Lucid CV

Lucid CV is a LuaLaTeX class for creating modern, ATS-friendly resumes (CVs). It combines a design inspired by PlushCV with a custom back-end supporting extensive customization, single- and two-column layouts, and, above all, designed and tested compatibility with common PDF parsers and text extractors.

---

Author: Patrik Goldschmidt (<goldschmidt.patrik@gmail.com>)
Maintainer: Patrik Goldschmidt (<goldschmidt.patrik@gmail.com>)
Bug reports: https://github.com/xGoldy/lucidcv/issues

---

The class's name reflects its main goal: a resume that is lucid, i.e., clear and easy to understand for humans and machines alike. However, many modern resumes prioritize visual appeal at the expense of machine readability, causing Applicant Tracking Systems (ATSs) and PDF extraction tools to misinterpret or extract their content incorrectly. Lucid CV aims to tackle this issue and provide the best of both worlds: a resume that is visually appealing to humans while remaining machine-readable.

Colors, fonts, text sizes, spacing, layout, and other design elements can be customized without compromising document parsability.

This README covers the class's features, setup, usage, and ATS-compatibility testing and should be sufficient for most use cases. For detailed design information, see docs.md (fully AI-generated).

Note: The Overleaf and CTAN packages include the class, sample, and README, but not the test suite, extended documentation in resources/, and the build.sh script. To obtain these files, e.g., for ATS-compatibility testing described in Section 5, get the full project from the GitHub repository.

Table of Contents

1. Features
2. Examples
3. Getting Started
4. Usage
5. Testing for ATS Compatibility
6. License
7. Support the Project

1. Features

• Modern, clean design. Inspired by PlushCV with bundled Inter, Source Sans 3 and Source Code Pro fonts, a single accent color driving the palette, section heading rules, and Font Awesome 5 icons in the contact line. These choices aim to remain visually appealing without compromising machine readability (see ATS Compatibility as a Design Priority).
• PDF/A-2b and tagged PDF. The output conforms to the PDF/A-2b standard for long-term archiving and includes a tagged document structure: headings use H2/H3, the photo is a Figure with alternative text, while the title, author, and subject are populated automatically from the header. This improves interoperability with PDF readers, parsers, and other applications.
• Customizable without touching the class. Accent color in any color model, margins, column ratio, font sizes, and the spacing between items, entries, and sections can all be configured from the preamble (see Preamble).
• One content file, multiple layouts. The same body can be typeset in one or two columns, and at regular or compressed density, by changing class options only. Column breaks are safely ignored in the single-column layout, so there is no need to maintain separate versions of the content.
• Header with contact fields and photo. Predefined contact fields (email, phone, address, location, homepage, LinkedIn, GitHub, ORCID) display their full service prefix (e.g., github.com/xGoldy) and can be turned into clickable links. Custom fields with icons can also be declared. A circular photo can be added to the header and enabled or disabled with a single class option.
• Flexible entry system. Resumes are composed of entries supporting stacked or one-line layouts, flush-right dates, and configurable per-entry spacing. Entry hierarchies (e.g., one company, multiple job positions) are supported by indented groups of sub-entries.
• Extra elements. Predefined skill chips, reference entries, dashed dividers, and an optional Last update timestamp can be added.
• Additional tooling. In addition to Overleaf support, the project includes a local build script and a comprehensive ATS test suite (see Testing for ATS Compatibility).

ATS Compatibility as a Design Priority

A PDF that looks good on screen says nothing about what a parser can extract from it. A regular PDF stores only text glyphs and their positions. Word spaces, line breaks, and reading order must then be inferred by individual extraction tools, which may produce different results.

Because many resume templates are validated primarily by visual inspection, their extracted text might contain merged words, icons turned into random characters, interleaved columns, or dates detached from their entries. In contrast, Lucid CV was designed and validated against what extractors actually return, with each relevant design decision measured rather than assumed.

• PDF tagging: Real spaces between words. Instead of having PDF extractors infer the space between words, the class stores word gaps as actual space characters (with LuaLaTeX and PDF tagging enabled). As a result, extractors do not rely solely on horizontal positioning, which helps prevent them from merging words (e.g., Loremipsumdolor).
• Icons extract as whitespace. During PDF-to-text extraction, icon glyphs can be extracted as random letters or icon names (LINKEDIN, Github). The class makes every icon extract as a plain space using two independent mechanisms, because no single one covers all extractors. Icons are never assigned text labels either, since extracting words that were never written is worse than extracting no word at all. The noicons option removes icons entirely for legacy systems.
• Readable contact values. Contact fields display a self-explanatory link (e.g., linkedin.com/in/your_id rather than only your_id), allowing parsers to identify the value without relying on the adjacent icon. Contact items are separated by an explicit | delimiter.
• No header fusing. The first name, last name, and tagline are separated by invisible space characters, so the header does not extract as FirstnameLastnameTagline. These spaces have no visual effect.
• Dates stay with their entries. Many resume templates place dates at the far right of the page, leaving a large horizontal gap between the entry text and its date. Some parsers may interpret such a gap as a separate column and consequently detach the date from the entry. To tackle this issue, Lucid CV connects right-aligned dates to the entry title with a dotted leader.
• No transformed text. Headings are not set in small caps or forced uppercase, avoiding glyph transformations that may extract differently from the text as written.
• Chips that survive OCR. Skill chips (tags) use an en dash delimiter and a border tint and width chosen based on Optical Character Recognition (OCR) measurements, helping OCR-based extraction preserve the skill list correctly.
• Single column by default. Two-column layouts may extract out of order in parsers that read text according to its position on the page. This is a general limitation of multi-column PDFs and cannot be reliably eliminated within the document itself. Therefore, the onecolumn layout is the default and twocolumn must be enabled explicitly with the understanding that PDF parsability may be degraded.

All of the above is verified by the test suite described in Testing for ATS Compatibility, which checks whether every resume phrase is extracted intact and in the correct order by 9 extraction engines in 13 configurations, including OCR. The technical details and measurements behind these decisions are documented in the ATS and accessibility notes of docs.md.

1. Examples

The template ships with a sample resume, contents.tex, written for a senior software engineering role and compiled from main.tex out of the box. It serves as a reference rather than a showcase, using every body macro provided by the class, with comments explaining each one. The sample persona, Adrian Castellanos, is fictional, as are the employers and referees. The photo is also AI-generated.

For additional inspiration, three real-world resume variants are shown in the README on GitHub.

1. Getting Started

Ready to create your own resume with Lucid CV? This section covers the two supported ways to use the class: Overleaf and local compilation on Linux.

LaTeX Engine Requirements

Lucid CV is based on LuaLaTeX, a modern TeX engine that extends LaTeX with an embedded Lua interpreter. The required LaTeX format version is 2026-06-01 or newer (TeX Live 2026 with current updates).

An older format will only produce a warning, but the resulting PDF may contain extra vertical space below list environments due to changes in how newer LaTeX kernels handle lists.

Overleaf

The easiest way to get started is to use Overleaf, a cloud-based LaTeX editor. A free Overleaf account is required.

The resume template is available at the following link:

Open Lucid CV on Overleaf

After opening the link, select File -> Make a Copy to create your own editable copy. In your copy, verify the following settings under File -> Settings -> Compiler:

Setting	Value
Main document	main.tex
Compiler	LuaLaTeX
TeX Live version	2026 (or newer)

Local Compilation (Linux)

For local compilation or for running the PDF extraction tests that simulate ATS behavior (see Testing for ATS Compatibility), use the build.sh script available in the GitHub repository. The build script is not included in the CTAN and Overleaf packages. Alternatively, latexmk main.tex can be used.

Command	Action
./build.sh	Build main.tex into resume.pdf in the same directory.
./build.sh path/to.tex	Build a different source file.
./build.sh path/to.tex out.pdf	Build a different source file and write the PDF to the specified path.
./build.sh --opt NAME ...	Add a class option for this build without modifying the source file. Repeatable, e.g., --opt twocolumn --opt compressed.

The build script searches for a TeX Live installation in the following order: a local installation in .texenv/bin/* next to the script, a user installation in ~/texlive/*/bin/*, and the default install-tl location /usr/local/texlive/*/bin/*. Within each location, the newest year is preferred. If none is found, lualatex from PATH is used. If your TeX Live environment is located elsewhere, add its path to the TEXLIVE_BIN_GLOBS variable in the script or create a symbolic link named .texenv pointing to your installation, e.g., ln -s /opt/texlive/2026 .texenv.

The script compiles the PDF in two passes (or three when a bibliography processed by biber is included), removing the auxiliary and log files after a successful build without warnings. If compilation fails, the script reports the error, displays the relevant output, and preserves the log file for further analysis.

1. Usage

This section describes LaTeX document settings, commands, and macros provided by Lucid CV. The commands below are grouped into four sections: class options, preamble, header, and body, depending on where the command should be placed.

Section	Where	What it controls
Class Options	In the brackets of \documentclass[...]{lucidcv}	Document-wide switches: links, photo, icons, heading rules, column layout, and compression.
Preamble	Between \documentclass and \begin{document}	Accent color, margins, font sizes, spacing, and PDF metadata.
Header	After \begin{document}, before the cvbody environment	Name, tagline, photo, and contact details, typeset by \makecvheader.
Body	Inside the cvbody environment, ideally in a separate file loaded with \input	The CV contents: layout breaks, sections, entries, and elements.

The final Warnings and Errors subsection lists what the class reports in the compilation log and what it deliberately leaves unreported.

Class Options

Set on \documentclass[...]{lucidcv}. Recommended: 10pt, a4paper. All other extarticle options are also accepted.

Class-specific options include:

Option	Effect
hyper	Enable clickable links (recommended).
photo	Show the header photo declared by \cvphoto.
noicons	Do not show icons. May improve ATS parsing on legacy systems.
headingrule	Draw a line under section headings (the default, width = 0.4pt). Adjust with \setlength{\cvrulewidth}{width}.
noheadingrule	Do not typeset the rule under section headings.
onecolumn	Use one body column (the default). The safest choice for ATS parsing.
twocolumn	Use two body columns. Affects the body only; the header remains unchanged. Geometry-sorting extractors may interleave the columns, reducing reading-order reliability and overall parsability.
compressed	Tighten the document layout. Individual settings can be overridden through configuration knobs described in Preamble. Also sets \cventry to oneline; an explicit oneline= value overrides this default.

In every document using Lucid CV class, two additional commands should be placed at the very top of main.tex before the \documentclass command:

\DocumentMetadata{pdfstandard=A-2b,tagging=on,lang=en-US}
\tagpdfsetup{math/mathml/luamml/load=false}

\documentclass[10pt,a4paper,hyper,photo,onecolumn]{lucidcv}

\DocumentMetadata configures the document for PDF/A-2b, enables PDF tagging, and sets the PDF language to US English. For a CV written in another language, change lang accordingly. PDF/A provides a standardized, self-contained document format intended for long-term preservation, while PDF tagging provides a semantic structure that can improve accessibility and machine processing. Together, these features contribute to robust PDF parsing and ATS compatibility.

The \tagpdfsetup command disables the math-tagging module. Since a CV should normally contain no mathematical content, this has no effect on the final output; it only suppresses a warning that would otherwise be produced. If you need unicode-math to map mathematical content to PDF tags, comment out or remove this line.

Preamble

In addition to the class options described above, the final output of a document using Lucid CV can be customized in many ways in the preamble (the lines between \documentclass and \begin{document}). The available commands are described in the following table and can also be found in main.tex together with explanatory comments.

Values in parentheses apply when the compressed class option is used. Size and spacing values are set with \renewcommand, e.g., \renewcommand{\cvnamesize}{34}; the remaining commands are used directly.

Command	Default	Description
\cvaccent[model]{color}	1D76E2	Template accent color. Model is HTML by default, but any other xcolor model works, e.g., \cvaccent[RGB]{29,118,226}, or \cvaccent[named]{OliveGreen} for a named color.
\geometry{...}	1.2cm margins, columnsep=1.1cm	Page margins and the gap between columns.
\columnratio{ratio}	0.65	Width of the left column in a two-column layout. Accepts a single number in the range (0, 1).
\cvnamesize	38	Font size of the name in the header, in pt.
\cvtaglinesize	15	Font size of the tagline, in pt.
\nametaggap	0.1em	Extra space between the name and the tagline.
\tagcontactgap	0.25em	Extra space between the tagline and the contact line.
\cvsectionsize	16 (13)	Font size of section headings, in pt.
\cvsubsectionsize	11 (10)	Font size of subsection headings, in pt.
\cvitemsep	0.15em (0.1em)	Gap between list items.
\cvitemtopsep	0.2em (0.1em)	Gap between an entry line and its first list item.
\cvsectiongapratio	1.0 (0.7)	Scales the space above each section heading.
\cvrulegaptrim	0em (0.25em)	Reduces the space around the section heading rule; a larger value is tighter.
\cvtitle{title}	name – Curriculum Vitae	PDF title metadata.
\cvauthor{author}	name	PDF author metadata.
\cvsubject{subject}	tagline	PDF subject metadata; falls back to "Curriculum Vitae" when no tagline is set.
\cvkeywords{a, b, c}	empty	Comma-separated PDF keywords.

Header

The header is defined by commands placed after \begin{document} and before the cvbody environment. These commands specify the name, tagline, photo, and contact details. \makecvheader then typesets the header from them. The name and tagline are also used as defaults for the PDF metadata described in the Preamble section.

Command	Description
\name{first}{last}	The name in the header. The first name is set in SemiBold, the last name in Light.
\fullname{name}	Alternative to \name that sets the whole name in a single font.
\tagline{text}	A single line below the name, typically containing a role or field.
\cvphoto[diameter][inset]{image}	Declares the header photo, cropped to a circle. The photo class option is required for it to be displayed. If no image is provided, a placeholder avatar is shown. The default diameter is 3cm. The photo is centered over the right column unless an inset from the right margin is specified. Must be placed before \makecvheader and followed by a blank line; otherwise, the next {...} argument may be interpreted as the image.
\cvpersonalinfo{fields}	Defines the contact fields below the tagline. All fields are optional.
\email, \phone, \mailaddress, \location, \homepage, \linkedin, \github, \orcid	Predefined contact fields, used inside \cvpersonalinfo. Web fields take an ID and print the service prefix before it (\github{xGoldy} prints github.com/xGoldy), so the value is self-explanatory to parsers. [noprefix] prints the value without the prefix while retaining the link.
\CvInfoField[scheme][prefix]{field}{icon}	Declares a new contact field, or redefines a predefined one. The link is built from the scheme, prefix, and value, e.g., \CvInfoField[https://][gitlab.com/]{gitlab}{\faGitlab} makes \gitlab{your_id} print gitlab.com/your_id.
\CvInfoField*{field}{icon}	Declares a field that takes its full link as a second argument when used, e.g., \mastodon{@user@instance}{https://instance.url/@user}.
\cvfieldicon{field}{icon}	Changes the icon of a field, including the predefined fields.
\makecvheader	Typesets the header. Place it after all other header commands.

In field values, _, &, $, ~, and ^ can be written literally; only # needs to be escaped as \#. A $ cannot appear in a link target at all.

Body

This section describes the macros for writing the CV body (the content itself). The cvbody environment is provided by Lucid CV and wraps a paracol environment when a two-column layout is used. In a single-column layout, \cvcolumnbreak is ignored without raising a LaTeX error. This makes it possible to use the same content file for different CV layouts.

Put the body in a separate file and include it with \input inside cvbody in main.tex.

Page Layout

Macros in this section control the page layout.

Command	Description
\cvcolumnbreak	Marks the start of the next column within a two-column layout. Ignored in a single-column layout.
\newpage	Starts a new page (in the current column only).
\cvdivider	Inserts a light dashed rule for separating entries.

Sections

Section headings are tagged in the PDF as second- and third-level headings (H2 and H3), allowing parsers and assistive technologies to recognize the document structure. Use them to mark the boundaries between CV sections. Many ATS platforms (and human reviewers) place strong emphasis on the About section, making it a good choice for the first section.

Command	Description
\cvsection{name}	Section heading. Prints a horizontal rule underneath by default; disable it with the noheadingrule class option.
\cvsubsection{name}	Subsection heading. Use it to group entries within a section, e.g., publication types.

Entries

An entry is the fundamental building block of a resume. A resume is composed of multiple entries, which can be used for work experience, education, projects, volunteering activities, and more.

In most cases, the \cventry macro is sufficient. More complex hierarchies can use \cvsubentry. Although \cvsubentry can technically be used on its own, it is recommended to place it inside a cvgroup environment, which adds indentation to visually communicate the hierarchy between entries and sub-entries.

This section describes the entry types and their keywords, followed by an example.

Command	Description
\cventry[keywords]{title}{side text}{date}	Base CV entry element. title is the entry name in bold, typically a job role. When followed by cvgroup/\cvsubentry lines, it can instead contain the organization name. side text is typeset beside the title, typically containing the organization, but it can also contain the location, using a delimiter of your choice, e.g., {organization --- location}.
\cvsubentry[keywords]{title}{side text}{date}	Sub-entry subordinate to a higher-level entry. Should be placed inside a cvgroup environment. Takes the same arguments and keywords as \cventry, except noscale. Set on one line by default; oneline=false stacks the date underneath, like a stacked \cventry.
\begin{cvgroup}[rule] ... \end{cvgroup}	Groups the sub-entries of a \cventry, e.g., positions held at one employer, and indents everything inside the environment, including sub-entries, paragraphs, and lists. rule (default: false) adds a thin vertical line along the indentation as decoration; the bare rule keyword enables it. A group should not contain footnotes, floats, or \cvsection headings.

Delimiters (pipe characters) are automatically inserted between the non-empty arguments on the first line of an entry. Empty arguments are omitted along with their delimiters. For instance, \cventry{title}{}{date} prints only the title and date.

Entry keywords:

Keyword	Default	Description
oneline	\cventry: false (true under compressed); \cvsubentry: true	Keeps the date on the title line instead of stacking it underneath. An explicit value overrides the default.
flushdate	true on one-line entries	Flushes the date right instead of placing it after a bar; flushdate=false disables this and keeps the date after a bar. Ignored on stacked entries. Leader dots fill the gap before a flushed date for ATS compatibility: without characters in between, some PDF parsers may interpret the flush-right date as a separate column and break the reading order.
askip	small gap; none on a section's first entry or a group's first sub-entry	Minimum vertical space above the entry. Takes a length (e.g., askip=5pt); the bare keyword uses the default gap, while an explicit value also applies to the first entry. For sub-entries, the default gap applies regardless of whether the preceding entry has content.
bskip	\cventry: small gap; \cvsubentry: 0pt	Vertical space below the entry. Takes a length; the bare keyword uses the \cventry gap. For a sub-entry, a bare bskip gives the following content the same small gap it would get below a \cventry (about 1pt more than with no bskip). Avoid it on a group's last sub-entry without content, as it would unnecessarily lengthen the rule.
noscale	–	Disables the 1.11× scaling that normally makes the entry font optically match the body text.

Example:

\cventry[oneline]{Employer}{City, Country}{09/2021 -- 06/2026}

\begin{cvgroup}[rule]
  \cvsubentry{Position 2}{}{09/2023 -- 06/2026}
  \begin{itemize}
    \item ...
  \end{itemize}

  \cvsubentry{Position 1}{}{09/2021 -- 08/2023}
  \begin{itemize}
    \item ...
  \end{itemize}
\end{cvgroup}

Elements

CV elements are predefined macros that carry content with additional formatting.

Command	Description
\cvref{name}{email}{mailing address}	Reference entry. The mailing address can be replaced with the referee's company name or other affiliation.
\cvtag{tag}	Single chip: text on a tinted background with a rounded border, e.g., for a skill. Use \cvtaglist for multiple chips on the same line.
\cvtaglist{a, b, c}	Comma-separated list of chips. Wraps across lines and preserves the delimiter when an item wraps. Brace an item to keep a comma inside it: \cvtaglist{{C, C++}, Python}. Designed to remain machine-readable, but ordinary comma-separated text is preferable when maximum parsability is desired.

Warnings and Errors

Lucid CV warns primarily about problems that may not be visible in the rendered PDF. Each warning marks a document that compiles and looks correct on screen, but has a problem in the PDF structure, a link, or the text presented to PDF extractors and ATSs. Problems that are directly visible in the PDF are generally not reported, because the rendered document already shows them. If the log contains a lucidcv warning, the PDF alone may not reveal what is wrong.

The class raises the following warnings and one error:

Condition	Type	Why it is reported
Not compiled with LuaLaTeX	Error	Nothing else in the class can work.
No \DocumentMetadata line at the top of the file	Warning	The PDF looks identical, but it is neither PDF/A nor tagged and does not contain real spaces between words.
\nametaggap too small (below roughly -6pt)	Warning	The header looks fine, but some parsers might merge the name and tagline into a single line.
A \CvInfoField scheme ending in neither : nor //	Warning	The printed text is correct, but the link target is broken.
\setcolumnwidth used instead of \columnratio	Warning	No column ratio can be read, so the header falls back to 0.65.
The photo option set without \cvphoto	Warning	The missing photo is visible, but its cause is not: the option only enables photo support, while \cvphoto declares the photo itself.
\cvphoto placed after \makecvheader	Warning	Same as above. The photo must be declared before the header is built.

The photo warnings are the only exception to the general rule: their result is visible, but the cause is specific to this class and may not be obvious from the PDF alone.

The following mistakes are not reported, because their result is already apparent from the PDF:

• A missing \name, \fullname, or \cvpersonalinfo: the header is typeset without them.
• A second \makecvheader: the first one is used and subsequent calls are ignored.
• A mistyped keyword value, e.g., oneline=ture or rule=yes: only a literal true is true and anything else is false, so the entry is typeset in its other form.
• An empty field value, e.g., \email{}: the icon is printed with nothing after it.

A mistyped keyword name, e.g., \cventry[onelin], is different: it stops compilation with a LaTeX error.

1. Testing for ATS Compatibility

Since Lucid CV is designed to be ATS-compatible, the tests/ directory contains a test suite to verify this claim. Because commercial ATSs are typically closed-source and do not provide direct access to their document-processing engines, their PDF parsing behavior cannot be tested directly. Instead, the test suite evaluates the resume using 9 PDF/OCR engines in 13 configurations, providing a practical estimate of PDF parsability across different extraction implementations.

Engine	Test mode	Description
Poppler (pdftotext)	poppler	Default mode. Widely used in Linux tooling and PDF-processing pipelines.
poppler-raw	Extracts text in content-stream order.
poppler-layout	Attempts to preserve the physical page layout.
MuPDF (PyMuPDF)	mupdf	Independent PDF-processing implementation.
PDFium (pypdfium2)	pdfium	PDF engine used by Chromium-based browsers.
pdfminer.six	pdfminer	Python-based PDF text extraction library commonly used for custom document-processing pipelines.
Ghostscript (txtwrite)	ghostscript	Ghostscript's text extraction; the strictest word-gap heuristic of the set.
pdf.js	pdfjs-old	Version 2.5.207, representative of older pdf.js integrations.
pdfjs-new	Current pdf.js version, used by Firefox.
Apache PDFBox	pdfbox	Default extraction mode. Widely used in Java-based document-processing pipelines.
pdfbox-sorted	Position-sorting mode (-sort).
Apache Tika	tika	Delegates PDF extraction to PDFBox, testing Tika's document-processing pipeline rather than an independent PDF engine.
Tesseract	tesseract	OCR performed on rendered pages, representing extraction from a PDF when its text layer is unavailable or unusable.

Commercial resume parsers such as Textkernel/Sovren, Daxtra, and HireAbility are proprietary systems whose internal PDF-processing pipelines cannot be reproduced locally. The test suite therefore does not guarantee compatibility with any particular commercial ATS, but consistent extraction across 13 configurations—including OCR—provides evidence of robust, machine-readable PDF content and a useful indication of general parsability.

Test Descriptions

The test pipeline takes one .tex file, extracts the expected visible phrases from its source, builds the PDF, and runs the following checks:

1. Real word spaces: Every gap between words is stored as an actual space character in the PDF, so extractors do not have to infer word boundaries from horizontal positioning.
2. Icons as whitespace: Icons extract as plain whitespace, never as random letters or icon names.
3. PDF/A-2b conformance: The PDF is validated with veraPDF (if installed), since a PDF/A claim in the file's metadata does not guarantee actual conformance.
4. Text extraction: Every expected phrase is found in the text.
5. Reading order: The extracted phrases appear in the same order as in the document, ensuring that dates remain associated with their entries and that columns do not interleave.
6. Word-gap characters (report only, never fails): The whitespace character returned between words by each extraction configuration is recorded, since some parsers do not treat non-breaking spaces as regular whitespace.

Results

The latest test report covers a full pipeline run on the three generic variants described in Section 2 (Examples): one column on two pages (1c2p), and two compressed one-page variants with one column (1c1p) and two columns (2c1p). All three pass every required test:

1. Real word spaces

1. Icons as whitespace

1. PDF/A-2b conformance (veraPDF)

1. Text extraction, 12 text-layer configurations

1. Text extraction, Tesseract (OCR)

1. Reading order of the body, 12 text-layer configurations

1. Word-gap characters

Test	1c2p	1c1p	2c1p
Phrases checked	91	76	73
pass	pass	pass
pass	pass	pass
pass	pass	pass
all complete	all complete	9 complete, 3 at 49–50/73
79/91	70/76	65/73
1.00	1.00	8 at 1.00, 4 at 0.71–0.86
plain spaces only	plain spaces only	plain spaces only

For the one-column variants, every text-layer configuration returns all phrases in document order. The only extraction losses occur with the two-column layout: poppler-layout, ghostscript, and pdfbox-sorted reconstruct lines across the full page width, causing the two columns to be spliced together and approximately one-third of the phrases to be lost. The same three configurations, together with poppler, also produce incorrect reading order because they sort or reconstruct text based on its position. These are known limitations of the respective extraction methods with two-column layouts and therefore do not cause the test suite to fail. If you want the safest choice for ATS parsing, use one column.

The Tesseract misses are OCR errors rather than problems with the PDF itself. Capital I is occasionally recognized as lowercase l in AI, while Czech and Slovak diacritics may be dropped by the English-only OCR model. This has limited practical significance because an ATS normally reads the PDF's text layer and uses OCR only when usable text cannot be extracted.

In addition, I manually tested the one-column PDFs produced by the template by uploading them to Workday, Greenhouse, and Lever ATSs during my own job search. In all cases, relevant fields, including contact information, education, work experience, and skills, were extracted correctly.

These results do not constitute a guarantee of compatibility with every ATS. However, they provide strong evidence that one-column PDFs produced by the class are robustly machine-readable across multiple independent PDF extraction implementations and have also performed correctly in several commercial ATSs in practical use.

Running the Tests Yourself

The tests/ directory is available only in the GitHub repository, not in the Overleaf or CTAN packages.

If you wish to run the PDF extraction tests yourself, the pipeline.sh script provides an end-to-end way to run the full test suite. First, set up the test environment with setup.sh. This script installs the required tools in two groups: system packages (Poppler, Ghostscript, Tesseract, and a Java runtime), installed through apt-get or dnf and requiring root privileges; and local dependencies (Python, pdf.js, PDFBox, Tika, and veraPDF), installed under tests/env/ without root access.

With --no-system, the system packages are skipped. Tests that depend on them are reported as not run rather than failing. This leaves 5 of the 13 configurations available (mupdf, pdfium, pdfminer, pdfjs-old, and pdfjs-new), while the PDF/A check falls back to its built-in approximation.

The TeX toolchain is not installed by the test setup script. The same LuaLaTeX environment used to build the class (see Local Compilation (Linux)) is required.

cd tests
./setup.sh                    # or: ./setup.sh --no-system (no root needed)

Once the environment is ready, run pipeline.sh against the top-level (main) .tex file of a Lucid CV project. It builds the PDF, generates the phrase list from the same source, runs all tests, and exits with a non-zero status if any required test fails. Class options can be passed with --opt; the option can be repeated to test different layouts without modifying the document:

./pipeline.sh ../main.tex                                    # builds and tests ../main.tex
./pipeline.sh ../main.tex --opt twocolumn --opt compressed   # with class options

Everything produced by a run (the PDF, the phrase list, extracted text from each engine, run.log, and a machine-readable pipeline.json) is written to tests/out/<label>/. The results of each test are reported directly to the console.

1. License

Copyright (c) 2026 Patrik Goldschmidt (<goldschmidt.patrik@gmail.com>)

License	Files
LaTeX Project Public License (LPPL) 1.3c	lucidcv.cls, main.tex, latexmkrc, README.md, resources/docs.md
MIT License	All other files, including build.sh, package.sh, and the ATS compatibility tests in tests/
SIL Open Font License	Bundled fonts in fonts/

See the LICENSE file for the complete license texts and additional information.

1. Support the Project

Although much of this class and its associated tooling was vibe-coded (Opus 5), the project still took 250+ hours of design, fine-tuning, manual verification, functionality testing, and documentation to bring this free public template to life.

The class was originally created to solve my own problem, but as a strong supporter of free and open-source software, I decided to release it for anyone looking for an alternative to inflexible free templates, ATS-unfriendly design tools, or premium templates that cost hundreds of dollars.

If you found the class useful, I would appreciate a message or simply a GitHub ⭐. If you feel that your gratitude cannot be expressed in words, you can also buy me a coffee on Ko-fi. Your support helps me maintain this template and create more open-source projects in the future. Many thanks!