \caption{Description of image content}
  \label{fig:label}
\end{figure}

Use traditional \caption + \label when:

• Not using memoir class
• Figure/table spans full text width
• Caption naturally belongs below content (e.g., wide tables)

Referencing figures and tables

Always use \cref{fig:label} (cleveref package) to reference figures and tables:

As shown in \cref{fig:memory-hierarchy}, secondary memory is slower but non-volatile.

The results in \cref{tab:benchmark} demonstrate...

Never hard-code references like "Figure 1", "Table 3.2", or use manual prefixes like Figure~\ref{fig:label}—let cleveref handle both numbering and prefixes automatically.

Verbatim and Code

• Use minted (preferred, syntax-highlighted; needs -shell-escape) or listings for code blocks
• Use \verb or \mintinline for inline code snippets
• Never paste code as normal text
• minted v3 (TeX Live 2024+) gotchas — see references/minted-v3-and-floats.md (search: outputdir, Pygments lexer): the [outputdir=...] package option is removed (now an error; load plain \usepackage{minted}), and a minted/\mintinline language argument split across a line break fails with Pygments lexer " python" is unknown. Keep each minted invocation's arguments on one line; wrap prose between \mintinline{...}{...} calls, never inside one.

Debugging ! LaTeX Error: Float(s) lost.

This fatal, location-less error (reported at \end{document}) in a dual beamer/article didactic+memoir build is typically a citation/footnote inside a slide-only \begin{frame}<presentation> frame (it suppresses output but still executes the body, so the margin footnote gets orphaned) — not citations in ordinary frames, which are fine. Run a bare pdflatex -halt-on-error pass to see the real error past latexmk/biber noise, and diagnose by minimal reproduction. Fix: wrap the slide-only frame in \mode<presentation>{...} (or filter the citation with \only<presentation>{...}). Caveat: the \mode<presentation>{...} wrap cannot be used on a [fragile] frame — the group's closing } falls outside the verbatim .vrb scan and the slides build then fails with ! Extra }, or forgotten \endgroup. For a fragile slide-only frame, drop [fragile] if it is not actually needed (no verbatim/minted/listings inside), otherwise keep \begin{frame}<presentation>[fragile] and guard each citation with \only<presentation>{\autocite{...}}. Full diagnosis: references/minted-v3-and-floats.md and the didactic-notes skill's references/footnotes-and-citations.md.

Dual beamer/article builds (one source, two outputs)

For papers compiled twice from the same content files (memoir + beamerarticle article AND a beamer slide deck, content files starting with \mode*), follow references/dual-beamer-article.md (search: pyblock frame, ProvideSemanticEnv, restatable, biber root). The four rules, in brief:

• Every PythonTeX pyblock goes inside a \begin{frame}[fragile] that renders in both modes — out-of-frame pyblocks are skipped by the slides job, desynchronising the shared PythonTeX session data (symptoms: swapped outputs, ?? PythonTeX ??).
• Research questions and hypotheses go in semantic, cleveref-labelled environments (didactic's question; declare e.g. hypothesis via \ProvideSemanticEnv after \DeclareTranslation), stated inside \begin{restatable}{question}{tag} (thm-restate) and restated with \tag* — with a beamer shim in the shared preamble; reference via \cref{rq:*}, never literal "RQ1"/"H2".
• Slide-only frames: \mode<presentation>{...} wrap (fragile caveat above); article-only prose sits outside frames; don't let slide bullet lists duplicate adjacent article prose.
• Biber reruns happen from the project root (biber ltxobj/<jobname>), never from inside the output directory.

Encoding and Fonts (engine-dependent)

Unicode/font build failures have opposite fixes on pdfLaTeX vs XeLaTeX/LuaLaTeX, so identify the engine first (grep 'This is' file.log). Full guidance in references/unicode-and-fonts.md (search: DeclareUnicodeCharacter, iftex, newunicodechar, fontenc, pdffonts):

• pdfLaTeX — ! LaTeX Error: Unicode character … (U+XXXX) not set up — a non-ASCII byte with no mapping (reversed quotes, dash variants, ≈, box-drawing). Fix in the preamble with \usepackage[utf8]{inputenc} then \DeclareUnicodeCharacter{XXXX}{...}, not by editing the source. (inputenc is what defines \DeclareUnicodeCharacter portably; without it you may get Undefined control sequence.)
• XeLaTeX/LuaLaTeX — the "not set up" error does not occur (native UTF-8), and \DeclareUnicodeCharacter is undefined there (inputenc is ignored). Guard pdfLaTeX-only mappings with \ifpdftex … \fi (iftex package); remap a glyph with \newunicodechar instead.
• pdfLaTeX — code renders as proportional serif (underscores extract as ˙, quotes turn curly) — a T1-only monospace font (e.g. Bera Mono) cannot select without \usepackage[T1]{fontenc}. Load it (inside the \ifpdftex branch).
• Build systems switch engines. latexmk -pdf vs -xelatex, and Makefile rules may be tangled from a .nw, so a make/submodule update can flip the engine — re-check the .log banner, don't assume.
• Diagnose, do not guess: pdffonts -f N -l N file.pdf shows which fonts a page embeds; pdftotext -f N -l N reveals the encoding via glyph→Unicode mapping.

Paths

• Always use forward slashes in paths: figures/diagram.pdf not figures\diagram.pdf
• Use platform-independent path specifications

Workflow for Writing LaTeX

When writing or editing LaTeX content:

1. Check file type: Are you in a .nw literate programming file? If yes, use [[code]] notation, not \texttt{...\_...}
2. Identify content structure: Is this a list of uniform items or term-definition pairs?
3. Choose semantic environment: Match the environment to the content meaning
4. Use proper commands: Leverage LaTeX's semantic commands rather than manual formatting
5. Verify cross-references: Ensure labels and references are descriptive and correct
6. Check for anti-patterns: Review for \textbf{Label:} in itemize, \texttt{..._...} in .nw files

When Reviewing LaTeX Code

Check for these common issues:

• Lists using \textbf{Label:} instead of description environment
• Hard-coded numbers instead of \ref
• Manual cross-reference prefixes (\S\ref, Section~\ref, Figure~\ref) instead of \cref
• Manual citation formatting instead of \cite commands
• Manual quotes ("...", '...', `...`) instead of \enquote{...}
• Verbatim source text written as \enquote{...} + a detached \cite/\autocite instead of \textcquote/\blockcquote (and, conversely, \textcquote/\blockcquote wrapping paraphrase that is not verbatim)
• Images without figure environment
• Code without proper formatting (listings/verbatim)
• Windows-style backslashes in paths
• Engine-specific Unicode/font setup not guarded by iftex — e.g. \DeclareUnicodeCharacter (pdfLaTeX-only) used unconditionally, which errors under XeLaTeX/LuaLaTeX (see references/unicode-and-fonts.md)
• pdfLaTeX: non-ASCII characters without a \DeclareUnicodeCharacter mapping, or \DeclareUnicodeCharacter used without \usepackage[utf8]{inputenc}
• pdfLaTeX: monospace/code rendering as proportional serif — missing \usepackage[T1]{fontenc} for a T1-only font such as Bera Mono
• Dual beamer/article build: pyblock outside a both-modes \begin{frame}[fragile] (desynchronises PythonTeX sessions), or literal "RQ1"/"H2" in prose instead of \cref to a semantic question/hypothesis environment (see references/dual-beamer-article.md)

Additional checks for .nw literate programming files:

• \texttt{..._...} or \texttt{...__...} → Should use [[...]] notation
• Description labels with underscores \item[FOO\_BAR behavior] → Use better label + [[FOO_BAR]] in text
• Any manually escaped underscores in code references → Use [[...]] instead

Beamer: Presentation vs Article Mode

When creating Beamer presentations that also generate article versions, use \mode<presentation> and \mode<article> to provide appropriate content for each format.

When to Split Content

Verbose text environments: Semantic environments (definition, remark, example, block) with more than 2-3 lines of prose are too verbose for slides.

Solution: Provide concise versions for presentations and full explanations for articles:

\begin{frame}
  \mode<presentation>{%
    \begin{remark}[Key Point]
      \begin{itemize}
        \item Concise point 1
        \item Concise point 2
        \item Concise point 3
      \end{itemize}
    \end{remark}
  }
  \mode<article>{%
    \begin{remark}[Key Point]
      Full explanatory paragraph with detailed reasoning and context
      that would overwhelm a slide but provides value in written form.

      Additional paragraphs can elaborate on nuances and implications.
    \end{remark}
  }
\end{frame}

Visual Elements

Side-by-side layouts using \textbytext:

• Presentation: \textbytext (non-starred, column width, works in beamer frames)
• Article: \textbytext* (starred, fullwidth, better for printed documents)

\begin{frame}
  \mode<presentation>{%
    \textbytext{Left content}{Right content}
  }
  \mode<article>{%
    \textbytext*{Left content}{Right content}
  }
\end{frame}

Principle

Slides need visual clarity and conciseness (bullets, short phrases). Articles can provide depth and explanation (full sentences, paragraphs). Design content appropriate for each medium.

Interactive Questions with Mentipy

For slide decks or educational articles that should embed live polls, use mentipy.latex from a PythonTeX block or another Python block. Prefer mc, open_text, scale, and word_cloud for student-facing questions, then run mentipy serve to publish the poll and recompile after URL changes so refreshed QR codes are embedded. Use layout="slide" for Beamer slides and layout="article" or layout="article+qr" for handouts.

Standard Preamble

For new LaTeX documents, use the standard preamble from references/preamble.tex. Copy it verbatim to your project's doc/preamble.tex and include it with \input{preamble} after \documentclass.

Document structure:

\documentclass[a4paper,oneside]{memoir}
\input{preamble}

\title{Document Title}
\author{Author Name}
\date{\today}

\begin{document}
\frontmatter
\maketitle
\tableofcontents

\mainmatter
% Content here

\backmatter
\end{document}

The standard preamble provides:

• Language support, bibliography, mathematics, and quotations
• Code highlighting and noweb support
• Cross-references and tables
• Common utilities such as enumitem, acro, and siunitx

This ensures consistent formatting across all documents and projects.

Remember

LaTeX is a document preparation system based on semantic markup, not a word processor. The goal is to describe what content is, not how it should look. Let LaTeX handle the formatting based on the semantic structure you provide.