Amazon Kdp Publishing

Amazon KDP Publishing

Complete reference for producing and verifying KDP-ready files across every KDP product type, plus a repeatable manuscript QA workflow for long fiction. Built from hands-on verification (real EPUBCheck runs, real page-size/font audits) — trust the numbers here over general knowledge, but re-verify anything price- or policy-related with a quick search, since KDP changes these periodically.

Step 0 — Identify the product type first

Everything downstream (trim size, bleed decision, file format) depends on what's being made. Ask if it's not obvious:

A single title can need up to three separate files: a print interior PDF, an ebook EPUB, and a print cover PDF — each with different rules. Never assume one file satisfies two formats.

Step 1 — Build to the right specs the first time

Read the relevant reference file(s) above before generating anything — margins, page-count math, and bleed decisions are cheap to get right up front and expensive to redo after a rejection. references/paperback-hardcover.md has the exact spine-width and full-cover-size formulas; use scripts/spine_calculator.py to compute them rather than doing the arithmetic by hand.

For the actual document build (docx/epub/pdf mechanics — reference-doc styles, page breaks, footers, gutter/mirror-margins XML, EPUB CSS, pandoc invocations), use the docx, pdf, and general markdown→pandoc techniques already in your toolkit; this skill tells you what values to plug in, not how to drive pandoc. The one KDP-specific gotcha worth restating: LibreOffice's docx→pdf export can drift the page size by ~1pt from what you specified (a twips-to-points rounding artifact) — always verify the exported PDF's exact page size with pdfinfo and hard-correct the MediaBox with pikepdf if it doesn't match the trim size exactly (see scripts/fix_pdf_pagesize.py).

Step 2 — Run the compliance checklist before telling the user it's ready

Don't just assert a file is compliant — verify it, the way you'd verify code by running it. For each file type:

Print interior PDF:

1. pdfinfo file.pdf | grep "Page size" — must exactly match trim size × 72 (e.g. 6×9" = 432×648pt). Fix with scripts/fix_pdf_pagesize.py if off by any amount.
2. pdffonts file.pdf — every font row should show emb: yes. If a font isn't embedded, KDP may substitute unpredictably.
3. Page count: pdfinfo file.pdf | grep Pages. Confirm ≥ 24 (paperback minimum), and pad to an even number with a blank page for clean binding (not a hard requirement — KDP auto-rounds — but best practice).
4. No encryption: pikepdf → pdf.is_encrypted must be False.
5. No crop marks, no bleed unless the product actually needs it (see Step 0 table).
6. Margins/gutter meet the minimums in references/paperback-hardcover.md for the actual page count.

Kindle EPUB:

1. Run real EPUBCheck, not just an XML well-formedness check — download it if not present (references/ebook-kindle.md has the exact command). A file can be valid XML and still fail EPUBCheck's structural rules.
2. Confirm nav.xhtml (HTML TOC) and toc.ncx both exist.
3. Confirm dc:title, dc:creator, dc:language are set correctly in content.opf.
4. Confirm the cover is registered with properties="cover-image".
5. No page numbers/headers baked into the CSS — Kindle is reflowable, those don't belong.

Print cover PDF (wraparound):

1. Dimensions must match the exact output of the spine calculator for the final page count, paper color, and binding (paperback vs hardcover) — recompute if the interior page count changes after the cover was built. This is the single most common rejection cause.
2. 0.125" bleed on all outer edges, all text/logos inside the 0.25" safe zone from trim.
3. CMYK color space, 300 DPI.
4. No crop marks/color bars in the delivered file.

Report results as a table (spec vs. actual, pass/fail) the way you would code review output — don't just say "looks good."

Step 3 — Manuscript QA workflow (for novels / long fiction)

Independent of file formatting, a long manuscript needs a content consistency pass before it's truly ready to publish. This is a distinct skill from formatting — do this on the text itself, in phases, before final file generation. Full method, phase breakdown, and the categories of errors most worth checking → references/manuscript-qa-workflow.md.

Quick summary: build a "story bible" of names/ages/dates/rules as you write or before you check; then re-read the manuscript in 2–4 chunks (never all at once) cross-referencing every number and relationship claim against it; log every finding before fixing anything; fix in small verified batches; rebuild and re-render deliverables only after all fixes land. This process caught (and fixed) real timeline and genealogy errors in prior work — it's worth the extra pass on anything over ~20,000 words.

Step 4 — ISBN, categories, rights

Quick reference only — verify current numbers before quoting a price. Full detail → references/isbn-and-metadata.md.

• KDP gives a free ISBN for paperback/hardcover only (not ebook, not required for low-content). It lists "Independently published" as publisher and can't be used outside KDP.
• Each format (paperback, hardcover, ebook, audiobook) needs its own ISBN if using paid ISBNs (Bowker in the US, ~$125 single / ~$295 for 10).
• Low-content books (journals, planners, notebooks) get a "Low-content" checkbox during KDP setup that changes some requirements — flag this to the user if the product is low-content.

Working notes / things this skill got right in practice

• The 6×9" gutter/margin XML for python-docx-generated files: pgMar with a separate w:gutter attribute, plus <w:mirrorMargins/> in settings.xml (not in sectPr — a common mistake, it belongs in settings, not the section properties).
• w:titlePg + two footerReference entries (type="first" blank, type="default" with a PAGE field) cleanly suppresses the folio on the title page without needing multiple sections.
• Pandoc's EPUB writer auto-wraps any pre-first-heading content (title page, copyright, dedication) into an implicit section titled after the doc's metadata title, which renders as an unwanted duplicate heading — hide it in CSS with section#<slugified-title> > h1 { display: none; }, not by hiding the whole section (that hides the front matter too).