What Is PDF/A? The Archival PDF and What OCR Adds to It

PDF/A is a restricted form of PDF for documents that must still open and look the same many years from now. ISO publishes it as the standard ISO 19005. A scanned page with an OCR text layer can be PDF/A, but only if the file is saved to the standard's rules and passes a validator.

What is PDF/A and who defines it?

PDF/A is a set of rules that narrows ordinary PDF so a file depends on nothing outside itself. The International Organization for Standardization publishes the rules as ISO 19005, in four parts issued between 2005 and 2020. A PDF/A file is still a PDF, and any PDF reader opens it.

The Library of Congress keeps a plain-language record of the format. Its description of the PDF/A family says the standards are developed and maintained by a working group, WG 5 of ISO Technical Committee 171, with representatives from government, industry and academia and support from Adobe. The Sedona Conference's glossary of eDiscovery terms gives the purpose in one sentence: PDF/A "preserves electronic documents visual appearance over time, independent of the tools and systems used for creating, storing, or rendering the files."

The idea behind every rule is self-containment. An ordinary PDF may borrow a font from the computer it is opened on, point to a file on a server, run a script, or sit behind a password. Each of those can fail in twenty years: the font is gone, the server is gone, the script will not run, nobody remembers the password. PDF/A removes the dependencies so that the page a clerk sees in 2046 is the page you saved.

Three things PDF/A is not:

• Not a different file type. The extension is still .pdf. You cannot tell a PDF/A file from its name.
• Not a promise about content. The standard governs how the file is built. It says nothing about whether the words are accurate or the scan is legible.
• Not a lock. Software can still edit a PDF/A file. Once edited and saved, the file may no longer conform.

What does PDF/A forbid and require inside a file?

PDF/A forbids anything that needs outside help or could change what the reader sees, and requires the file to carry everything used to draw its pages. In short: no encryption, no scripts, no audio or video, and all fonts embedded.

The Library of Congress lists the constraints shared by PDF/A-1, PDF/A-2 and PDF/A-3:

The encryption rule surprises people most. A password-protected PDF cannot be PDF/A, so "secure it and archive it" are two requests that one file cannot satisfy. If a document must be both protected and preserved, the protection has to come from where the file is stored.

Notice what the list leaves out. PDF/A does not forbid document metadata, comments or an invisible text layer. A file can conform to the standard and still carry an author name, a markup note, or words sitting under a black rectangle. Conformance is not a privacy check, and a file bound for a long-lived archive deserves one. The post on hidden text in a PDF covers what to look for before a file leaves the firm.

PDF/A-1, PDF/A-2 and PDF/A-3 and the letters after them

The number is the part of the standard, and the letter is the conformance level. PDF/A-1, 2 and 3 differ in which PDF features they allow. The letters a, b and u say how much the file promises beyond its appearance. "PDF/A-2b" means part 2, level b.

The parts

A later part does not retire an earlier one. The Library of Congress page on PDF/A-2 says it "was not intended to make PDF/A-1 obsolete, but to add support for features that would be valuable in some circumstances." A PDF/A-1b file made in 2009 is still a valid PDF/A-1b file.

PDF/A-3 is the part to be careful with. Because it lets any file ride inside the PDF, a spreadsheet or an email can travel with the page that summarizes it. The Library of Congress page on PDF/A-3 records the unease of archives about that, including a 2014 report from the National Digital Stewardship Alliance that recommended PDF/A-2 for documents with nothing embedded. If you have no attachment to embed, you have no reason to choose part 3.

PDF/A-4 replaces the levels with two profiles. PDF/A-4f allows embedded files in any format and succeeds PDF/A-3. PDF/A-4e is intended for engineering documents.

The levels

• Level b covers visual appearance only. The page will look the same. Nothing is promised about the text or its structure.
• Level u, introduced with PDF/A-2, is level b plus the requirement that all text in the document have Unicode equivalents. Text can be searched and copied reliably.
• Level a satisfies every requirement in the standard, including the structure of the document, which is what tagging records.

PDF/A-1 has only levels a and b. The Library of Congress adds a warning about level a: a file that conforms to it does not necessarily meet the separate accessibility standard, PDF/UA. Tagged and accessible files are explained in the post on how to make a PDF readable.

Can a scanned page with an OCR text layer be PDF/A?

Yes. A scanned page is a picture, and PDF/A allows pictures. Adding recognized text behind the picture does not break the standard. Level b is the level typically used for files made from scanned pages, and whether a given file conforms depends on the software that saved it.

Two statements from the Library of Congress frame the question. Its family description says level B conformance "is the level typically used for PDF/A files created from scanned pages." Its page on PDF/A-1 says the format does not preclude scanned page images at level b, but that "such files do not necessarily support indexing of the document text or extraction of text for quotation."

Put together: PDF/A and searchable are separate properties. A scan saved as PDF/A-1b with no OCR is a valid archival file that nobody can search. OCR supplies the text. PDF/A supplies the guarantee about appearance. For a closed matter you will want both, and you have to ask for both. What a searchable file holds is covered in the post on the searchable PDF.

Store a court opinion as PDF/A and you may still have nothing to search. Add text if you need the holding, the citation, or a statute the opinion quotes, then check those strings against the picture. You read the precedent named in that citation from the page, after the conformance report is done.

One federal bankruptcy court states the choice of level for scans directly. The PDF/A questions and answers of the Bankruptcy Court for the Eastern District of Pennsylvania say PDF/A-1a "requires structure in a document and is best to use for electronic documents", while PDF/A-1b "does not require structure in a document and is best to use for scanned documents or documents where the structure is unknown."

A working order for a scanned file

Take a fictional example. The firm Okafor Lindqvist is closing the file in Marsh v. Tollgate Storage and its retention policy calls for PDF/A. The file includes 140 scanned pages of correspondence.

1. Keep the scan as received. Work on a copy.
2. Run OCR and save a searchable PDF. Check the text with a few searches for names and amounts you can see on the pages.
3. Convert the searchable PDF to PDF/A in a program that offers it, choosing level b, or level u in part 2 if the program supports it.
4. Validate the converted file. A save option is not proof.
5. Compare a page. Conversion can recompress images or change colors. Put a page with a signature or stamp next to the original and look.
6. Search again. Confirm the text layer survived the conversion.

Schema Law's part in that sequence is step 2. Its searchable PDF export is your original file with a hidden text layer. Do not assume that export, or any OCR tool's output, is PDF/A until a validator has said so.

Which courts and archives ask for PDF/A

Whether PDF/A is asked for depends on the court. Idaho's electronic filing rule accepts a text-searchable PDF or PDF/A, and the federal filing system accepts files that meet the standard. The U.S. National Archives lists PDF/A among its formats for digitized permanent records. Read your own court's rule.

A few examples show the range. They are illustrations, and each may have changed since we read it.

• A state rule that accepts either. Idaho's Rule 6 of the Rules for Electronic Filing and Service calls for a text-searchable PDF or a text-searchable PDF/A file, converted directly to PDF where possible. It excepts scanned attachments and exhibits, which may be standard PDF and need not be text-searchable.
• The federal filing system. The Administrative Office's guide to PDF functionality for public filers in NextGen CM/ECF lists a file that "Meets the PDF-A standard" among those accepted. The same guide lists what the system rejects, including files that contain JavaScript, are encrypted or password-protected, contain internal attachments, or contain audio and video. That overlaps heavily with what PDF/A forbids. The guide repeats that local courts may add requirements of their own.
• The National Archives. NARA's tables of file formats for transferring permanent records tell agencies digitizing paper to "Select any version of PDF/A that meets project requirements" and not to use the attachments feature in PDF/A-3 or PDF/A-4.

To find your own requirement, look in three places: the court's electronic filing rule or administrative procedures, the technical page of its filing portal, and any standing order of your judge. Search each for "PDF/A", "archival" and "text-searchable". The last term has its own meaning in court rules, set out in the section on what a court means by text-searchable.

If a client's outside counsel guidelines or records schedule asks for PDF/A, ask which part and level they expect. "PDF/A" with no number is an incomplete instruction.

How do you check that a file really conforms to PDF/A?

Run the file through a PDF/A validator and read the report. A file can announce itself as PDF/A and still break the rules, because the announcement is only a marker that the saving program wrote into the file's metadata. Validation tests the whole file against the standard.

One validator is free and built for the purpose. veraPDF describes itself as "a purpose-built, open source, file-format validator covering all PDF/A and PDF/UA parts and conformance levels". The veraPDF getting-started documentation describes a graphical program for trying files by hand and a command-line version for automating large batches. Some commercial PDF editors include their own conformance checks.

A routine that fits a law office:

1. Validate against the level you were asked for. A file that passes 2b has not been tested for 1b.
2. Save the report with the file. The date and result are your record that the file conformed when you stored or filed it.
3. Read the first failure. Check it against the rules the standard sets: a font that is not embedded, a leftover form script, transparency in a part 1 file, or encryption.
4. Fix at the source if you can. Re-export from the word processor with the PDF/A option, or convert again from the unprotected original. Repairing a converted file is harder than making a clean one.
5. Validate again after any change. Adding a Bates stamp, a signature or a page can break conformance.

Two limits are worth stating. A validator checks rules. It does not check that the right exhibit is in the file or that the OCR text matches the page. And a banner in your PDF reader saying the file "claims compliance" with PDF/A is reporting the marker, which is a claim, and has run no test.