The difference in one sentence

PDF/A-2b and PDF/A-3b describe the same archival file — PDF/A-3 simply allows you to embed other files inside it. If your thesis is one PDF and nothing else, 2b is the narrower, safer answer; if the repository asks you to attach data, code or the source Markdown, 3b is the variant that permits it.

What each variant allows

VariantBased onAttachmentsTagged structure
PDF/A-1bPDF 1.4NoNo
PDF/A-2bPDF 1.7NoNo
PDF/A-3bPDF 1.7YesNo
PDF/A-3aPDF 1.7YesYes
PDF/A-4e / 4fPDF 2.04f onlyOptional

Two things follow from that table and both matter for a thesis. First, 2b and 3b are not "more" and "less" compliant — a 2b file with no attachments is exactly as valid as a 3b file without them. Second, the "a" variants (3a) additionally require a tagged structure, which is a different requirement from archival compliance and belongs to the accessibility track.

Which one a repository usually asks for

Most university repositories and library systems were designed before PDF/A-3 and PDF/A-4, and their instructions name 1b or 2b. Read the instruction literally: if it says 2b, a 3b file may be rejected even though it is a superset in capability.

Three practical cases:

  • "PDF/A-1b" — this is the one to look at twice. 1b is based on PDF 1.4 and forbids transparency; many modern tools cannot produce it without rewriting the file. MarkdownGo does not produce PDF/A-1b: the exporter writes PDF/A-2b, 3b, 3a, 4e or 4f. If your repository insists on 1b, use the tool they provide or ask whether 2b is accepted.
  • "PDF/A-2b" — the default here. No attachments, native 1.7 file, embedded fonts.
  • "PDF/A-3b with data files" — only when they explicitly want the dataset or source attached.

What actually fails a PDF/A check

Variant choice rarely causes a failure. These five do:

  • Fonts not embedded. Every glyph in the text layer must come from an embedded program. Subsetting is fine; referencing a system font is not.
  • No Unicode mapping. A font without a ToUnicode entry still looks right and copies as garbage — the file can pass visually and fail extraction.
  • Transparency in 1b. Only relevant if you were asked for 1b; it is the main reason 1b is hard to produce from modern HTML rendering.
  • Missing XMP metadata. PDF/A requires an XMP packet declaring the conformance level; a plain /Title entry is not enough.
  • Images in unusual colour spaces or with interpolation enabled. PDF/A-2b 6.2.8 forbids /Interpolate on images; the Markdown export turns it off during its PDF/A clean-up pass, but a PDF built elsewhere may still carry it.

Exporting PDF/A from Markdown

  1. Set pdf_a: true in the document's front matter, or choose a variant in the export settings. The default is PDF/A-2b.
  2. Export from the studio. The PDF/A copy is generated natively — no Ghostscript rewrite — and 3a additionally gets a tagged structure.
  3. Keep the ordinary PDF as your reading copy; the archival copy is the one you submit.

How to verify before you submit

Two levels of checking, and they are not interchangeable:

  • The source check — the free checker reads your Markdown and reports missing alt text, skipped heading levels, empty links and a missing title. It says nothing about PDF/A; it catches the problems that are cheaper to fix in the file.
  • The file check — run the exported PDF through veraPDF or the repository's own validator. That is what decides the conformance level, and only that check can tell you whether the XMP claim matches the file.

The step-by-step checklist for the accessibility side is on the PDF/A guide and the thesis accessibility guide.

What this article does not do

  • It does not pick the variant for you. The repository's instruction wins over every recommendation here.
  • It does not convert an existing PDF. Everything starts from Markdown; a PDF built elsewhere has to be re-exported from its source, not patched.
  • It is not a guarantee of acceptance. A passing validator means the file-level requirements are met, not that a reviewer will accept the submission.

If your school is asking for tagged output instead of archival output, start with the thesis page and its localized samples. The two downloadable skeletons — thesis template and report template — both ship with the front matter that sets the export profile.