PDF/A parts and conformance levels
A PDF/A claim always has two halves. The part (1, 2, 3 or 4) says which edition of the standard the file follows. The conformance level (a, b or u) says how much it promises. "PDF/A-1b" is part 1, level B.
The four parts
| Part | Standard | Based on | Levels | What it adds |
|---|---|---|---|---|
| PDF/A-1 | ISO 19005-1:2005 | PDF 1.4 | a, b | The original. The most restrictive: no transparency, no layers, no JPEG 2000, no embedded files. |
| PDF/A-2 | ISO 19005-2:2011 | PDF 1.7 (ISO 32000-1) | a, b, u | Transparency, layers, JPEG 2000, and embedding other PDF/A files. Adds level U, which is level B plus Unicode mappings for all text. |
| PDF/A-3 | ISO 19005-3:2012 | PDF 1.7 (ISO 32000-1) | a, b, u | PDF/A-2 plus the ability to embed files of any format. The basis for hybrid e-invoices such as ZUGFeRD and Factur-X. |
| PDF/A-4 | ISO 19005-4:2020 | PDF 2.0 (ISO 32000-2) | none (plus PDF/A-4e, PDF/A-4f) | Drops the a/b/u lettering: conformance is simply PDF/A-4, with optional -f for embedded files and -e for engineering content. Requires Unicode mappings outright. |
The three levels
Level A implies level U, while level U implies level B. Each level is a superset of the one below it.
- b, basic: Reliable visual reproduction. Everything needed to render the page correctly is in the file.
- u, unicode: Level B plus a Unicode mapping for every glyph, so text can be extracted and searched reliably. PDF/A-2 and -3 only.
- a, accessible: Level U plus a tagged structure hierarchy and a defined reading order, so the document is machine-understandable rather than only machine-renderable.
Level A requires document tagging and is considerable authoring work, as no tool can invent a correct reading order for a scanned page or a complex table. If a requirement says "PDF/A" without a level, it almost always means level B.
Choosing one
If someone told you which one, use that. Archives, courts and tax authorities name a part and level for a reason, and they are the ones doing the validating.
If nobody told you, PDF/A-1b. It is the strictest and the most universally accepted, so it is the safest thing to have produced. Attaching machine-readable data is the exception: that needs PDF/A-3, covered in the e-invoicing guide. And if you need accessibility use level A.
What gopdfrab covers
For now, just PDF/A-1b. The profile covers ISO 19005-1 clauses 6.1 to 6.9, and it passes the full Isartor suite and the veraPDF corpus.
Frequently asked
What is the difference between PDF/A-1a and PDF/A-1b?
Level B ("basic") guarantees the page renders reliably. Level A adds accessibility: a tagged structure hierarchy, a logical reading order, and Unicode mappings for all text, so the document can be reflowed and read aloud. Level A is strictly harder and much rarer.
Which PDF/A version should I use?
PDF/A-1b if the receiving side does not say. It is the most widely accepted and the most restrictive, so a PDF/A-1b file is generally acceptable where a later part is expected. PDF/A-3 if you need to embed a machine-readable payload, such as an e-invoice.
Is a PDF/A-1b file also a valid PDF/A-2b file?
In practice yes, because PDF/A-2 relaxes constraints rather than adding them. It is not automatic in the other direction: a PDF/A-2 file using transparency or JPEG 2000 will fail PDF/A-1b.
Which parts does gopdfrab support?
PDF/A-1b only, for both verification and conversion. PDF/A-2, PDF/A-3 and PDF/A-4 are on the roadmap and not implemented.