Tagged, accessible PDFs from LaTeX

A screen reader finds the headings, the table cells and the description of a figure in a PDF only through its tags. In this lesson’s builds, one LaTeX report passed all 1,727 rules of veraPDF’s PDF/UA-2 profile after a single LuaLaTeX run, while pdfLaTeX’s first run of the same file left content in 133 places neither tagged nor marked as decoration.

You will switch tagging on with one line above \documentclass, choose the engine, add the alternative text, the header row, the title and the language that only you can give, and check your packages before you start. Every build was recorded with TeX Live 2026, read back with poppler and validated with veraPDF 1.30.

One line above \documentclass switches tagging on

The whole switch is the first line of report.tex: \DocumentMetadata{lang=en-IN, pdfstandard=ua-2, tagging=on}. LaTeX’s tagging instructions place \DocumentMetadata before \documentclass: it sets up the document’s metadata and the standards it should meet before the class and the packages are read. Its keys say what to build: lang the document’s main language, pdfstandard the standard the file will claim, here PDF/UAPDF/UA: ISO 14289, the standard for accessible PDF files: tagged content, a title and a structure that assistive software can follow. PDF/UA-1 builds on PDF 1.7 and PDF/UA-2 on PDF 2.0. More in “A one-page resume in LaTeX that parses”-2, and tagging=on that every paragraph, heading, figure and table is tagged as it is typeset. The rest of the report is ordinary LaTeX, with one line for the table that the third section explains:

% report.tex: a one-page report with a figure, a table, a formula and a link, tagged for PDF/UA-2.
% Build it twice with LuaLaTeX (the second run completes the MathML of the formula); pdfLaTeX builds it too.
\DocumentMetadata{lang=en-IN, pdfstandard=ua-2, tagging=on}
\documentclass{article}
\ifdefined\directlua \usepackage{unicode-math}\fi % LuaLaTeX writes MathML for formulas with unicode-math
\usepackage{graphicx}
\usepackage{hyperref}
\title{Monsoon rainfall at Station A}
\author{Asha Verghese}
\date{}
\begin{document}
\maketitle
\section{Summary}
Most of the rain at Station A, an invented station, falls from June to September.
\begin{figure}[ht]
  \centering
  \includegraphics[width=0.6\linewidth, alt={Bar chart of monthly rainfall: under 160 mm a month
    from January to May, 418 mm in June, a peak of 655 mm in July, then falling to 306 mm in September.}]{chart.png}
  \caption{Monthly rainfall at Station A.}
\end{figure}
\section{The wettest months}
\tagpdfsetup{table/header-rows=1}
\begin{table}[ht]
  \centering
  \caption{The four monsoon months.}
  \begin{tabular}{lrr}
    Month & Rain (mm) & Rainy days \\
    June & 418 & 21 \\
    July & 655 & 27 \\
    August & 541 & 25 \\
    September & 306 & 17 \\
  \end{tabular}
\end{table}
The total for the four months is $418 + 655 + 541 + 306 = 1920$~mm. The data are described at
\url{https://example.com/station-a}.
\end{document}

from session import files, read, run, write

# report.tex built twice by LuaLaTeX: what pdfinfo reads back, the language and the PDF/UA claim inside the file, and
# veraPDF's verdict against PDF/UA-2. Last, a copy with a second \DocumentMetadata line above the first.
files("report/report.tex", "report/make_chart.py", "vera.py")
write("twice.tex", "\\DocumentMetadata{tagging=on}\n" + read("report.tex"))
run(r"""
python3 make_chart.py
lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
pdfinfo report.pdf | grep -E "^(Title|Tagged|PDF version)"
qpdf --qdf --object-streams=disable report.pdf uncompressed.pdf
grep -a "/Lang" uncompressed.pdf
pdfinfo -meta report.pdf | grep -E "pdfuaid:part|<rdf:li>en-IN"
python3 vera.py ua2 report.pdf
!lualatex -interaction=nonstopmode -halt-on-error twice.tex > /dev/null
grep -E "^! " twice.log
""")

$ python3 make_chart.py
$ lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ pdfinfo report.pdf | grep -E "^(Title|Tagged|PDF version)"
Title:           Monsoon rainfall at Station A
Tagged:          yes
PDF version:     2.0
$ qpdf --qdf --object-streams=disable report.pdf uncompressed.pdf
$ grep -a "/Lang" uncompressed.pdf
  /Lang (en-IN)
$ pdfinfo -meta report.pdf | grep -E "pdfuaid:part|<rdf:li>en-IN"
   <pdfuaid:part>2</pdfuaid:part>
     <rdf:li>en-IN</rdf:li>
$ python3 vera.py ua2 report.pdf
report.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
$ lualatex -interaction=nonstopmode -halt-on-error twice.tex > /dev/null
[exit status 1]
$ grep -E "^! " twice.log
! LaTeX Error: Two \DocumentMetadata declarations.
!  ==> Fatal error occurred, no output PDF file produced!

The figure is a 360-pixel bar chart that make_chart.py draws as pixels only, so the PDF can say what it shows only through its alternative text. pdfinfo read the result back: a tagged PDFtagged PDF: A PDF that also stores its structure, such as headings, paragraphs and reading order, so screen readers and text extraction can follow it. Typst writes tagged PDF by default. More in “LaTeX and Typst: what they are and when to use each” of PDF version 2.0, with the title from \title. The catalog holds /Lang (en-IN), the XMP metadataXMP metadata: The block of XML inside a PDF that describes the file, such as its title, its author and, in a PDF/A file, the part and conformance level it claims. pdfinfo -meta prints it. More in “Archival PDF/A and final submission checks” holds pdfuaid:part 2 and dc:language, and veraPDF found every rule of the profile met. Step through where each key went:

The values of lang are BCP 47BCP 47 language tag: A code for a language, with its region or script when needed, such as en-IN for English as written in India, hi for Hindi or ta for Tamil. PDFs, web pages and the language settings of LaTeX and Typst use these codes. language tags, the codes that also name regional forms, such as en-IN for English as written in India or hi for Hindi. If you leave the language out, LaTeX2e News 43 says the code silently takes the main language of babel or polyglossia, or en when neither is loaded, so a document in Hindi that loads neither package would claim to be English. The W3C’s technique for the catalog’s /Lang gives the reason it matters: screen readers can load the correct pronunciation rules.

Because the line comes first and only once, it belongs to whoever owns the top of the file. A template that ships the line leaves it out of the documents built on it, and a document that needs another key changes the template’s line instead of adding a second one:

LuaLaTeX tags in one run; pdfLaTeX needs two and drops spaces

The tagging instructions name both pdfLaTeX and LuaLaTeX, and call LuaLaTeX the preferred engine for new documents. The tagpdf manual gives the reasons: LuaLaTeX requires fewer compilations, while pdfLaTeX needs more of them to resolve the cross-references that tagging creates. The first session builds the same report with each engine and checks the PDF of each run; the second builds two copies, one with a bare \hrule under the title and one without unicode-math:

from session import files, run

# The same report built by pdfLaTeX (report.tex) and by LuaLaTeX (lua.tex, a copy, so each engine starts in a folder
# of its own files), with a copy of each engine's first PDF kept before its second run; after LuaLaTeX's first run, the
# MathML file it wrote is listed. veraPDF checks all four files against PDF/UA-2; then how each final structure tree
# reads the words around the line break after "The", and how many formulas carry MathML in pdfLaTeX's final file and
# in LuaLaTeX's first and final files.
files("report/report.tex", "report/make_chart.py", "vera.py")
run(r"""
python3 make_chart.py
cp report.tex lua.tex
pdflatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
cp report.pdf pdflatex-run1.pdf
pdflatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error lua.tex > /dev/null
cp lua.pdf lualatex-run1.pdf
ls *mathml.html
lualatex -interaction=nonstopmode -halt-on-error lua.tex > /dev/null
python3 vera.py ua2 pdflatex-run1.pdf report.pdf lualatex-run1.pdf lua.pdf
pdfinfo -struct-text report.pdf | grep -a "data are"
pdfinfo -struct-text lua.pdf | grep -a "data are"
qpdf --qdf --object-streams=disable report.pdf report-qdf.pdf
qpdf --qdf --object-streams=disable lualatex-run1.pdf lua-run1-qdf.pdf
qpdf --qdf --object-streams=disable lua.pdf lua-qdf.pdf
!grep -a -c "application#2fmathml+xml" report-qdf.pdf
!grep -a -c "application#2fmathml+xml" lua-run1-qdf.pdf
grep -a -c "application#2fmathml+xml" lua-qdf.pdf
""")

$ python3 make_chart.py
$ cp report.tex lua.tex
$ pdflatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ cp report.pdf pdflatex-run1.pdf
$ pdflatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error lua.tex > /dev/null
$ cp lua.pdf lualatex-run1.pdf
$ ls *mathml.html
lua-luamml-mathml.html
$ lualatex -interaction=nonstopmode -halt-on-error lua.tex > /dev/null
$ python3 vera.py ua2 pdflatex-run1.pdf report.pdf lualatex-run1.pdf lua.pdf
pdflatex-run1.pdf: PDF/UA-2 + Tagged PDF validation profile: NOT compliant, 1726 rules passed, 1 failed
  8.2.2-1 (133 checks): Content is neither marked as Artifact nor tagged as real content
report.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
lualatex-run1.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
lua.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
$ pdfinfo -struct-text report.pdf | grep -a "data are"
        "mm. Thedata are described at"
$ pdfinfo -struct-text lua.pdf | grep -a "data are"
        " mm. The data are described at "
$ qpdf --qdf --object-streams=disable report.pdf report-qdf.pdf
$ qpdf --qdf --object-streams=disable lualatex-run1.pdf lua-run1-qdf.pdf
$ qpdf --qdf --object-streams=disable lua.pdf lua-qdf.pdf
$ grep -a -c "application#2fmathml+xml" report-qdf.pdf
0
[exit status 1]
$ grep -a -c "application#2fmathml+xml" lua-run1-qdf.pdf
0
[exit status 1]
$ grep -a -c "application#2fmathml+xml" lua-qdf.pdf
1

from session import files, read, run, write

# Two more copies of report.tex: one with a bare \hrule under the title, built twice by pdfLaTeX and once by LuaLaTeX
# (rule-lua.tex), and one without unicode-math, built twice by LuaLaTeX, with its warning and its count of formulas
# that carry MathML.
files("report/report.tex", "report/make_chart.py", "vera.py")
write("rule.tex", read("report.tex").replace("\\maketitle\n", "\\maketitle\n\\hrule\n"))
run(r"""
python3 make_chart.py
cp rule.tex rule-lua.tex
pdflatex -interaction=nonstopmode -halt-on-error rule.tex > /dev/null
pdflatex -interaction=nonstopmode -halt-on-error rule.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error rule-lua.tex > /dev/null
python3 vera.py ua2 rule.pdf rule-lua.pdf
sed '/unicode-math/d' report.tex > plain-maths.tex
lualatex -interaction=nonstopmode -halt-on-error plain-maths.tex | grep -A1 "Package tagpdf Warning"
lualatex -interaction=nonstopmode -halt-on-error plain-maths.tex > /dev/null
qpdf --qdf --object-streams=disable plain-maths.pdf plain-maths-qdf.pdf
!grep -a -c "application#2fmathml+xml" plain-maths-qdf.pdf
""")

$ python3 make_chart.py
$ cp rule.tex rule-lua.tex
$ pdflatex -interaction=nonstopmode -halt-on-error rule.tex > /dev/null
$ pdflatex -interaction=nonstopmode -halt-on-error rule.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error rule-lua.tex > /dev/null
$ python3 vera.py ua2 rule.pdf rule-lua.pdf
rule.pdf: PDF/UA-2 + Tagged PDF validation profile: NOT compliant, 1726 rules passed, 1 failed
  8.2.2-1 (1 check): Content is neither marked as Artifact nor tagged as real content
rule-lua.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
$ sed '/unicode-math/d' report.tex > plain-maths.tex
$ lualatex -interaction=nonstopmode -halt-on-error plain-maths.tex | grep -A1 "Package tagpdf Warning"
Package tagpdf Warning: Neither unicode-math nor lua-unicode-math has been
(tagpdf)                loaded.
$ lualatex -interaction=nonstopmode -halt-on-error plain-maths.tex > /dev/null
$ qpdf --qdf --object-streams=disable plain-maths.pdf plain-maths-qdf.pdf
$ grep -a -c "application#2fmathml+xml" plain-maths-qdf.pdf
0
[exit status 1]

Three of those rows are differences you would never see on the page. Tagging makes pdfTeX put a space glyph between words, a real space glyphreal space glyph: A space character drawn from a font between two words, which a PDF stores so that copying, search and screen readers can find where one word ends. TeX itself only moves on between words; with tagging on, pdfTeX and LuaTeX add these glyphs. that text extraction and screen readers can find, but the space at the end of a line is not written, so poppler read the tags as Thedata. LuaLaTeX wrote the space. LuaLaTeX with unicode-math also attached MathMLMathML: The W3C's markup language for mathematics, which records a formula's structure, such as which part is a fraction or a power, so that browsers can lay it out and assistive technology can read it. More in “Maths fonts, symbol lookup and accessible equations” to the formula, which a screen reader can read as maths, but only on its second run: the first wrote the MathML to lua-luamml-mathml.html, and its PDF carried none. The pdfLaTeX file carried none at all. And the tagpdf manual says LuaLaTeX’s Lua mode marks everything left unmarked as an artifactartifact: Content of a tagged PDF that is not part of the document's text, such as a page number, a decorative rule or a background, marked so that screen readers and text extraction can skip it. More in “A one-page resume in LaTeX that parses”, so the bare rule passed, while pdfLaTeX left it as content that is neither tagged nor decoration: rule 8.2.2-1.

Only you know the description, the header row, the title and the language

The tagging code tags what LaTeX’s commands tell it: a \section becomes an H1, a figure a Note with a Caption. It cannot know what a picture shows, which row of a table holds the headings, what the document is called or which language it is in. tags.py, from the alternative text lesson, prints the structure treestructure tree: The tree of tags in a tagged PDF, such as headings, paragraphs, lists and links, in reading order, which screen readers follow. pdfinfo -struct-text prints it with the text of each tag. More in “A one-page resume in LaTeX that parses” that poppler reads; the session then builds two copies that each leave one of those out:

"""Prints the structure tree that `pdfinfo -struct-text` reads from a tagged PDF, given on standard input, in short:
one line per tag, indented as in the tree, with the alternative text of a Figure in square brackets and the text of
the tag in quotes. The numbers poppler gives each tag, its attributes, its references to link annotations and its empty
text runs are left out.

  pdfinfo -struct-text report.pdf | python3 tags.py
"""
import re
import sys

tags = []  # [depth, tag, alt, text]
# pdfinfo prints some titles as raw UTF-16 bytes; they are not text a reader hears, so they are dropped.
for raw in sys.stdin.buffer.read().decode("utf-8", errors="replace").splitlines():
    body = raw.lstrip(" ")
    depth = (len(raw) - len(body)) // 2
    if not body or body.startswith("/") or body.startswith("Syntax") or body.startswith("Object "):
        continue
    if body.startswith('"'):
        text = body.strip().strip('"').strip()
        # A run of text belongs to the nearest tag above it that is less indented.
        owner = next((t for t in reversed(tags) if t[0] < depth), None)
        if text and owner:
            owner[3] = (owner[3] + " " + text).strip()
        continue
    match = re.match(r"(\w+)(?: <[^>]*>)?(?: \[\"(.*)\"\])?", body)
    if match:
        tags.append([depth, match.group(1), (match.group(2) or "").strip(), ""])

if not tags:
    print("no structure tree: the PDF is not tagged")
for depth, tag, alt, text in tags:
    line = "  " * depth + tag
    if tag == "Figure":
        line += f" [{alt}]" if alt else " (no alternative text)"
    if text:
        line += f' "{text}"'
    print(line)

from session import files, read, run, write

# The structure tree of the LuaLaTeX build, shortened by tags.py; then two copies of report.tex that leave out what
# only the author can give: the table's header row, and the title.
files("report/report.tex", "report/make_chart.py", "vera.py", "tags.py")
text = read("report.tex")
write("no-header.tex", text.replace("\\tagpdfsetup{table/header-rows=1}\n", ""))
write("no-title.tex", text.replace("\\title{Monsoon rainfall at Station A}\n", "").replace("\\maketitle\n", ""))
run(r"""
python3 make_chart.py
lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
pdfinfo -struct-text report.pdf | python3 tags.py
lualatex -interaction=nonstopmode -halt-on-error no-header.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error no-header.tex > /dev/null
pdfinfo -struct-text no-header.pdf | python3 tags.py | grep -a -E "T[HD] " | head -4
lualatex -interaction=nonstopmode -halt-on-error no-title.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error no-title.tex > /dev/null
!pdfinfo no-title.pdf | grep "^Title"
python3 vera.py ua2 no-header.pdf no-title.pdf
""")

$ python3 make_chart.py
$ lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error report.tex > /dev/null
$ pdfinfo -struct-text report.pdf | python3 tags.py
Document
  Part
    P
      P "Monsoon rainfall at Station A"
    P "Asha Verghese"
  Sect
    H1 "Summary"
      Span "1"
    Part
      P "Most of the rain at Station A, an invented station, falls from June to September."
  Sect
    H1 "The wettest months"
      Span "2"
    Part
      P "The total for the four months is mm. The data are described at ."
        Formula "418+655+541+306=1920"
        Link "https://example.com/station-a"
  Sect
    Note
      Caption
        Lbl "Figure 1:"
        P "Monthly rainfall at Station A."
      Part
        P
          Figure [Bar chart of monthly rainfall: under 160 mm a month from January to May, 418 mm in June, a peak of 655 mm in July, then falling to 306 mm in September.]
  Sect
    Note
      Caption
        Lbl "Table 1:"
        P "The four monsoon months."
      Part
        P
        Table
          TR
            TH "Month"
            TH "Rain (mm)"
            TH "Rainy days"
          TR
            TD "June"
            TD "418"
            TD "21"
          TR
            TD "July"
            TD "655"
            TD "27"
          TR
            TD "August"
            TD "541"
            TD "25"
          TR
            TD "September"
            TD "306"
            TD "17"
        P
$ lualatex -interaction=nonstopmode -halt-on-error no-header.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error no-header.tex > /dev/null
$ pdfinfo -struct-text no-header.pdf | python3 tags.py | grep -a -E "T[HD] " | head -4
            TD "Month"
            TD "Rain (mm)"
            TD "Rainy days"
            TD "June"
$ lualatex -interaction=nonstopmode -halt-on-error no-title.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error no-title.tex > /dev/null
$ pdfinfo no-title.pdf | grep "^Title"
[exit status 1]
$ python3 vera.py ua2 no-header.pdf no-title.pdf
no-header.pdf: PDF/UA-2 + Tagged PDF validation profile: compliant, 1727 rules passed, 0 failed
no-title.pdf: PDF/UA-2 + Tagged PDF validation profile: NOT compliant, 1726 rules passed, 1 failed
  8.11.1-1 (1 check): Metadata stream does not contain dc:title

The tree holds everything the author gave. The Figure carries the alt text in square brackets, the formula is a Formula tag, the URL a Link, and the first row of the table is three TH cells because of the line before the table, \tagpdfsetup{table/header-rows=1}. Without that line every cell was a TD, so a screen reader could not name the column of 655, and veraPDF still passed all 1,727 rules: no rule can know that the first row was meant to head the columns. Without \title, pdfinfo found no title and veraPDF failed rule 8.11.1-1, “Metadata stream does not contain dc:title”.

The tagging instructions put the first row plainly: every graphic must be marked either as an artifact or get an alternative text. The table key works the same way for a whole document when it sits in the preamble, or for the next tables when it sits before them, as here; by default, the instructions say, all cells are data cells.

Check every package against the tagging status list first

Tagging changes how LaTeX builds headings, floats and tables, so a package that patches those parts may break the build or the tags. The LaTeX Project’s tagging statustagging status list: The LaTeX Project's list of how each package and class works with tagging switched on: compatible, partially compatible, currently incompatible, no support or unchecked, with comments on what to do instead. list gives each of more than 2,200 packages and classes one of five words: compatible, partially compatible, currently incompatible, no support (the project expects never to support it), or unchecked. These three copies of the report load a package each:

from session import files, read, run, write

# Three copies of report.tex: one that loads titlesec, one that loads pdfx, and one that loads multirow and asks for
# LaTeX's tagging status report with the check-tagging-status key. Last, tlmgr says whether this TeX Live has the
# latex-tagging-status package, which holds the data the report is filled from (--only-installed: no network).
files("report/report.tex", "report/make_chart.py")
text = read("report.tex")
write("titlesec.tex", text.replace("\\usepackage{hyperref}\n", "\\usepackage{hyperref}\n\\usepackage{titlesec}\n"))
write("pdfx.tex", text.replace("\\usepackage{hyperref}\n", "\\usepackage{pdfx}\n\\usepackage{hyperref}\n"))
write("status.tex", text.replace("tagging=on}", "tagging=on, check-tagging-status}")
      .replace("\\usepackage{hyperref}\n", "\\usepackage{hyperref}\n\\usepackage{multirow}\n"))
run(r"""
python3 make_chart.py
!lualatex -interaction=nonstopmode -halt-on-error titlesec.tex > /dev/null
grep -E -A1 "sectioning command .section$|^! " titlesec.log
lualatex -interaction=nonstopmode -halt-on-error pdfx.tex > /dev/null
grep -E -A1 "loading of package pdfx" pdfx.log
lualatex -interaction=nonstopmode -halt-on-error status.tex > /dev/null
lualatex -interaction=nonstopmode -halt-on-error status.tex > status.txt
sed -n '/Status report/,/6. Unclassified/p' status.txt
sed -n '/6. Unclassified/,/=====/p' status.txt | grep -c "^ [a-z0-9-]*\.sty$"
tlmgr info --only-installed latex-tagging-status | grep -E "^(package|installed):"
""")

$ python3 make_chart.py
$ lualatex -interaction=nonstopmode -halt-on-error titlesec.tex > /dev/null
[exit status 1]
$ grep -E -A1 "sectioning command .section$|^! " titlesec.log
Package titlesec Warning: Non standard sectioning command \section
(titlesec)                detected. Using default spacing and no format.
--
! Package titlesec Error: No format for this command.

--
!  ==> Fatal error occurred, no output PDF file produced!
$ lualatex -interaction=nonstopmode -halt-on-error pdfx.tex > /dev/null
$ grep -E -A1 "loading of package pdfx" pdfx.log
Package pdfmanagement Warning: The loading of package pdfx is disabled.
(pdfmanagement)                It is not compatible with the PDF management.
$ lualatex -interaction=nonstopmode -halt-on-error status.tex > /dev/null
$ lualatex -interaction=nonstopmode -halt-on-error status.tex > status.txt
$ sed -n '/Status report/,/6. Unclassified/p' status.txt
 Status report of the tagging support
======================================
 Details at https://latex3.github.io/tagging-project/tagging-status
 Report erroneous entries at https://github.com/latex3/tagging-project

 A. CLASS
 ********

 article.cls is not in the status list

 B. PACKAGES, LIBRARIES, etc
 ***********

 1. Unsupported
 --------------------------
 NONE

 2. Currently incompatible
 --------------------------
 NONE

 3. Partially compatible
 --------------------------
 NONE

 4. Compatible
 --------------------------
 NONE

 5. Unknown
 --------------------------
 NONE

 6. Unclassified files with extension .sty
$ sed -n '/6. Unclassified/,/=====/p' status.txt | grep -c "^ [a-z0-9-]*\.sty$"
43
$ tlmgr info --only-installed latex-tagging-status | grep -E "^(package|installed):"
package:     latex-tagging-status
installed:   No

titlesec stopped the build: it warned that \section was a non-standard sectioning command, then failed with “No format for this command”, because headings are now built from templates, as LaTeX2e News 43 describes. The PDF management code refused to load pdfx at all; the status list says to set standards with \DocumentMetadata keys instead. The check-tagging-status key printed a status report at the end of the run, but it classified nothing: article.cls “is not in the status list”, and all 43 package files came out unclassified. The last command shows why: CTAN describes the latex-tagging-status package as the data behind such checks, and this TeX Live did not have it. Install it with tlmgr where you use the key, and look packages up on the online list as well.

The order of work follows from the table. Step through it, with what this lesson’s sessions met at each stage:

Practise: check a preamble and a table before you tag

Two checks that a script can run before a build and after it. The first reads a .tex file and lists what keeps it from being tagged, the slips of this lesson among them; the second finds the tables whose first row has no TH cell in the structure tree that tags.py prints. Each step has sample tests you can run in the browser and a first hint:

def preamble_problems(source, status):
    """Return what keeps a LaTeX file from being tagged, in the order of the checks."""
    # Replace this line: read the file's lines, its \DocumentMetadata and its packages.
    return []

from preamble import preamble_problems

STATUS = {
    "graphicx": "compatible",
    "hyperref": "partially-compatible",
    "unicode-math": "compatible",
    "titlesec": "currently-incompatible",
    "multirow": "currently-incompatible",
    "pdfx": "no-support",
    "booktabs": "compatible",
}

REPORT = r"""% report.tex
\DocumentMetadata{lang=en-IN, pdfstandard=ua-2, tagging=on}
\documentclass{article}
\usepackage{graphicx}
\usepackage{hyperref}
\begin{document}
Text.
\end{document}
"""


def test_report():
    """finds nothing to fix in the lesson's report.tex"""
    assert preamble_problems(REPORT, STATUS) == []


def test_titlesec():
    """names titlesec with its status when the report loads it"""
    source = REPORT.replace(r"\usepackage{hyperref}", "\\usepackage{hyperref}\n\\usepackage{titlesec}")
    assert preamble_problems(source, STATUS) == ["titlesec: currently-incompatible"]


def test_list_of_packages():
    """reads every name of \\usepackage[opts]{a,b} and keeps their order"""
    source = REPORT.replace(r"\usepackage{graphicx}", r"\usepackage[final]{pdfx,booktabs,multirow}")
    assert preamble_problems(source, STATUS) == ["pdfx: no-support", "multirow: currently-incompatible"]


def test_missing_metadata():
    """reports a file without \\DocumentMetadata once, without checks 3 and 4"""
    source = REPORT.replace("\\DocumentMetadata{lang=en-IN, pdfstandard=ua-2, tagging=on}\n", "")
    assert preamble_problems(source, STATUS) == [r"\DocumentMetadata must come before \documentclass"]


def test_metadata_after_class():
    """reports \\DocumentMetadata after \\documentclass, and still reads its keys"""
    source = "\\documentclass{article}\n\\DocumentMetadata{pdfstandard=ua-2}\n\\begin{document}\nx\n\\end{document}\n"
    assert preamble_problems(source, STATUS) == [
        r"\DocumentMetadata must come before \documentclass",
        "tagging is not switched on",
        "no document language",
    ]


def test_twice():
    """reports a second \\DocumentMetadata"""
    source = "\\DocumentMetadata{tagging=on}\n" + REPORT
    assert preamble_problems(source, STATUS) == [r"\DocumentMetadata is used twice", "no document language"]


def test_spaces_and_language_key():
    """accepts tagging = on with spaces, and the language key instead of lang"""
    source = REPORT.replace("lang=en-IN, pdfstandard=ua-2, tagging=on", "language = en-IN, tagging = on")
    assert preamble_problems(source, STATUS) == []


def test_comments():
    """ignores a commented-out package and a commented-out key, but not an escaped \\%"""
    source = REPORT.replace("tagging=on}", "} % tagging=on").replace(
        r"\usepackage{hyperref}", "\\usepackage{hyperref}\n% \\usepackage{titlesec}\nRain fell by 40\\% \\usepackage{multirow}"
    )
    assert preamble_problems(source, STATUS) == ["tagging is not switched on", "multirow: currently-incompatible"]


def test_unknown_package():
    """says nothing about a package missing from the status map"""
    source = REPORT.replace(r"\usepackage{graphicx}", r"\usepackage{rainfallplot}")
    assert preamble_problems(source, STATUS) == []

        Table
          TR
            TH "Month"

def tables_without_headers(tree):
    """Return the numbers of the tables (1, 2, …) whose first row has no TH cell."""
    # Replace this line: walk the lines, find each Table and its first TR.
    return []

from headers import tables_without_headers

WITH_HEADER = """Document
  Sect
    Note
      Part
        Table
          TR
            TH "Month"
            TH "Rain (mm)"
            TH "Rainy days"
          TR
            TD "June"
            TD "418"
            TD "21"
        P
"""

WITHOUT_HEADER = WITH_HEADER.replace('TH "', 'TD "')

TYPST = """Document
  H1 "2. The wettest months"
  Table
    Caption
      Span "Table 1: The four monsoon months."
    THead
      TR
        TH "Month"
        TH "Rain (mm)"
    TBody
      TR
        TD "June"
        TD "418"
"""


def test_header_row():
    """finds no problem in the lesson's report, whose first row is three TH cells"""
    assert tables_without_headers(WITH_HEADER) == []


def test_no_header_row():
    """reports table 1 when every cell is a TD, as in the copy without header-rows"""
    assert tables_without_headers(WITHOUT_HEADER) == [1]


def test_typst_thead():
    """reads the first row inside a THead, as Typst writes tables"""
    assert tables_without_headers(TYPST) == []


def test_later_header_does_not_count():
    """looks at the first row only, even when a later row has a TH"""
    tree = WITHOUT_HEADER.replace('TD "June"', 'TH "June"')
    assert tables_without_headers(tree) == [1]


def test_two_tables():
    """numbers tables in document order and reports only the second"""
    tree = WITH_HEADER + WITHOUT_HEADER.replace("Document\n", "")
    assert tables_without_headers(tree) == [2]


def test_table_without_rows():
    """reports a table that has no TR at all"""
    tree = "Document\n  Table\n    Caption\n      Span \"Empty\"\n  P \"After the table.\"\n"
    assert tables_without_headers(tree) == [1]


def test_no_tables():
    """returns an empty list for a tree without tables"""
    assert tables_without_headers('Document\n  H1 "Summary"\n  P "Text."\n') == []

Then practise the four ideas of the lesson, from the line above the class to the status list. Each answer you give is explained on the spot, wrong ones included, and the ideas you miss come round again:

Interview questions

Warm-up (fresher to 2–5 years): what does \DocumentMetadata{lang=en-IN, pdfstandard=ua-2, tagging=on} change in a LaTeX PDF?

It must be the first line, above \documentclass, and it loads LaTeX’s PDF management and tagging code before the class. tagging=on makes LaTeX record a structure tree while it typesets: headings, paragraphs, figures, table cells and links, in reading order. lang=en-IN writes the catalog’s /Lang and the metadata’s dc:language, which screen readers use to pick a voice. pdfstandard=ua-2 makes the file claim PDF/UA-2 in its XMP metadata, which validators check. The line also produces PDF 2.0 and, Overleaf’s notes on TeX Live 2026 add, T1 font encoding. In this lesson’s run the report passed all 1,727 rules of veraPDF’s PDF/UA-2 profile, but the author still had to add alternative text, the table’s header row and a title.

Two harder questions follow: moving a department’s thesis template to tagged PDFs, and a report that passes veraPDF while a screen-reader user cannot read its tables. Decide your first step before you open each one:

Key takeaways

• \DocumentMetadata{lang=…, pdfstandard=ua-2, tagging=on} goes once, above \documentclass, and gives the file its structure tree, its language and its PDF/UA-2 claim; a second one is an error.
• LuaLaTeX passed PDF/UA-2 after one run, wrote the space at a line break, attached MathML with unicode-math on its second run and marked a bare rule as an artifact; pdfLaTeX needed two runs and still missed those three.
• The tagging code cannot know what a picture shows, which row heads a table, the title or the language, and veraPDF passed a table without its header row, so add them yourself and read the structure tree.
• Look up every package on the tagging status list before you switch tagging on: titlesec stopped the build, pdfx was refused, and without its data package this TeX Live’s own status report could not classify a single package.