Publication

Turning a directory of chapters into something a print-on-demand service will accept.

Two jobs, deliberately split

Job	Owned by	Why
Manuscript — concatenation, body typesetting, profile conformity	This pattern	It is downstream of authorship and needs the manifest
Print geometry — trim, gutter, spine, cover wrap, pre-flight	kdp-publishing	It is downstream of the finished interior, and it is already solved

The split is not administrative. A cover cannot be built until the interior is final, because spine width is page count multiplied by paper caliper. A cover built from an estimated page count produces a book whose spine type sits off-centre, and nothing on screen reports it — you find out when the proof copy arrives.

So print geometry is handed off, last, to a tool that refuses to take a page count and measures the built PDF instead. Reimplementing that here would mean maintaining a second copy of the one calculation in this pipeline where being approximately right costs a print run.

The division runs both ways. kdp-publishing states plainly that it does not write your book; this pattern does not typeset it. Typesetting and authorship have different failure modes, and tools that bundle them tend to be good at neither.

The handoff artefact

book.py assemble writes two files:

• dist/manuscript.md — every drafted chapter concatenated in manifest order, with part dividers, front matter from the manifest, and undrafted chapters marked as comments rather than silently omitted
• dist/manuscript.json — title, author, snapshot date, chosen profile, handoff target, word count, and per-chapter status

The sidecar exists because the next tool should not have to parse prose to learn what it is being handed, and because "which chapters were actually in this build" is the first question asked when a proof looks wrong.

Typesetting profiles

typeset/profiles/*.toml describe how the text is set: measure, type families, sizes, leading. They deliberately do not describe trim, gutter, spine or bleed.

One value in a profile leaks upward into authoring: the measure. The measure fixes the maximum monospace line length that will not wrap in print, and that cap belongs in the style guide.

It is a measurement, not a convention. A book authored against Crown Quarto's 5.62in measure and later moved to KDP's 6x9 will wrap its listings, and the symptom appears in a proof rather than in a build. Re-derive the cap whenever the trim or the monospace size changes; never carry the number forward because it was already written down.

Channel differences that matter

• Lulu offers Crown Quarto (7.44 × 9.68in); KDP does not. KDP's 7.5 × 9.25in is not offered by Lulu. A book targeting both needs a profile per channel, and the two will have different page counts.
• Colour interior changes cost per page substantially. It is a spec decision taken during definition, not a production decision discovered at upload.
• Greyscale interiors require that no figure carries a distinction by hue alone — a graphics style guide constraint that has to be set before thirty figures are drawn.

Pre-editing

The review stage sits between drafting and publication and is skippable by explicit user choice.

Worth stating clearly: skipping it is a decision about what the book is for. A personal reference printed for one shelf and a book sold to strangers do not warrant the same scrutiny, and the pattern does not pretend otherwise. What it does insist on is that the choice was made rather than defaulted into.

The mechanical gates are not skippable. book.py lint --strict runs before production regardless, because an unresolved [VERIFY:] marker in a printed reference work is not a style question.