What a PDF keeps after you delete a page

Deleting a page in most PDF editors does not remove it from the file. A PDF can be saved by appending: the changed parts go after the existing bytes with a new cross-reference section pointing back at the old one, and nothing earlier is taken out. The reader shows you the newest version and hides the rest. Open the file in STRATA, which walks that chain and shows every earlier version still inside it (in the browser, with nothing uploaded), and then build a fresh document with STAPLE if you need the page genuinely gone.

Why a deleted page can still be in the file

A PDF is not rewritten from scratch every time it is saved. The format allows an incremental save: the objects that changed are written after everything that was already there, followed by a fresh cross-reference table whose /Prev entry points back at the previous one. The result is a file that ends with the current document and begins with every version it used to be.

This is not a defect. It is what makes signing a PDF possible at all (a signature covers a byte range, and later changes go after it so the signed bytes stay signed), and it is why a large document saves instantly instead of rewriting two hundred megabytes.

It does mean that “delete page 4, save” can produce a file which still contains page 4, plus a note saying it is no longer part of the document. Every reader honours the note. A reader that did not, or a tool that walks the chain, does not have to.

What else survives the same way

Pages are the obvious case and rarely the one that causes trouble. The same mechanism keeps:

• Comments and annotations that were removed, including the ones somebody wrote to a colleague before the file went outside.
• Form field values from earlier fills, which is how a form template can arrive carrying the previous applicant's details.
• The earlier state of anything that was edited: a figure, a date, a bank account number, a name.
• Redactions applied as black rectangles. A rectangle drawn over text is an object on top of the text, not a removal of it, and in an incrementally saved file the pre-redaction version may be present twice over. This is the mechanism behind a long run of published-document embarrassments.

How to find out what is in a file you have

1. Look at it. STRATA finds every version boundary in the file and, for each consecutive pair, compares the document properties, the fonts each page uses, the decoded page content, the annotations, the form values, the signatures and how many bytes each one actually covers, and renders every page at the same scale to compare them pixel against pixel. It runs in the tab; the file is not uploaded, which matters because the documents worth checking this way are contracts, disclosure bundles and statements.
2. Check the size. A three-page PDF of two megabytes is telling you something. Not proof (images are large), but worth a look.
3. Try the crude test. Open the file in a text editor and search for %%EOF. A PDF that has been saved once contains it once. Several occurrences means several saves, each with whatever it carried. This finds the common case and misses the rest, which is why it is third.

Getting the page genuinely out

The reliable answer is to produce a new file rather than to edit the old one, because a new file has no chain to carry.

• Export or “save as” rather than save, where your editor offers it. Some editors rewrite the whole file on Save As and append on Save, and the difference is invisible in the interface.
• Extract the pages you want into a fresh document. STAPLE builds a new PDF from the pages you choose, so what was not chosen was never in the output.
• Flatten it. FLATTEN burns filled-in form fields into the page, so there is no field left holding a value.
• For a real redaction, remove the content rather than the view of it. A box over text is not a redaction. Take the words out before the PDF is made, or cover the area in the picture and rebuild the page from pixels, which is what SCREENSHIELD does for a screenshot.

Then check the result with STRATA before it goes out, rather than assuming. That step takes thirty seconds and is the only part of this that verifies anything.

If you are on the receiving end

The same mechanism works in your favour when a document arrives and the question is what it used to say. A disclosure bundle, a contract that came back amended, a report whose figures moved: the earlier versions are frequently still in the file, along with who saved it and when, and nobody sending it had any idea.

Where the file matters, record what you received before you examine it. PROVENANCE takes a digest and a note of where it came from, so the copy you analysed can be shown later to be the copy you were given. LINEAGE does the equivalent job for a Word file, which keeps its own editing history by a different mechanism.