BASALT · JOURNAL

Bates numbering on a Mac without Acrobat

2026-08-06 · bates numbering mac

Bates numbering is a sequential unique identifier stamped on every page of a production so that any page can be cited without ambiguity. It sounds like a formatting task. It is closer to assigning primary keys, and the failure modes are the ones you would expect from a database with no uniqueness constraint.

macOS ships nothing that does it. Preview cannot, Automator's PDF actions cannot, and Quartz filters operate on rendering, not on stamping sequences. Here is what the job actually requires and how to do it on a Mac.

What the number has to guarantee

Three properties, and every rule below follows from them.

Unique. No number appears on two pages, ever, across the whole matter. Not just within a volume. If you produce volume one ending at ABC000450, volume two starts at ABC000451.

Sequential and gapless. The receiving party uses continuity to know they have everything. A gap reads as a withheld page, which is fine only if that is what you meant and it is on the privilege log. Accidental gaps generate correspondence.

Permanent. The number must survive downstream handling: copying, printing, re-saving, extracting a page range. A citation to ABC000112 has to resolve to the same page a year later.

Everything else, the prefix, the padding width, the corner it sits in, is convention. Useful convention, but convention.

Format conventions worth following

A prefix identifying the producing party, usually two to five letters, then a zero-padded number. SMITH000001. Pick the padding width from the expected size of the whole production and add a digit. Six digits handles a million pages and costs nothing; four digits looks tidy until the production hits 10,000 pages and you have to choose between an inconsistent format and a renumber.

Place the number in the bottom-right margin. It is the conventional position and, more practically, it is the least likely corner to collide with content. If a page has content to the edge, move the stamp for the whole production, not for that page, so the position stays predictable.

Keep confidentiality designations separate from the Bates number. A protective order designation can change during a matter; the Bates number cannot. Stamping "CONFIDENTIAL" as part of the same string forces a renumber if the designation is later withdrawn.

The page geometry traps

This is where naive stamping scripts go wrong, and the failures are visible only after you have stamped a few thousand pages.

MediaBox origin is not always (0, 0). The spec allows any rectangle. A page whose MediaBox is [0 0 612 792] and one whose MediaBox is [-12 -12 600 780] describe pages of the same size in different coordinate frames. A stamp drawn at a fixed offset from the origin lands in a different place on each.

CropBox governs what is displayed. If a page has a CropBox smaller than its MediaBox, the visible page is the CropBox. Compute the stamp position from the CropBox, falling back to the MediaBox when there is no CropBox, or your numbers will drift off the visible page on cropped pages.

/Rotate rotates the rendered page, not the coordinate system. A page with /Rotate 90 displays sideways relative to its content coordinates. A stamp placed at the bottom-right of the content space appears in a different corner, rotated, in the viewer. You have to apply the inverse rotation to the stamp's transformation matrix so the number reads horizontally in the corner the reader sees.

Page sizes vary within a document. Productions mix letter, legal, A4, and scanned pages of odd dimensions. Any position hardcoded in points rather than computed per page will be wrong on some of them.

Existing content state. Appending drawing operators to an existing content stream inherits whatever graphics state that stream ended in: a clipping path, a non-default color, a transformation matrix. Wrap the stamp in q and Q to save and restore the graphics state, and reset the matrix explicitly, or your number may be clipped away, invisible, or drawn at a strange scale.

Content stream, not annotation

You can put text on a page two ways: append operators to the page content stream, or add a free text annotation.

Use the content stream. An annotation is a separate object that any PDF editor can delete in one action, that may or may not print depending on its flags, and that some processing pipelines strip wholesale. A Bates number that a downstream tool can silently remove violates the permanence property.

If you also want the number to survive text extraction, so a receiving party can search for ABC000112 and land on the right page, it has to be real text with a font resource and a ToUnicode mapping, not a path-drawn outline or a rasterized image of the number.

Doing it with command line tools

There is no single Mac tool for this, so the common approach is to generate an overlay and merge it.

Generate a one-page-per-page overlay PDF where each page contains only the number, positioned correctly, then merge it onto the source. pdftk-java calls this multistamp:

pdftk-java source.pdf multistamp overlay.pdf output numbered.pdf

The overlay itself is where the work is. A Python script using pypdf to read each page's CropBox and /Rotate, and reportlab to draw the corresponding overlay page at the right size and offset, is maybe eighty lines. That script has to handle the geometry cases above; if it does not, it works on your test file and fails on the production.

For a production spanning multiple documents, the counter has to persist across files in the order the documents will be produced, which means the script drives the whole set, not one file at a time. Sorting matters: a naive directory listing sorts doc10.pdf before doc2.pdf, and a production numbered in that order is technically valid and practically confusing.

After stamping, verify: extract the text of every page, pull the Bates string, and assert that the sequence increments by exactly one with no repeats. That check takes a few lines and catches a mis-sorted set, a restarted counter, and a page that failed to receive a stamp.

Order relative to redaction

Endorse before you redact.

The privilege log has to point at Bates ranges, and a redaction log generated from marks placed on already-endorsed pages is consistent by construction. Endorse afterward and you have two independent passes whose agreement is assumed rather than enforced. This ordering is part of the broader sequence in a redaction workflow for law firms, and it is what makes building a privilege log from the redactions you already made possible.

One consequence to plan for: because the number is applied first, a redaction tool that rewrites the page content must preserve the stamp. That is normal if the redaction only removes the glyph runs intersecting a mark, since the stamp sits in the margin. It is not normal if the tool rasterizes and re-emits pages, which can degrade the number's searchability.