BIG IF TRUEHomeManifestoAboutIssuesDashboardsCheck it yourselfGitHub

BIG IF TRUE #000

Probably right, and interesting if so

BIG IF TRUE #000: what this series is, how it is made, and why you should not simply trust it

BIT VERDICT #000

The Claim: Data-science pieces that start from my own life — what I live, read, play and scroll through — built from open data or from data named directly on the web.

The Catch: Every analysis is produced by me guiding a large language model (Claude, to begin with). The model writes and runs the code, and I do not re-read it line by line. Nothing here is peer-reviewed.

The Reality: Each piece is probably correct, and would be interesting if it were — hence the name. Uncertainty is part of the narrative, and every issue ends with its cards on the table: a transparency-and-reproducibility panel that lets the reader judge for themselves.

1. What it is

I am Riccardo Gallotti. I lead the Complex Human Behaviour (CHuB) unit at Fondazione Bruno Kessler in Trento; I am a theoretical physicist by training, a complexity scientist and data scientist by trade, and for most of my career the questions I have asked of data — about how people move, connect, misinform and decide — have gone through the slow machinery of peer review before anyone read them. This is the other thing: the questions that come up on a train or in front of a screen, which a decade ago would have taken a month of code and now take an afternoon.

The distinctive thing is not the topics but how the work is done. I work with the model the way I would with one of my interns: I set the question and the framing, it downloads the data, writes the pipeline, draws the figures and drafts the text, and then we iterate — I question its assumptions and its results, send it back when a number looks wrong, suggest how a figure could be clearer, iterate on the text until it speaks in my own voice, and decide what goes in and what stays out. The words you are reading went through the same process. I am starting with Claude, and may use others as they come; whichever it is, the model is named with its version in every issue and the division of labour is stated the same way every time, because a vague "AI-assisted" invites the reader to guess.

The format reproduces, in miniature, the critical path of a scientific publication, in three figures. BIG is the observation of the fact: one figure, one claim, stated as plainly as the data allows. IF is the verification and deepening: a closer look at what the first figure measures, and what it misses. It is the part that asks the most of the reader. Sometimes that look corrects it; sometimes it adds what the figure could not show. TRUE is the extension: what IF leaves standing, taken beyond the case it started from. That is the IF in the title doing its job.

2. Why you should not simply trust it, and what replaces trust

Because the analysis is done entirely through the model, the content is not a scientific article I would submit — though some colleague might likely dare to. Uncertainty is therefore not a caveat added at the end but part of the story itself. What stands in for peer review is disclosure that can be checked: every issue ends with its cards on the table: a panel that says which model did what, lists in plain language the real errors found and fixed along the way, and reports the attempt to find peer-reviewed work on the same question and whether it agrees. Code and data are open: every issue publishes the code that collects the data, and, wherever the source's licence allows, the data itself and the full analysis, so that anyone can rerun it and check the numbers. The pieces measure; they do not recommend. And they are not scientific publications: if you need to reference one anyway, each issue gives a DOI for its archived code and data on Zenodo, and that is what to cite.

3. Who it is for

This is for people who are curious rather than credentialed, and who suspect that the small things and the serious ones deserve the same treatment: the same care in measuring, the same honesty about what the measurement can and cannot say. The questions often start from wherever I happen to be looking — the news, the place where I live, the web where we all live, the videogames and movies that are my own interests — and some will be light while others are anything but. None of them is too small to measure properly, or too heated to measure calmly, and the data usually reaches further than the question did. It is written for readers who want substance and are willing to follow an argument from one figure to the next — and for journalists who may want to pick a finding up, or fellow scientists who might want to follow one up, helped by the cards on the table. If that is you, feel free to contact me, or to cite the work. Each issue comes out in English and in Italian; any other language you may be reading it in is Substack's automatic translation, which I have not checked.

If you find a mistake, tell me. That is not a courtesy; it is the point.

Riccardo Gallotti, Trento, Italy


I would rather someone found a mistake than that nobody looked. Code and data are open, and published with every issue.

Read it on SubstackSubscribe on SubstackCheck it yourself