Appendix — How this was made
08 September 2026
A note on method, in line with my guide “How to Use AI in Your Project” (V1.7). A series that argues judgement is the check on the machine would be hollow if it hid its own machine, or the checking.
The plan
The goal. A white paper, and a personal essay series for the website, on what a university is for when the content is free — grown out of preparing for a higher-education panel.
Where the thinking came from. The thesis is mine, reached the way the afterword describes: a decade of working with students, hiring them, watching what an education produced, and thinking it through. Observation, not research, and not AI. The conclusions existed before any tool was opened. What I did not have was the history checked, the argument put in order, the prose in my own voice, and the field’s prior work mapped.
The zone. Zone 4 in the guide’s terms — intentional. Think first; AI as a tool; record everything; every conclusion mine.
Execution
| Tool | Task | How the output was used and verified |
|---|---|---|
| Claude (Anthropic) — drafting | Scaffolds and section drafts, for the paper and the posts. | Nothing was accepted as written. Every draft was rewritten in my own voice; most were rewritten heavily or discarded. I edit against a written voice specification I keep; where the draft and my rewrite diverged, I logged the difference and corrected the spec. The prose was the part I trusted the machine with least. |
| Claude (web search) — research | Checking facts and dates: the WEF skills ranking, the TEQSA resource, Australian higher-education funding history, the founding of Bologna, al-Qarawiyyin, Humboldt and Melbourne, Socrates in the Phaedrus, the NAPLAN research, Kahneman and Klein on intuition, and the prior literature behind the note on novelty. | Every fact that mattered was checked against a primary or independent source, and named in the paper’s Notes & sources. Contested claims were marked as contested. |
| Code (Node.js, docx, HTML) — production | Building the branded paper and the hub page from a script. | The build is the document; each version dated and reproducible. |
| Me — everything that decides | The thesis, the concerns, the conclusion, the structure, the title, what stayed and what went. | Not verified. Decided. My name is on it, not the tool’s. |
Record, and what was discarded
One continuous conversation; dated, versioned files. The voice specification is itself versioned — the record of teaching the machine my voice, and of where it kept failing.
More was discarded than kept, most of it prose. Two failures recurred, and were cut every time. The machine wrote me as someone who was right all along, when the honest version has me getting it wrong first. And it reached for manufactured wit — a tidy closing line with nothing behind it. Both are in the voice spec now. The one idea I think might be new — a paradox about what regulation protects and what it leaves exposed — I set aside myself, because I am not sure enough of it to build on.
The test
My own guide sets one test: if someone asked you to walk through how you reached any conclusion — tool by tool, decision by decision — could you? For every line here, I can. The drafting was Claude’s; the research was Claude’s and checked by me; the thinking, the conclusions, and the final shape of every sentence were mine.
This appendix was drafted by Claude from the record of the work, then rewritten and approved by me. The method in one sentence.
This post is part of The Third Nexus Series — What a university is for when the content is free.