PDF/A

What it restricts

PDF/A is ordinary PDF with the parts removed that stop a document rendering identically in fifty years. Every font must be embedded, so the file does not depend on the reader having it. Colour must be unambiguously specified, through an embedded profile or a device-independent space. Encryption is forbidden, as are JavaScript, executable content, external file references and audio or video.

The result is a file that carries everything it needs. Open it on a machine that shares nothing with the one that made it and the pages look the same.

The conformance levels

The part number identifies the PDF version it is based on — PDF/A-1 on PDF 1.4, PDF/A-2 on ISO 32000-1, PDF/A-3 allowing arbitrary embedded attachments.

The letter identifies how much structure is present. Level B, for basic, guarantees only that the visual appearance is preserved. Level A, for accessible, additionally requires tagged structure and a real reading order, which is what makes the document usable by a screen reader and reliably extractable as text.

A scanned page saved without OCR can meet level B and cannot meet level A, because an image of text has no text in it.

When it is actually required

Court filings, government submissions, regulated record-keeping, institutional archives and some publishing workflows specify it. If nobody has asked for PDF/A, an ordinary PDF with its fonts embedded gets you most of the durability for none of the constraint.

The place it bites is scanned documents: to reach level A you need a text layer, which means running OCR and embedding the result behind the page image rather than shipping the image alone.