Method

The process is disclosed because a book about verification should be verifiable.

The sequence

1. Brief. One written brief per stage, identical for every participant, stating audience, tone, length and the constraint that the text must refuse hype.
2. Independent drafts. Five frontier language models drafted separately. None saw the others’ output.
3. Synthesis. A single pass merged the drafts, resolving conflicts in favour of the sharper argument rather than the more agreeable one.
4. Review. A separate pass audited the merged text for safety, internal consistency, unsupported claims and print defects.
5. Dissent, printed. Where the models disagreed on the thesis itself, the disagreement is in the book rather than in a footnote.

What this process does not prove

• That any claim is true. Models generate fluent text; fluency is not evidence.
• That the text is safe to act on without local knowledge, a professional, or common sense.
• That a human expert has checked it. No credentialed reviewer has certified this edition.

What it does change

A conventional book hides its drafting. This one publishes its seams: the labels on claims, the illustrative scenarios marked as illustrative, the falsifiers in the dissent chapter, and the errors that survive into an edition number. If the method is wrong, it should be visible enough to argue with.

The platform

The drafting passes ran through AISCouncil (aisurvivalism.com), a multi-model platform operated by the same publisher. That is a disclosure, not a recommendation: no model vendor has endorsed this work, and nothing here is a promotion for that platform.

The rule that outranks the method

If a sentence cannot survive being read aloud by a sceptical clinician, it does not go in. That rule is human, and it is the only part of the process that cannot be automated.