latex-tables

LaTeX Tables

Create publication-quality tables using booktabs conventions for academic manuscripts.

Table Style Rules

• No vertical rules. Use \toprule, \midrule, \bottomrule from booktabs only.
• Proper spacing. Use \addlinespace for logical row grouping, never \\[6pt] hacks.
• Left-align text columns, right-align numeric columns. Use S columns from siunitx for decimal alignment.
• Minimal ink. Remove all unnecessary lines and borders.

Supported Table Types

Results Table

• Rows = methods/models, columns = metrics
• Bold the best result per column (\textbf{})
• Statistical significance markers: * (p<0.05), ** (p<0.01), *** (p<0.001)
• Add table footnote explaining marker meaning via \tnote{}

Comparison Table

• Side-by-side feature or property comparison
• Checkmarks (\checkmark) and dashes (---) for binary features
• Grouped column headers via \multicolumn{}

Summary Statistics Table

• Descriptive stats: mean, SD, median, range, N
• Use siunitx S column type for aligned decimals
• Horizontal grouping with \cmidrule{}

Multi-Page Table

• Uses longtable package for tables exceeding one page
• Repeated header row on each page with \endhead
• Continued footer with \endfoot

Landscape Table

• Uses sidewaystable from rotating package for wide tables
• Automatically rotates 90 degrees on the page

Multi-Column and Multi-Row Support

• \multicolumn{n}{alignment}{content} for spanning columns
• \multirow{n}{width}{content} for spanning rows (requires multirow package)
• Combine with \cmidrule{start-end} for partial horizontal rules under grouped headers

Data Ingestion

From CSV

1. User provides path to .csv file
2. Parse headers and rows automatically
3. Detect numeric vs text columns for alignment
4. Apply booktabs formatting
5. Place output in tables/ directory

From JSON

1. Accept array-of-objects JSON format
2. Keys become column headers
3. Values populate rows
4. Same auto-detection and formatting as CSV

From Inline Data

1. User describes data in natural language or pastes raw text
2. Parse into structured rows/columns
3. Confirm structure with user before generating LaTeX

Required LaTeX Packages

\usepackage{booktabs}       % Professional table rules
\usepackage{multirow}       % Row spanning
\usepackage{siunitx}        % Decimal-aligned columns
\usepackage{rotating}       % sidewaystable
\usepackage{longtable}      % Multi-page tables
\usepackage{threeparttable} % Table notes and footnotes

Output Format

LaTeX Output

\begin{table}[htbp]
  \centering
  \caption{Comparison of model performance on benchmark dataset.}
  \label{tab:results}
  \begin{threeparttable}
    \begin{tabular}{l S[table-format=2.1] S[table-format=2.1] S[table-format=1.3]}
      \toprule
      {Method} & {Accuracy (\%)} & {F1 (\%)} & {p-value} \\
      \midrule
      Baseline   & 78.3 & 75.1 & 0.042\tnote{*} \\
      Our Method & \textbf{85.7} & \textbf{83.2} & {---} \\
      \bottomrule
    \end{tabular}
    \begin{tablenotes}
      \item[*] p < 0.05 compared to Our Method.
    \end{tablenotes}
  \end{threeparttable}
\end{table}

Word-Compatible Output (planned, not implemented)

DOCX table emission is planned, not implemented. The shipped generator, templates/word/build-docx.js, produces a title page, numbered headings, paragraphs, and lists from sections/*.md; it emits no tables. The mapping below is the specification for that work, not a description of what runs today. Never tell a user a table was written into their DOCX.

• Generate the equivalent table structure with the docx library's Table
• Map booktabs rules to border styles (top/bottom thick, mid thin; Word has no \toprule)
• Preserve bold best results and alignment
• Include table caption and notes as separate paragraphs

Until that ships, deliver the LaTeX table and hand the user this mapping so they can place the table in Word themselves. examples/visualization-latex/latex-results-table.md shows the expected shape of that hand-off.

Journal Adaptation

• Read manuscript/config.yaml for target journal
• Adjust table width to single-column or double-column as required
• Apply journal-specific font size (e.g., \small, \footnotesize)
• Respect maximum table count limits if specified
• Use tabular* or tabularx for full-width tables when journal requires it

Style Presets

• Named presets (default, nature, ieee) are defined once in references/figure-styles.md. This skill does not duplicate those definitions; it reads that file for the concrete values.
• This skill consumes the preset's booktabs conventions, font size, rule weights, and caption placement, applying them to the tabular output generated here.
• Presets resolve in one precedence order, highest first: explicit Style: line > trigger phrase > journal inference from manuscript/config.yaml > default.
• Presets restyle only: they never alter data, values, or the "(synthetic, for demonstration)" labeling on placeholder data.

Style: line (accepted input)

The invocation may carry an explicit Style: line, which outranks every other selector:

Turn results.csv into a results table, bold the best per column.
Style: nature

• Style: nature, Style: ieee: apply that preset.
• Style: default, or no Style: line at all: the no-op path, exactly the current behavior described above (zero change). An omitted Style: line is never an error.
• Any other value: do not guess and do not improvise a preset. Say it is not defined, list the presets that are (default, nature, ieee), and ask which to use.

Trigger phrases and journal inference

• Trigger phrases that select a non-default preset when no Style: line is given: "nature style", "in Nature format", "for submission to ", "IEEE format".
• Journal inference: with neither a Style: line nor a trigger phrase, check manuscript/config.yaml for the target journal and map it to a preset family (Nature portfolio titles map to nature, IEEE titles map to ieee, anything else falls back to default).
• State which preset was applied and why.

Workflow

1. Determine table type from user request
2. Collect data (CSV, JSON, inline, or user description)
3. Confirm column headers, alignment, and grouping with user
4. Load references/figure-styles.md and resolve the style preset (explicit trigger phrase, or journal named/read from manuscript/config.yaml, or default); state which preset was applied and why
5. Generate LaTeX code with all required packages noted, styled per the resolved preset
6. Save to tables/table-<name>.tex
7. Validate compilation by running scripts/latex-compile.py (or latex-compile.sh on POSIX), which uses whichever TeX engine is installed (tectonic recommended, or latexmk/pdflatex from TeX Live, MiKTeX, or MacTeX)
8. If Word output is requested, say that DOCX table emission is planned and not implemented (see "Word-Compatible Output" above), then deliver the LaTeX table plus the docx-library mapping for the user to apply by hand

Alt Text

Alt text is a REQUIRED output of this skill, delivered with every table (and every preset variant). Write one or two sentences describing the DATA content: the rows and columns compared and the headline result (for example "Macro-AUROC for supervised vs self-supervised methods at 1%, 10%, and 100% labeled fractions; self-supervised leads at every fraction"). Never describe styling (booktabs rules, fonts); presets restyle only, so all variants of one table share ONE data description plus at most a one-clause style note. Do not begin with "Image of". See the Alt Text section of references/figure-styles.md. This matters when the table is exported as an image; a native LaTeX or DOCX table is read directly by assistive tech and needs no separate alt text.