CV

Several CV variants built from one shared content database.

Variant	Driver	Published at
Academic / postdoc	main-postdoc.tex	main.pdf
Quant / data science	main-quant.tex	main-quant.pdf

Design

main-postdoc.tex      driver: prose, WHICH entries appear, in WHAT order
main-quant.tex        driver
style/
  cvkit.sty           machinery only: the record store and \use... commands.
                      Knows nothing about content or about which variants exist.
  variant-postdoc.sty the academic LOOK + academic renderers
  variant-quant.sty   the industry LOOK + compact renderers
content/              the facts. Definitions only; these files print nothing.

The content files behave like a small .bib: they define entries and output nothing. A driver cites the ones it wants.

% content/publications.tex
\definepub{ntk}
  {Quantitative Understanding of PDF Fits and their Uncertainties}
  {\me, L. Del Debbio, R. Kenway}
  {Published in: \textit{Eur.Phys.J.C 86 (2026) 6, 722}}{2512.24116}

% main-quant.tex -- selection and order live here, not in the content
\cvsection{Selected Publications}
\begin{cvlist}
  \item[] \footnotesize Six peer-reviewed publications since 2024. [...] \normalsize
  \usepub{ntk}
  \usepub{runii}
\end{cvlist}

Three consequences worth keeping:

• Content never mentions a variant. There are no \if-style switches in content/. Adding a fourth CV touches zero content files.
• Selection is a driver decision. Showing a different subset of papers is editing a list of \usepub lines, not adding tags or filter rules.
• Look is a variant decision. Each style/variant-*.sty owns its own geometry, fonts, headings and renderers, so two CVs can be genuinely different-looking documents rather than one template with tweaks.

Registers

The one case where a fact needs different wording per audience is handled by registers, which are tones of voice rather than variant names:

\definexp{nnpdf}{NNPDF Collaboration}{Developer}{since 11/2023}
\xpdesc{nnpdf}{academic}{Contribute to the maintenance and development of...}
\xpdesc{nnpdf}{industry}{Maintain and extend a production Python/C++ codebase...}

The driver asks for the tone it wants: \usexp{nnpdf}{industry}. Organisation, role and dates still exist in exactly one place.

Available commands

Define (in content/)	Cite (in a driver)
\definepub{key}{title}{authors}{venue}{arxiv}	\usepub{key}
\definexp{key}{org}{role}{dates} + \xpdesc{key}{register}{prose}	\usexp{key}{register}
\defineedu{key}{degree}{grade}{dates}{institution}{supervisor}{detail}	\useedu{key}
\definetalk{key}{title}{url}{date}{event}	\usetalk{key}
\defineaward{key}{name}{dates}{desc}	\useaward{key}
\defineskill{key}{label}{body}	\useskill{key}

Citing a key that does not exist is a hard build error, not a silent omission.

Adding a variant

1. Copy a driver, e.g. main-industry.tex.
2. Copy a style/variant-*.sty and give it whatever look you want.
3. Add industry to matrix.variant in both workflow files.

No content files change.

Building locally

latexmk -pdf main-postdoc.tex
latexmk -pdf main-quant.tex

Checking a CV

cicd/spelling.sh main-quant.pdf    # spelling + leftover [TODO: ...] placeholders
cicd/Length.sh   main-quant.pdf    # fails above two pages

Both run in CI on every push and pull request, for both variants.

The spell checker reads the rendered text of the PDF rather than the LaTeX source. That is deliberate: entry keys, command names, lengths and arXiv IDs are not words, and checking the source meant every one of them had to be whitelisted in dict/words before CI would go green. Checking the PDF means dict/words holds domain vocabulary only, and adding a \definepub{key} never touches it.

Placeholders render in red as [TODO: ...] and are a hard failure, so a CV with an unresolved placeholder cannot pass.