How These Papers Were Made

How These Papers Were Made

Each paper in this collection begins with a question rather than a topic, because a thesis question decides which sources matter and which do not. The method then moves through five steps, and each step produces a document that the next step depends on.

Step	What happens	What it produces
1. Collect	A manual search of Google Scholar and arXiv finds the sources, and a source enters the list only with a DOI or a downloadable PDF	A verified reading list
2. Ask	A thesis question frames that list	One question that the paper must answer
3. Annotate	A Vancouver annotated bibliography records what each work informs and what it refutes	A map of the evidence on both sides
4. Draft	An AI system writes one paragraph on each work from its annotation	A paragraph per source
5. Form	The bibliography and the paragraphs combine into a white paper, an IMRaD paper, or both	The manuscript

Why the method runs in this order

The order matters because AI drafting is strong at organization and weak at provenance. A model asked for a paper from a blank page will produce fluent text with sources that may not exist; however, a model handed a verified reading list and a bibliography that already records what each work says has far less room to invent, which means the author's judgment sits in steps 1 to 3 and the machine's labor sits in step 4. The annotated bibliography also forces a harder discipline than a literature review, because it must record what refutes the thesis as well as what informs it.

The papers copy no text, and they cite every table and figure.

What the audit found

Verifying a source before use does not guarantee that its identifier survives drafting. The 2026 paper tested this directly by re-checking the fourteen sources of the 2025 review: nine carried an incorrect DOI or preprint identifier, although their titles and authors were accurate, and one source could not be located under any title, author, or identifier. The 2025 paper remains in this collection with its reference list as first written, and a note, because the error and its correction are both part of the record.

The consequence is a final check. A last pass resolves every identifier in a finished manuscript against the source read in step 1, because drafting can damage what collection verified.

Status

These papers are manuscripts. None has gone to a journal or through peer review, because they exist to inform the design of the repositories that accompany them. The findings therefore stand as working research, not published results.

See the Reading Guide for where to begin.