Two editions, not a translation: write from one spine

A second-language book edition should inherit the verified argument, not the first edition’s finished sentences. Chapter 7 of The Agentic Author calls that argument and its proof the shared spine. Both editions draw from it. Each writes its own prose for its own readers.

In this piece, translation means the sentence-led workflow the chapter warns against: hand over finished English paragraphs and ask for Dutch equivalents. Chapter 7 calls the failure structural, not a matter of the translator’s skill: a line fed finished sentences returns converted sentences. Give the second writer prose to convert and the first edition’s examples and rhythm travel along unnoticed. Give them claims, evidence, structure and terms, and they can make a distinct edition while the reasoning stays intact.

What stays shared, and what gets rebuilt?

The spine is more precise than a topic and less restrictive than a finished manuscript. Put it on paper before the second line starts.

Shared across editions	Written for each edition
The argument and framework	Sentences, paragraph rhythm and idiom
Verified claims, their sources and their limits	A style profile for the intended readers
The chapter map where order matters	Examples that carry the point for those readers
Defined concepts and paired glossary terms	The words used to explain those concepts

In the book’s own production, the English edition is the master: it is drafted first, and it is the home of record for the argument and the verified claims. Master names where the argument is kept, not a higher standard of quality. The second edition still has to stand on its own.

An evidence ledger makes the proof portable. Each row records one claim, the source that supports it, how strong that source is and what would retire the claim. Send the second writer the claim with its qualifier, not merely a citation. If the source supports “during the rollout,” the new edition must not turn that into “because of the rollout.”

What does the second edition have to make for itself?

Give the edition its own style profile. Name its reader, form of address, tone, sentence rhythm, preferred and banned terms, borrowed-word policy and rule for localizing examples. Chapter 7 insists that each profile is written independently, toward its own readers; neither is derived by translating the other. A translated profile would hand the second writer someone else’s voice instructions.

The chapter sketches what a Dutch profile for a founder’s book might set: the informal je-form rather than the formal u, a nuchter tone (level-headed and unsentimental), and an anglicism policy that keeps loanwords Dutch professionals already use, such as red team, while rendering others in Dutch. The book’s own Dutch edition keeps red team and writes stijlprofiel and woordenlijstbeheer for style profile and glossary control. The Dutch version of this article was recast the same way, not translated sentence by sentence.

The paired glossary does a different job: it keeps each defined concept stable while its name changes. Where the second language offers several plausible terms, choose the one that carries the intended meaning and record the pair. A writer can then change a metaphor or paragraph break without silently changing the concept.

Choose examples for the readers named in the brief. A language does not identify a single reader or market. Specify the role and context first; then change the scene while preserving the point it teaches.

A new real-world example also makes new factual claims. Chapter 7 puts it plainly: a localized example has to be as true and as checkable as the one it replaces. Open a source for the new example, or mark it as hypothetical.

What goes wrong when finished prose becomes the input?

Chapter 7 names the failure in one line: feed the agent English sentences, ask for Dutch ones, and two lines have collapsed back into one. The second edition has become a translation, whatever the workflow is called.

The output can still be fluent. Fluency does not show whether it reads as native writing, whether a qualifier survived or whether the original example fits the new reader. It looks finished before anyone has decided.

Ask one question before drafting: is the second writer reading the settled argument, or converting the first writer’s sentences? Chapter 7 calls the target bilingual parity: equivalent argument, claims and quality, with freedom in voice, examples and idiom. Parity does not require matching word counts, sentence boundaries or a twin for every paragraph.

A worked example: one claim, two reader routes

This example is hypothetical. An author is writing a book for team leads about checking claims before acting on them. The English edition sets its scene in a software support team, the Dutch edition in an equipment-repair business. Those settings are choices in this brief, not claims about everyone who reads either language.

The shared claim comes from chapter 6 of The Agentic Author: a citation points to a source, but the cited passage must actually support the sentence. Every traced claim in that chapter takes one of three dispositions: keep (verified as written), reduce (rewritten to fit the source) or cut (removed). Both editions teach that sequence: state the claim, read the source, then keep, reduce or cut it. The rule and standard of proof stay fixed; the scene can change.

In the English edition’s fictional scene, a software team writes: “The new ticketing tool caused the complaints.” Its imaginary incident report records complaints during a rollout but does not establish their cause. After reading the report, the writer reduces the sentence to what the source can bear: “Complaints were recorded during the rollout.” The source can carry a timing claim, not a causal one.

In the Dutch edition’s fictional repair workshop, a draft blames late repairs on a new parts supplier. An imaginary delivery log records late parts arrivals but does not establish why every repair finished late. The writer reduces the claim to the arrivals the log actually records; a claim about the cause of all delays needs more evidence.

Both documents are invented and supply no evidence about a real business. In a real edition each needs its own check; the software report could never verify the supplier story. The paired glossary has work to do as well. In The Agentic Author, chapter 6’s keep, reduce and cut appear in the Dutch edition as behouden, bijstellen and schrappen: different words, one check.

Which claims travel, and which need another check?

Chapter 7 lets a language-independent verified claim cross into the second edition with its proof and limits attached. The second writer need not reopen the source because the language changed, but must state the claim within its verified reach. If the source supported an association, the new edition cannot call it a cause.

Two cases demand fresh work. First, a claim tied to law or regulation has a jurisdiction. The second edition must check whether that claim applies to its market. Second, a localized example brings new facts, so its company, document or event needs its own source.

Time adds another limit. Chapter 10 treats the evidence library, glossary and style profiles as reusable assets, while warning that a claim verified for the first book may be stale by the third. Before reusing an old claim in either edition, check whether its source or the world it describes has changed. A shared spine is a controlled record, not permanent permission to repeat a fact.

How do you check parity without forcing identical sentences?

Use two readings that answer different questions. First, have someone competent in the second language read the draft with the first edition closed. Do the voice, terms and examples read as writing for that reader? Ask them to mark sentences that feel imported or require knowledge the intended reader was never given. Chapter 7 assigns certification of native voice to a native ear, not to the agent that drafted the text.

Then compare the edition against the spine. Check the conclusion, every load-bearing claim with its qualifier, the framework’s steps, the paired terms and the order of argument where it matters. An edition can pass a language read and still overstate a claim. In the chapter’s own example, a Dutch draft let one claim stand a shade more confidently than the hedged English version allowed, while each sentence read fine on its own; the gap showed only when the two arguments were laid side by side. An edition can also pass that comparison and still sound imported. Both readings are needed.

The check is easier when every difference has a reason. A different example serves a named reader. A different paragraph order helps that reader follow the same argument. A different term is recorded in the glossary. “It reads well” is not enough to explain a changed claim.

Where can an agent help?

Give the agent the spine, the edition’s style profile and the paired glossary. It can draft a fluent first pass so the writer is not facing a blank page, flag a term drifting from its agreed equivalent, and surface an example that will not land in the target market. None of that gives it authority to certify its own prose or its own account of a source.

Chapter 7 gives the reason. The system optimizes for the likely next word, not for the sentence a writer in that language would have chosen, and it can shift a claim by a degree while rephrasing, with no flag. The author or a native-language collaborator must make the language judgment, and a person must open the sources for new factual claims. If nobody on the project can judge the second language, the edition has no native review gate yet. Fluent output is a draft, not evidence that the gate has passed.

What are the three failure signs?

Chapter 7 names three ways a bilingual production line breaks.

Translating instead of producing. The second writer receives finished paragraphs and is asked to convert them sentence by sentence. The result is translationese: correct, but foreign. The repair is to supply the spine, an independent style profile and a brief written for the second edition’s readers.

Letting the editions drift apart. Both editions sound natural on their own, but a qualifier disappears or a defined term shifts its sense across the border. Compare claims and paired terms against the shared spine, rather than judging each edition in isolation.

Shipping one edition and a shadow. One edition gets source checks, a style profile, localized examples and a native reviewer; the other gets only a fluent draft. Give both editions a gate that can fail them. If the second cannot clear its own gate, it is still in production.

The BooksOS overview sets out seven production stages and one rule: agents draft, gates decide. Chapter 7 applies that rule to the second language. What does the author hand to the second line, and who may say that line is done?

A brief you can copy

Shared claim and its exact limits:
Evidence and source passage that support it:
Order of reasoning that must stay:
Terms that must mean the same thing in both editions:

Second edition's reader and purpose:
Voice and reader-address rules:
Example to replace or build, with its own source to check:
Local claim or rule that needs a new jurisdiction check:
Native reader who will review the draft:
Person who will compare the draft with the spine:

The first four lines preserve the book’s argument. The next six force the second edition to make its own choices and name the checks those choices create. Leave a line blank if it does not apply; do not fill an evidence line with a plausible source you have not opened.

Try this today (20 minutes)

Choose one page you intend to publish in a second language. Spend five minutes writing its spine in a few lines: the claim, the evidence that supports it and the order in which the page makes its case. Include the claim’s limit, especially any word such as “may,” “some” or “in this sample” that the second edition must keep.

Spend the next ten minutes listing what the second-language reader needs that the first page does not give them. Name an example that reader would not recognize, a term that needs an agreed equivalent, an assumed context that needs explaining, and any rule that may change by jurisdiction. For each proposed real example, write the source you would have to open. If you do not have one, mark it unverified.

Use the last five minutes to write a two-sentence brief for the second edition’s writer: what must remain true, and what they are free to rebuild. Give that brief and the spine to someone who knows the second language. If the only useful instruction you can write is “translate this page,” the shared argument is not explicit enough yet. That is the finding to fix before drafting.

Cite this:Two editions, not a translation: write from one spine.Len P. van der Hof. https://lenvanderhof.com/en/blog/two-editions-not-a-translation/ · Published 30 October 2026.