Reading one
Reading one publishes Friday 2 October 2026.
The paper, the question, the number and the figure appear here on that date, with the appendix, the inputs and the record's DOI.
Every Reading has the same four blocks
Four blocks, every issue, in this order, so that a reader who has seen one piece can read the next in fifteen seconds.
| Block | What it carries |
|---|---|
| What the paper says | Their words, their numbers, their p value or confidence interval. |
| What your brain actually wanted to know | The clinical question, stated as a clinician would ask it. |
| The translation | A probability, against a stated decision line. |
| What would change it | Honestly, in two sentences. |
The format is not redesigned. One paper, one question, one number, one picture — a second interesting question is a note towards a future Reading rather than an extension of the present one.
The Readings
Reproductions and comments
No check or comment has been received yet: one arrives by e-mail to eurostrokes@gmail.com and is published here with the sender's name, affiliation and date, once they have agreed in writing.
Open for independent check
Everything a Reading rests on is published with it: the counts as the source paper printed them, with the page and table they came from; the script, seeded; the execution log and the environment it ran in; the reconciliation between the engines; and the figure. All of it sits under CC BY 4.0 at a Zenodo record with a DOI, and downloads from this page as a PDF.
The invitation is open to anyone with the credentials to take it up — statisticians, methodologists, epidemiologists, clinicians who work with these methods. What a checker is asked for is one afternoon. Take the counts from the paper itself — the record says which page and table — reproduce the headline probability by whatever method you trust, and write back saying whether you agree with the number, with your name and your affiliation. Checks follow publication: a Reading goes up complete on its date, and the first verified reproduction is named on it when it arrives.
What is offered in return is the name, the affiliation and the date on the face of the Reading and in the Zenodo record's notes, as arithmetic independently reproduced by; co-authorship on any peer-reviewed submission the check contributes to; and first refusal on the biostatistics co-lead role for the Centre's pre-registered re-analyses. No money moves in either direction. A check that finds an error is published on the piece with the error, the correction and the checker's name, unless the checker asks for the name to be withdrawn — which is the part that makes the offer worth taking.
The same invitation stands for anyone who wants to work on the statistics rather than check them. The Centre needs a biostatistician on its re-analyses.
One route in for now: write to eurostrokes@gmail.com. A signed comment route is being built for this page, so that a check can be posted here under its author's own name rather than sent privately; it opens when it is ready, and a comment will be read before it appears.
How a Reading is made
A Reading takes one published paper and one clinical question. The counts come out of the paper as printed; the decision line is written down before any probability is computed, and the piece says so in its own text; the prior is a single reference prior, declared on the page; and the seed is fixed and published, so the run reproduces exactly.
The reasoning behind the method, and the published objection to it, are set out on Bayesian synthesis in stroke medicine, which is the method annex of this page. What reproducibility means for work of this kind, and why stable comes before better, is on Reproducibility.
Authorship, and the role of AI models
This applies to every Reading on this page. Each Reading's own record carries the specifics of that piece — what was run, in which languages and engines, with which seed — in its statistical appendix.
I choose the paper, the question and the line before I compute anything, and I fix it in writing first. I am teaching myself Bayesian statistics as I go, a work in progress, and I run part of the arithmetic myself. The rest — the code that produced these numbers — was written and executed with Claude, an Anthropic AI model, working from my specification, in two languages and four engines that must agree before a number leaves the folder. Every number here is in the execution log published beside it. The reading, the decision line and the judgement are mine; no model chose the question. None of this has been peer reviewed, which is why it is open for anyone with the credentials to check.
What this is not
This is not a peer-reviewed publication. There is no ISSN, no editorial board, no editorial process and no submissions, and a Reading is not a re-analysis: it adds no new data, it rebuilds none of the original model, and it makes no claim that the authors of the source paper erred. The original paper answered the question it asked, and every Reading says so, in the authors' own terms, before it does anything else. A reader who finishes a Reading thinking less of the source paper has been badly served.
The comments on a Reading are not peer review either. They are read before they appear, and what they are for is the arithmetic.
Contact
For anything about the Centre that is not a check — a paper you think deserves a Reading, a question about the method, or the statistics role — write: eurostrokes@gmail.com