'Books are not science.' Correct. Here is what the book is doing on the evidence page.

Michael Darius Eastwood · Independent AI alignment researcher

Published 3 July 2026

The reader is right: a trade-published book is not a scientific artefact in the technical sense. It has not been peer-reviewed as a paper; it does not carry a dataset; its citations are journalistic in density rather than academic. If the programme were offering the January 2026 book as a proof of anything, the objection would land. It is not doing that. Understanding what the book is for on the evidence chain closes the objection cleanly.

What the book actually is

A trade-published book, print edition 2 January 2026 and ebook edition 6 January, ISBN 978-1806056200, 470 pages, 122,251 words. Its role in the evidence chain is documentary. On a specific date, publicly and irrevocably, a specific arrangement of arguments and a specific vocabulary, The Chokepoint, the HARI Treaty, the Three Ethical Loops, the U equals I times R squared presentation, entered the public record with a date fixed by the publisher's own indelible record. The book is a monument, not a proof. Its evidentiary value is the fact of its publication at that time, not the strength of any individual argument inside it.

Why timestamped monuments matter

Because several convergences the register catalogues match theses stated in the book: the Chokepoint chapter's account of infrastructure-as-leverage, and subsequent events including the US export-control action against a deployed commercial AI model, Macron's Evian critique and the EU Chips Act 2.0 package. If the book did not exist, or existed only in unpublished manuscript, the reader would have to trust the author about which arguments preceded which events. Because it exists as a trade publication on a fixed date, no such trust is required; the reader can walk into a bookshop and buy the same argument in the same words.

Where the proofs actually sit

Not in the book. The technical claims live in Papers I to XI and their supplementary materials, which are the things the field is entitled to review as science. The book states the framework in accessible prose so a general audience can follow it; the papers state it in the notation and with the assumptions that let a technical audience test it. The evidence page separates these clearly: the book is listed as artefact A3 with its ISBN, page count and word count; the papers are listed as A4 with their DOIs. Neither the book nor the papers do the other's job.

What the objection catches, and what it does not

It catches an author who cites a book to settle a technical dispute. That would be an error and it is not happening here. It does not catch the use of a timestamped, trade-published artefact to fix a claim's public appearance on a specific date. The Vatican encyclical, the Anthropic co-founder's brake-pedal argument, and the export-control action all followed publications inside the book. The register calls them convergent, not derived, and the book is what makes the "before" of that comparison verifiable rather than asserted. That is a legitimate role, and it is the only role the record asks the book to play.