Appendix A — Appendix A: AI Assistance Disclosure

A.1 What was AI-assisted

Draft zero of this book was produced with substantial assistance from Claude, an assistant made by Anthropic. The assistant generated first-pass prose for the original twenty-two chapters and the index, following a detailed editorial prompt written by the author. The author then reviewed, revised, and (in many places) rewrote the output. The book’s framework, chapter scope, running examples, and exercise design originate with the author. The prose does not.

The 2027 restructure was also AI-assisted. Working from the author’s course design for INFO 4871/5871 (Spring 2027) and the author’s article on public interest data infrastructuring, Claude reordered the chapters into module parts, wrote the part pages, revised chapter framing, cross-references, cases, and exercises to fit the new structure, moved tutorials between chapters, and drafted the new chapter on the limits of the framework. That revision is under author review. Commits from the restructure are marked in the repository history, and new bibliography entries from it are flagged for verification.

This disclosure exists because a book about public interest data science should not hide how it was made. The same principle that asks a model card to document training data asks an author to document drafting practice.

A.2 What was not AI-assisted

The bibliography’s selection reflects the author’s reading, not the assistant’s. The running examples (Boulder and Denver municipal records, a synthetic worker-log fixture, a class-wide FOIA tracker, the regulations.gov walkthrough) are authorial choices grounded in the companion course. The “Missing Manual” and “In the Public Interest” callout convention is the author’s. The genre structure, and particularly its commitment to producing submittable artifacts, is the author’s pedagogical argument.

The assistant was asked not to invent citations. Where a suggested reference could not be verified, the bibliography flags the entry for follow-up rather than recording a fabricated one. Readers who find a citation that cannot be located should open an issue.

A.3 How to verify

The prompt used to generate draft zero is preserved in the repository’s issue history under the label prompt-archive.
Each chapter’s first commit is the AI-generated draft. Later commits include human revisions and the AI-assisted 2027 restructure; commit messages say which. Use git log --follow ch-NN-slug.qmd to inspect a chapter’s editorial history.
Chapter-by-chapter revision notes are maintained in claude.md and in pull-request discussions.

A.4 Why disclose

Because readers deserve to know. Because other authors deciding whether to use similar tools need honest reports of what worked and what did not. Because a book that argues for documentation of automated systems must document its own use of one. And because the professional lineages this book draws on (journalism, law, accounting, planning, engineering) all insist on provenance: whose work, whose hand, whose judgment.

A.5 Versioning

This disclosure was last updated on the date of the most recent commit to this file. See git log appendix-ai-disclosure.qmd for the authoritative version history.

--- title: "Appendix A: AI Assistance Disclosure" --- ## What was AI-assisted Draft zero of this book was produced with substantial assistance from Claude, an assistant made by Anthropic. The assistant generated first-pass prose for the original twenty-two chapters and the index, following a detailed editorial prompt written by the author. The author then reviewed, revised, and (in many places) rewrote the output. The book's framework, chapter scope, running examples, and exercise design originate with the author. The prose does not. The 2027 restructure was also AI-assisted. Working from the author's course design for INFO 4871/5871 (Spring 2027) and the author's article on public interest data infrastructuring, Claude reordered the chapters into module parts, wrote the part pages, revised chapter framing, cross-references, cases, and exercises to fit the new structure, moved tutorials between chapters, and drafted the new chapter on the limits of the framework. That revision is under author review. Commits from the restructure are marked in the repository history, and new bibliography entries from it are flagged for verification. This disclosure exists because a book about public interest data science should not hide how it was made. The same principle that asks a model card to document training data asks an author to document drafting practice. ## What was not AI-assisted The bibliography's selection reflects the author's reading, not the assistant's. The running examples (Boulder and Denver municipal records, a synthetic worker-log fixture, a class-wide FOIA tracker, the regulations.gov walkthrough) are authorial choices grounded in the companion course. The "Missing Manual" and "In the Public Interest" callout convention is the author's. The genre structure, and particularly its commitment to producing submittable artifacts, is the author's pedagogical argument. The assistant was asked not to invent citations. Where a suggested reference could not be verified, the bibliography flags the entry for follow-up rather than recording a fabricated one. Readers who find a citation that cannot be located should open an issue. ## How to verify - The prompt used to generate draft zero is preserved in the repository's issue history under the label `prompt-archive`. - Each chapter's first commit is the AI-generated draft. Later commits include human revisions and the AI-assisted 2027 restructure; commit messages say which. Use `git log --follow ch-NN-slug.qmd` to inspect a chapter's editorial history. - Chapter-by-chapter revision notes are maintained in `claude.md` and in pull-request discussions. ## Why disclose Because readers deserve to know. Because other authors deciding whether to use similar tools need honest reports of what worked and what did not. Because a book that argues for documentation of automated systems must document its own use of one. And because the professional lineages this book draws on (journalism, law, accounting, planning, engineering) all insist on provenance: whose work, whose hand, whose judgment. ## Versioning This disclosure was last updated on the date of the most recent commit to this file. See `git log appendix-ai-disclosure.qmd` for the authoritative version history.