Improve accessibility

[ryanlovett]

These changes improve accessibility scoring from two different accessibility checkers. The main differences are using lualatex instead of xetex and adding tags. I needed to install TeX Live 2025. I don't know LaTeX very well, and relied on @rhstanton's work in https://github.com/rhstanton/accessible_LaTeX/. All shortcomings are mine.

The tools I used to measure accessibility.

1. ally, used within my institution's LMS, which scans uploaded PDFs
2. verapdf, used via docker

   myst build --pdf
   docker run --platform linux/amd64 --rm -it -v .:/data ghcr.io/verapdf/cli:latest \
     --format json -df ua2  --loglevel 0  \
     exports/myst-latex.pdf > verapdf-analysis.json
   

My local myst project referenced my local fork:

version: 1

project:
  title: Accessibility Test
  authors:
    - name: Your Name
  exports:
    - format: pdf
      template: ../lapreprint
      output: exports/myst-latex.pdf
    - format: tex
      template: ../lapreprint
      output: exports/myst-latex.tex
  toc:
    - file: index.md

site:
  template: book-theme

My index.md test file contained:

---
title: foo
authors:
  - name: Your Name
exports:
  - format: pdf
  - format: typst
---
+++ { "part": "abstract" }

This is a placeholder abstract.

+++

## Code

:::{code-cell} python
hello = "hello"
there = "there"
phrase = f"{hello}, {there}!"
print(phrase)
:::

## header 1

some math

$$
\int f(\theta)d\theta
$$

## header 2

:::{csv-table} a table
:header: "col1", "col2"

4, "a"
5, "b"
:::

(I tried several table variants besides csv-table.)

When I generated verapdf json output with the current lapreprint code and summarized with:

$ cat /tmp/myst-latex-upstream.json | jq '
      .report.jobs[0].validationResult[0]
      | { compliant, failed_rules: .details.failedRules, failed_checks: .details
.failedChecks }
    '
{
  "compliant": false,
  "failed_rules": 10,
  "failed_checks": 43
}

When I ran it with my changes:

$ $ cat /tmp/myst-latex-fixes.json | jq '
      .report.jobs[0].validationResult[0]
      | { compliant, failed_rules: .details.failedRules, failed_checks: .details
.failedChecks }
    '
{
  "compliant": false,
  "failed_rules": 1,
  "failed_checks": 4
}

The remaining failure is related to tables and tagging, and I believe that would need to be fixed in myst.

Cc: jupyter-book/mystmd#2743

myst-latex-upstream.pdf
myst-latex-fixes.pdf

[rowanc1]

Thanks @ryanlovett!

[cycomachead]

I'm late to this, but I think we should probably start targeting pdf 2.0 standard documents. 2.0 is required for MathML and UA-2 is designed to work with 2.0. (Older readers can still open 2.0 documents, and even though they score less on checkers, I think that will be an anti-pattern soon enough.)