The book came first

Before texlode, there was a book: How To Publish Your Book with LaTeX: How Serious Authors Can Leave Word Behind. The writing finished long before it shipped; what followed was months of polishing. texlode’s first commit landed on 27 January 2026, partway through that stretch, and the book went on sale the following month.

So the book is not marketing material for texlode. It could not be: the text was written before texlode existed. It is instead a dated, specific description of a problem, written by someone with no solution to sell yet.

What the book was written against

The audience section of the preface is the clearest statement of the problem I have ever managed:

The book’s answer to that was a template, the LODE template, which the preface describes honestly: it “handles 90% of the complexity; you will just set up a few configuration parameters in separate settings files.”

That claim was true, and it is still true. A good template does absorb most of the pain of a book-length LaTeX project. But it taught me exactly where the remaining 10% lives, because I had to write chapters about it.

The part a template cannot fix

A template is a set of files. It cannot change what happens after you press compile. So the book had to teach people to live with these:

I wrote a whole reference chapter on interpreting LaTeX error messages. That chapter is good. It should not need to exist.

The 2026 edition is a specification

The preface of the 2026 edition lists what was new in that installment. I wrote it as a changelog. It reads now like a feature backlog:

• ›Tolerance and line-breaking optimization
• ›Page-breaking penalties and widow/orphan control
• ›Dynamic margin adjustment based on page count
• ›Hyphenation fine-tuning and custom patterns
• ›Float and image positioning controls
• ›Automated cover generation for all platforms
• ›Significantly expanded indexing methodology
• ›Enhanced debugging tools with detailed error message explanations

Every one of those is me pushing a template as far as a template goes: teaching a reader to set a value, compile, look, and set it again. They are all knobs that want a live preview and do not have one. That list is what you write when the format itself has become the constraint.

What texlode changed

texlode keeps a LuaTeX engine running and re-typesets a single paragraph in a few milliseconds, instead of rebuilding the document from scratch. Almost everything below follows from that one decision.

I presented the engine behind it at TUG 2026, the TeX Users Group conference, with the measured recompile times per paragraph type:

Two of these go past what the book describes as possible. Per-paragraph typesetting control is not a LaTeX workflow anyone teaches, because without a live preview it would be unusable. And tagged, accessible output (PDF/UA, which the book treats as a requirement still arriving under the EU Accessibility Act) is built in rather than a project to take on.

What the book still does better

The book gets you published. texlode gets you as far as the door. After a download it now reads your document and tells you what to type into KDP (format, trim size, bleed, language, ISBN, page count) and warns you if the trim you chose is not offered in the binding you picked. What it cannot do is the upload: Amazon publishes no API for that, so the last step is yours on every tool, ours included. For Lulu, texlode goes one step further: Lulu checks your print files against its print requirements before you upload them, and the upload is still yours.

The book still owns everything around the form. Account setup, ink and paper, the royalty decision, expanded distribution, what to do when the proof copy arrives wrong: a chapter of judgment that a generated checklist does not replace.

The book also explains where the editor assumes. Bleed, gutter, leading, kerning, recto and verso: the book defines 62 such terms and says why each one matters. texlode shows you a “bleed on all sides” checkbox and trusts you to know.

Some chapters are now older than the tool. The compilation-strategies chapter, the EPUB toolchain walkthrough and the local-editor setup guides describe a workflow texlode replaces. They were accurate when written, which is the cost of writing the problem down before solving it.

One difference is a straight tradeoff rather than a gap. The book teaches minted for syntax-highlighted code. texlode runs its engine with shell escape disabled, which is what stops a document from executing arbitrary commands on our servers, and minted needs exactly that capability. Highlighted code listings are a genuine loss, and a deliberate one.

Then the book became the test corpus

The polishing outlasted the writing by months, and partway through it the polishing moved into texlode. This repository’s git log still carries the editor’s own autosave entries from 17 February 2026: Edit chapter 'Introducing latex': -1 lines, then +2 lines, then -6 lines, a run of small corrections against Chapter 1. The editor’s first real job was finishing the book that argued for building it.

It never stopped being that. Chapter 2 is the fixture our keystroke-latency benchmark runs on, picked because it is mid-size real prose rather than synthetic filler.

Chapter 1 earned its place differently. A TikZ picture followed by a table triggers a font-recursion crash in the LuaTeX engine (\pgf@selectfontorig calling itself until the input stack blows), so the benchmark skips that chapter. Crash replays from it go back to April 2026 and the bug is still open. The book that described the problems also turned out to be the thing that found them.

That is the closest thing to a fair test I could have asked for.

Why this is the same project

The book’s epigraph is from Peter S. Beagle: “What is magic for? What use is wizardry if it cannot save a unicorn?” I used it to make a point about tools: LaTeX cannot write your book for you, it can only stop the machinery from getting in your way.

The template got that 90% of the way. The remaining stretch was never going to yield to better documentation, because the thing in the way was the compile-and-wait loop itself, and no amount of explaining makes waiting shorter. So I stopped writing chapters about the workaround and built the tool instead.

The book is still the better teacher. It is where the reasoning is: why bleed matters, how to approach an index, what the typesetting engine is doing when it breaks a paragraph. texlode is narrower on purpose: it removes the friction the book had to teach people to tolerate.