Research studio
How we write research
Every working paper follows one house standard. It exists so that a reader can check each claim, rerun what we measured and see what we changed.
The standard
- A paper reads like a paperA4, two columns, Newsreader for the text and Inter for headings and captions. Page one is a title block with the abstract, keywords, metadata, a report ID and a cite line. Chapters are numbered, and the back matter carries references, disclosure, revisions and how to cite.
- Each chapter answers one questionA chapter opens with a one sentence question, builds one mental model with at least one diagram, and ends on three key takeaways. Checklists move to the appendix.
- Every claim says how it is knownA small tag follows each substantive claim: M measured by us, S sourced from a primary source, D derived with the arithmetic shown, O opinion, E estimate. Estimates are labelled as estimates.
- Primary sources, datedAnything load-bearing cites a primary source with the date we read it. When sources disagree we show them side by side. Vendor language is labelled as vendor language.
- Method and limits are printedEach paper states what was run, on which versions, how many times, and what was not tested. Pass criteria are written before an experiment runs, and a result that surprised us is reported as found.
- Diagrams are drawn as codeFigures are built from source in one house palette: the data path in solid ink, the control plane dashed in steel blue, failure paths dashed in orange. One idea per diagram, every edge labelled.
- Plain and vendor neutralNo marketing and no filler. Tools are named when they are test subjects, with a version. The body never recommends our own products; the disclosure says what the studio builds.
- Versions and corrections in publicA report ID such as GLINR-WP-2026-101 never changes; the version does. A major version changes conclusions, a minor adds material, a patch corrects. Corrections are recorded on the paper with the old and new wording.
- Free to read, cite and reuseEach paper ships as a web page and a typeset PDF, with BibTeX, RIS and CITATION.cff, under CC BY 4.0. Cross-posts point their canonical link back to the paper.
- Read the papersLong-form papers on the failure modes we meet running our own stack: restores, DNS changes and schema migrations. Each one states how every claim is known, cites primary sources and says what it did not test.