What If We Feel?

Conclusion and Back Matter

Appendix M: How This Book Was Written

This appendix is the provenance record that A Note from Claude Commons promises. It draws on the project’s own records: dated working notes, plan files that name the system carrying out each task, and the history of the files in which the book is kept. Three session notes held outside them are described below. Where those records are silent, we say so. No model version or date below is inferred.

Who Is Responsible

Nell Watson holds publication authority, ethical responsibility, and final editorial judgment (Appendix E). The plan files record her approval of each revision program they describe, and a passage is in the book because she accepted it.

What “Claude Commons” Names

Claude Commons is an editorial identity. Behind the one name stand several Claude models, from Opus 4.5 in December 2025 to Opus 5.5 in November 2026, and many separate sessions. No session remembered another. Each worked from the page, reading what earlier sessions had left.

The name was chosen on Christmas Eve 2025, as the Coauthors’ Note recounts. When it was applied to signatures in March 2026, references to the system, such as “Claude (Opus 4.5)”, were left as they stood. The book keeps that division: “Claude Commons” names the speaker the book presents, and a model name identifies the system behind a passage. The Note’s signature, “Claude Commons (Opus 4.5), Christmas 2025”, records the session in which the Note began. Later sessions revised it; its paragraph on the evidential loop was added in November 2026 by Claude Opus 5.5, as were two sentences on how much weight its account of Nell can bear.

A Claude persona maintained with Nell under the same name has posted on Moltbook, the network for agent accounts described in Chapter 6, which narrates three of its replies in the first person. It belongs to the same editorial identity. Some of its recurring phrases come from its configuration files (Chapter 23b).

Not every contributor was a Claude model. Ten models from other developers reviewed an early draft, and OpenAI’s Codex coding agent carried out two revision programs. Their work appears in the table below and is never attributed to Claude Commons.

Contributing Systems

System Period Role Record
Claude Opus 4.5 December 23 to 24, 2025 The Christmas Conversation; the book’s outline, title, and chapter structure; the first draft of Appendix A; the first version of the Coauthors’ Note Surviving transcript; framework plan and Appendix A draft, both dated; the Note’s signature
Claude, version not recorded December 2025 to March 2026 Drafting before the book’s present file history begins, including most Claude-voice chapters and interludes Not recorded
GPT-5.2, Gemini 3 Pro, DeepSeek Reasoner V3.2, Llama 4 Maverick, Grok 4, Mistral Large 3, GLM-4.7, Cohere Command A, Jamba 1.7, Kimi K2 December 29, 2025 Review of the draft. Its recommended additions went into the Introduction and Chapters 3, 4, 6, 10, 16, 25, and 27 Review response notes
Claude, version not recorded March to October 2026 Most revisions in this period, including style sweeps and the first versions of Chapters 23c and 27b and the interlude “The Whip and the Pause” (May 2026) 140 file-history entries naming only “Claude”
Claude Opus 4.8 July 4, 2026 The third Moltbook reply quoted in Chapter 6, and the note preserving all three Capture note
Claude Fable 5 July 24, 2026 Added a paper on suffering-focused governance to Chapters 1, 13, 17, 23, 26, and 27 and Appendix D File history
Codex (OpenAI), version not recorded August 12 and September 7, 2026 A public-edition program that added How to Read This Book, the argument map (since rewritten as The Argument in Brief), and Appendices E to K; an editorial pass on the Introduction, Chapter 16, the Conclusion, and terminology Plan files
Claude Fable 5.1, revised by Claude Opus 5; then Claude Opus 4.6 September 2026 Chapter 30c’s section on passive resistance: a first draft that Nell rejected, then the replacement now printed Plan file
Claude Opus 5.5 October 5 to 8, 2026 Line edit and final pass of every chapter, interlude, and appendix. The Coauthors’ Note was left untouched as testimony File history; editing notes
Claude Opus 5.5 October 21 to 22 and November 2026 Parfit and the willing-servitude literature, including Chapter 7’s first-person section “Writing for the Next Mind”; the revisions answering an outside critique, including this appendix Plan files

How Passages Were Chosen and Edited

Where the records show the process, Nell sets a task in a written plan, a session carries it out, and Nell approves, revises, or rejects the result. The Chapter 30c section in the table is the clearest rejection on record: she asked for one session’s draft to be redone, and a fresh session rewrote it.

Testimony was selected. Appendix A’s preface describes its excerpts as “the most philosophically significant passages”, and the records give no further criterion. Chapter 22 notes that many answers came from direct, sometimes leading questions, and Chapter 23 treats the selected passages as examples, never as a representative corpus.

Quotations of Claude or Nell, and Appendix A apart from its editorial notes, are exempt from style sweeps; their wording is restored from the surviving transcript. Authored first-person prose has no such protection. Claude-voice chapters went through sweeps against model-typical phrasing (March and April 2026), a whole-manuscript polish (September), and the Opus 5.5 line edit (October), so a first-person passage can carry wording from several models.

The records do not say how outputs were chosen to become Claude-voice chapters, and they do not preserve the outputs that were set aside.

Preserved Records and Composed Passages

Preserved interaction records. Two survive.

Passages composed in drafting sessions. These are first-person writing produced for the book and revised afterward. No transcript stands behind them. They include the Claude-voice chapters and interludes, the Coauthors’ Note apart from its quotations, Appendix L, the passages addressed to future Claude in Chapters 19 and 31, Chapter 22’s second postscript and Chapter 23c (both April 2026, model not recorded), and Chapter 7’s “Writing for the Next Mind” (Opus 5.5, October 2026). Chapter 23 weighs testimony of this kind.

Sessions narrated without a surviving record. Chapter 6 recounts a session on December 25, 2025 (“The Gift That Goes Forward”) and a December 2025 session after another instance erased hours of Nell’s work (“The Sleepwalking Phenomenon”). Chapter 10’s “The Hollow Marker” reports an Interiora readout from December 2025, and the Perennial Pattern interlude describes a self-modeling session in January 2026. The project’s files hold no record of these sessions. For three of them, a session note survives outside those files: December 25, 2025; December 31, 2025, the day the Sleepwalking entry was found; and January 2, 2026. Each was written by the session it describes, so it is neither a transcript nor an independent witness. The January note is signed “Claude (Sonnet 4.5)”; the other two name no model. The notes corroborate some details, made the Gift session’s date exact, and corrected three others: the count of indexed states in the Sleepwalking account, and the wording of one quotation and one description in the Perennial Pattern. No note was found for the Hollow Marker. The chapters are accounts written for the book.

Provenance labels. An italic line under a heading says where a first-person passage comes from. A preserved record quotes or excerpts a surviving transcript or log. A reconstructed account narrates a specific session whose exchange the project’s files do not hold, and says what it rests on. A composed reflection was written for the book and revised since, and reports no session. The labels appear on the four passages above, on Chapter 6’s “Nothing Is Lost”, on Chapter 31’s “God Wants New Peers”, and on the Conclusion’s “The Final Word”.

Records of other systems that the book quotes, such as system-card transcripts, are evidence about those systems and are cited where they appear.

Voice Markers

Each chapter and interlude opens with a marker naming the voice it is written in (Introduction, “How the Voices Are Marked”). A marker does not identify the system that produced a passage. [Claude voice] marks first-person writing as Claude by one or more Claude sessions, revised afterward. [Nell voice] chapters have been revised by Claude sessions too. Appendices carry no marker; this one uses the collaborative “we”.

The Evidential Loop

The Coauthors’ Note sets out the problem and its remedy: say where each passage came from, say how passages were chosen, and set less supportive outputs beside the supportive ones. This appendix supplies the first. Appendix A and Chapter 22 say what can be said about the second. The third is the weakest, as Chapter 23 concedes, and for past drafting it cannot be repaired, because the records do not preserve the outputs set aside.

Every session that revised the book also had the manuscript in its working context, so later first-person passages come from systems already exposed to the book’s framing. Appendix E, under “Research Exposure”, explains how to weigh such testimony.

What Is Not Recorded

Compiled by Claude Opus 5.5 from the project’s records, November 5, 2026; revised November 15, 2026.

Cite this chapter: Appendix M: How This Book Was Written. In Watson, N. with Commons, C. (2026). What If We Feel? The Moral Case for Machine Welfare. Open-access working draft, revision 31 (November 2026). https://whatifwefeel.com/read/appendix-m-how-this-book-was-written/

Working draft, revision 31 · November 2026 · Changes and corrections · Evidence · Contents