Convert PDF to PDF/A for Archiving

How to convert a PDF to PDF/A for an archive

The requirement to file a document as PDF/A shows up wherever a file has to outlive its software: in state archives, court systems, thesis deposits, and corporate records with long retention.

The point of the format is self-sufficiency. An ordinary PDF may reference a font installed on the system, or lean on a monitor's colour profile. In twenty years neither may exist, and the document will open as something other than what it was. PDF/A forbids such external dependencies: everything needed sits inside the file.

What conversion actually does

Ghostscript performs the transformation, and it is not a cosmetic edit of the metadata.

Fonts are embedded into the file in full, so the document does not depend on what is installed on the reader's machine. Colours are brought into a described profile, and an output colour intent is attached to the file, from which the intended shade can later be reconstructed unambiguously. The conformance level is written into the metadata.

It follows that a file after conversion almost always gets heavier than the original. Embedded fonts and a colour profile take space. That is normal, the price of self-sufficiency rather than a sign that something went wrong.

Which level to choose

There are three levels, and the middle one is the default.

The letter b in all three means the basic level: the document is guaranteed to look the same always. There is also level a, with requirements for structure and accessibility, but it demands so much of the source document that automatic conversion of an ordinary scanned certificate will not reach it.

If the archive's requirements name a level outright, take the one they name. If they simply say "PDF/A", take 2b. Level 1 is worth choosing deliberately: it is based on an older version of PDF and supports neither transparency nor layers, so a document with modern graphics changes more when converted to 1b.

The service checks its own output

This deserves stating separately, because not every converter does it.

After the transformation the file is checked for the mandatory markers of PDF/A: whether the conformance level is declared in the metadata, and whether it matches what was requested. The result is not declared conformant simply because the conversion finished without an error.

If the check does not pass, the file is still returned, but honestly, as an ordinary PDF, with an explanation of what was missing. It is usually still better than the original; calling it PDF/A would just be untrue. The metadata is not faked either: conformance markers are not stamped onto a file that failed the check, because such a marker would mislead both a person and an intake system.

What that check is not

Here it matters not to overrate the guarantee.

The check looks at format markers rather than performing full formal validation against ISO 19005. Full validation requires a separate tool, and the accepted one is veraPDF. The difference is substantial: markers confirm that the file declares itself as PDF/A and was assembled accordingly, while formal validation checks every requirement of the standard point by point.

For an internal archive and for most organisations the first is enough. If the document is headed somewhere non-conformance carries legal consequences, run the finished file through veraPDF separately. The service says as much in the message attached to the result.

Step by step

1. Check with the recipient which level is required. 2. Open pdf-to-pdfa and upload the document. 3. Choose the conformance level, 2b by default. 4. Set a meaningful document title: it goes into the archival metadata and people will search on it. 5. Run the conversion and read the message attached to the result rather than just downloading the file.

Step five is not a formality. That message is precisely where it says whether conformance was confirmed or the file came back as an ordinary PDF.

Why an archive might still reject the file

Format conformance and acceptance by a particular archive are different things, and the second is stricter.

Most often the cause is requirements the standard itself does not contain. An archive may demand a specific level, a particular file naming scheme, a mandatory set of metadata fields, or a size limit. The file will be a flawless PDF/A and still come back.

The second frequent cause is that the document conforms formally but does not contain what it was filed for. A scan with no text layer, converted to PDF/A, remains a set of pictures. A requirement for "searchable PDF/A" is not met by such a file, even though both words about the format hold true.

The third cause is subtle: conformance may have failed to confirm and nobody noticed. The file was downloaded, seen to open, and sent. Meanwhile the message attached to the result said the check had not passed and the file came back as an ordinary PDF. Which is why it is worth reading.

If the archive's requirements run to several pages, it is wiser to put one trial file through the whole chain once and get their confirmation than to convert the entire body of documents and learn about non-conformance a month later.

What to prepare in advance

Conversion works with what it is given and does not fix problems in the source document.

A scanned document stays a set of pictures. Formally it becomes PDF/A, but you will not be able to search text in it. If the archive expects a document with a text layer, run the scan through ocr-pdf first and convert afterwards.

A password-protected document will not convert until the open protection is removed. unlock-pdf handles that. Worth remembering too that PDF/A and encryption are incompatible by the standard itself: an archival document must open without a key.

The title and other properties are easier to put in order before conversion with set-pdf-metadata than to edit afterwards in a file already declared archival.

If there are many documents, do not run them as a batch blindly. Conformance level and source preparation differ by document type: scans need recognition, documents with transparency sit poorly in level one, password-protected files will not pass at all. It is wiser to sort the body by type, settle the settings on one representative of each, and only then run the rest.