The archiving standard, and who supplies what
An archival PDF carries everything it needs to be read inside itself, so that opening it decades from now takes nothing but the file. This page says what the standard requires and what we write for you automatically; what is left for you to state is said on each of the eight pages of detail.
Summary
A PDF is not obliged to carry everything it needs. It may name a font it does not carry, give a colour as numbers without saying what device those numbers were chosen for, or point at a picture stored somewhere else. Today all of that works. But decades from now, the font may be gone from the machines, today's screen may be nowhere to be found, and the address of the picture may no longer answer. The document will then be impossible to display as it was written.
A standard was written to remove those three risks at once. It is called PDF/A, and its promise holds in one sentence: everything the document needs to be displayed is inside the document itself. Public archives, courts and accounting departments ask for it by name.
Most of it happens without being asked. You state that the document is archival: the engine carries the fonts in, describes the colours, writes the document's own description, and refuses to write the file as soon as it meets something the standard forbids, instead of letting it pass quietly. What stays with you is a short list, and every page of this section says exactly where the line falls.
The 10 claims
Two questions settle which one you want. Does the document have to be read out loud by software? Does it have to carry another file inside it? If neither, the plainest claim is the right one.
| Claim | Published text | Version it settles | What it adds | Value to send in an API request |
|---|---|---|---|---|
PDF/A-1B |
ISO 19005-1 | PDF 1.4 | The page looks the same for as long as the file exists, and part 1 is the strictest: no transparency, no layers, no file carried inside it. |
pdfa1b
|
PDF/A-1A |
ISO 19005-1 | PDF 1.4 | The same, and the document carries the structure of its own text, so it can be read aloud and searched by paragraph rather than by word. |
pdfa1a
|
PDF/A-2B |
ISO 19005-2 | PDF 1.7 | The page looks the same for as long as the file exists, and the document asks for nothing beyond that. It carries another file only where that file is itself archival. |
pdfa2b
|
PDF/A-2U |
ISO 19005-2 | PDF 1.7 | The same, and every letter drawn on the page maps back to the character it stands for, so the words can be searched for and copied out even where the document carries no structure of its own. |
pdfa2u
|
PDF/A-2A |
ISO 19005-2 | PDF 1.7 | The same again, and the document carries the structure of its own text, so it can be read aloud and searched by paragraph rather than by word. |
pdfa2a
|
PDF/A-3B |
ISO 19005-3 | PDF 1.7 | The page looks the same for as long as the file exists, and the document may carry another file of any kind inside it. This is the claim an electronic invoice is written under. |
pdfa3b
|
PDF/A-3A |
ISO 19005-3 | PDF 1.7 | The same, and the document also carries the structure of its own text, so it can be read aloud and searched by paragraph rather than by word. |
pdfa3a
|
PDF/A-3U |
ISO 19005-3 | PDF 1.7 | The page looks the same for as long as the file exists, the document may carry another file of any kind, and every letter drawn maps back to the character it stands for. |
pdfa3u
|
PDF/A-4 |
ISO 19005-4 | PDF 2.0 | The edition built on the current version of the format. It has neither level A nor level U: it is asked for without a letter. It says what it is in the description packet alone, where the earlier parts say it in two places, and it takes a carried file only where that file is itself archival. |
pdfa4
|
PDF/A-4F |
ISO 19005-4 | PDF 2.0 | The same edition, for a document that carries files of any kind. It asks for at least one. |
pdfa4f
|
What every archival document here carries
This holds whichever part is claimed. Nothing on it has to be asked for a second time.
- The document carries the fonts it uses, so the page displays the same on a machine where those fonts were never installed.
- An output intent is written whether or not one was asked for, so the greys and the reds mean the same thing on the next screen as on this one.
- The document's description — its title, the moment it was written, the part it claims and the year of that edition — is written into the document, and the claim is stated in the place the part being claimed requires.
- A report can be asked for before anything is written: it comes back with every finding at once, each naming the published text it answers to and the font, layer, page, link, field or cell it is about.
- A document that claimed a part and breaks one of its rules is refused rather than written wrong, and the refusal names the rule.
- Everything the document leads to is inside the document. Every part of the standard allows an instruction opening a file that sits beside it instead; the team declines that instruction under an archival claim, so the document stays readable on its own the day the folder around it is gone.
- The verdict belongs to two checkers that had no hand in writing the file, veraPDF and Ghostscript, and both are free to run on the document you download, from your own machine. Before a change enters the library or the service, every document claiming an archival part goes to veraPDF, and a refusal stops the check where it stands.
What every part of the standard rules out
Each of the bans below removes a way a document could stop being readable later; each edition's own page says which ones it adds and the few it relaxes.
- Encryption of any kind: an archived file opens without a password.
- Programs inside the document, and anything that launches a program.
- Sound, video and anything else that is not ink on a page.
- Anything the document does not carry itself: a font, a picture or a page that would have to be fetched from an address.
- A font named but not carried, whether it is one of the fourteen reading software is assumed to own or one of your own.
- An annotation that does not carry the drawing of how it is to appear.
The four editions, and the page each one has here
The family has been published four times, in 2005, 2011, 2012 and 2020, and every one of the four is still in force: whoever receives the document names the edition they require. Each has a page of its own here, saying what that edition asks for, what it rules out, and where it stands with us today.
-
PDF/A-1
—
ISO 19005-1, 2005 The first edition, published in 2005, is the one that set the rule the four editions have applied ever since: everything the document needs to be read is inside the file. The fonts are in it, the colours are described in it, the pictures are included in it. Nothing to fetch elsewhere, neither over the network nor on the machine that opens it. -
PDF/A-2
—
ISO 19005-2, 2011 The second edition, published in 2011, holds to the same rule as the first and lets a document use what the format had gained in the six years between them. -
PDF/A-3
—
ISO 19005-3, 2012 The third edition, published in 2012, differs from the second by a single clause — and that clause is the reason electronic invoicing across Europe is built on it. -
PDF/A-4
—
ISO 19005-4, 2020 The fourth edition, published in 2020, is the first built on the current version of the PDF format. It simplifies the claim rather than adding to it.
Technically
What the claim settles document-wide, the two moments the same rules run at, and the values the request format takes today.
The claim is a document-level statement. It settles the PDF version, it makes the XMP metadata be written, it makes an output intent be written even where you stated none, and it makes blocking a set of rules that, without it, would only produce findings.
There are two moments and they run the same rules. A report can be asked for on the document as composed — nothing is written, nothing is changed, and the document need not have claimed anything — and it comes back with every finding at once, each naming the published text, what stands in the way and which font, layer, page, link, field or cell it is about. Then, at write time, a document that claimed a part of the standard is refused on the first finding rather than written wrong.
Through the service, the field to send is called archive, and it takes the ten values the engine knows how to write: pdfa1b, pdfa1a, pdfa2b, pdfa2u, pdfa2a, pdfa3b, pdfa3u, pdfa3a, pdfa4 and pdfa4f.
Nothing here calls a document conforming. An empty report is not a certificate: what the engine does not know to look for, it does not report. The verdict belongs to a checker that did not write the file — one that every archival document goes to before a change enters the library or the service, and that you can run on your own copy without asking the team anything.
Family by family
Eight pages, one family of requirements each, at the same three reading levels as this one: the summary, Technically, then Clause by clause. Each ends on the published texts it answers to.
The requirements group into eight families. Take them in this order and each one rests on the last: choose the claim, then the fonts, then the colours, then the page itself, then what its marks stand for, then what it carries, then what it says about itself, and last how the whole thing is checked.
The eight families
- Which part of the archiving standard to claim The archiving standard is a family, not a single rule. Four of its parts can be claimed here, ten claims between them, and choosing takes two questions.
- Every font the document uses is embedded in the file A document that carries its own fonts opens the same way on a machine that has never had that font installed. That is the first thing the standard asks for, and the one that changes the file most.
- Colours that mean the same thing on every device Numbers on a page become a precise colour only once the device they were chosen for is named. An archival document carries that device's profile inside it.
- A page that comes out the same for everyone The promise of an archival document is that the page read today is the same page decades from now. It holds only where the file settles everything the page shows, leaving nothing to the reading software.
- The document says what its marks stand for The content of a page records where each letter is drawn, not what it means. Level A of the archiving standard and the accessibility standard (ISO 14289) ask the document to say it: which group of words is a title, which is a paragraph, and in what order to read them.
- A document that carries other files inside it ISO 19005-3 (PDF/A-3) exists because a document sometimes has to carry another file inside it, and that file has to come back out years later, whole.
- What the file says about itself A document still opened decades from now is found again by what it says about itself, not by the folder it happened to be in.
- Checking the claim, before and after Claiming a standard is one line inside the file. That line is not enough on its own: for the claim to be worth anything, an independent checker — one you run yourself, and one the team did not write — has to accept it.
Where to go next
Why archiving requirements exist, and who is obliged to follow them Which colour profiles are accepted as an output intent The electronic invoice, which is an archival document carrying its own data How a request asks for an archival document
See a document that carries its own structure See an archival form