About
We model football from data. Here is why, and who does it.
Docteur Bet applies statistical methods to football, and publishes what they produce — including when the result is disappointing.
Our mission
Football produces an enormous amount of data, and very little analysis anyone can actually verify. Most of the tips in circulation come from a conviction stated with confidence, not from a method that can be traced. The reader has no way of knowing whether they are looking at a model, an average, or an opinion.
We start from a simple principle: apply to football the rigour expected of a model in quantitative finance or actuarial work. Known assumptions, a published method wherever possible, limits owned up to, and performance measured over time rather than on a single call.
That means showing what does not work. A 60% title probability is wrong four times in ten. A model that knows only results is blind to injuries and transfers. A yield measured over a hundred bets proves nothing. These limits are written on the site, not relegated to a footnote.
A figure that does not say what it is worth is worth nothing. It is the only editorial rule we impose on ourselves, and it decides everything else.
It takes two forms: public content, free and documented, showing what our models say matchday after matchday; and an app, for those who want a method rather than a hunch.
Leadership
Docteur Bet is led by its founder, Lucas da Silva, a doctoral researcher in mathematics, who oversees the construction and calibration of our models — from the public Dixon-Coles model through to the proprietary model behind the app.
Football serves here as a testing ground for established statistical methods: count models, constrained estimation, temporal validation. Not as a pretext for tips.
The company
Docteur Bet is a French company. We operate an app published on the App Store and on Google Play, a computation chain that refits six European leagues on every matchday played, and an API that feeds this site as well as our partners.
Our computations are dated, our settings displayed, and our result history available in full. We would rather our seriousness be verified than asserted.
Why “Docteur Bet”
The name owns what we do: applying a research approach — hypotheses, method, verification — to a field where that rigour is rarely visible. It is not wordplay; it is a commitment about how we work.
What we stand for
Transparency, where it is possible. The model behind this site is academic and published: Maher (1982), Dixon-Coles (1997). We do not hide how it works, and no in-house method is used here without being described. What stays proprietary — the app's model — is proprietary for a reason we own rather than by reflex: it is what carries the product's value. Its performance, though, remains verifiable.
Limits, always stated. A probability is not a certainty. We repeat it deliberately often, on the site as in the app, because that is what separates a serious tool from a sales argument.
Substance before style. We prefer a figure that is right and slightly less spectacular to one that is impressive and fragile. That choice sometimes costs a good line, and it builds trust over time.
Two models, not to be confused
This site and the app do not share a model. The site publishes an open Dixon-Coles model you can reproduce. The app rests on a proprietary model that takes operators' odds as input and pursues a different aim. Their figures do not coincide, and that is normal.
Where to follow us
Analysis, simulations and behind-the-scenes work on the models are published here, on YouTube and on Instagram. The links are at the foot of every page, and the account depends on the language you are reading.
Join us
We look for people who like to check
What matters here is not the line on a CV but the ability to doubt a figure, trace where it comes from, and write clearly what you found. We work in a small group on demanding subjects, and everyone sees the effect of what they produce.
- Statistics and modelling — count models, time series, out-of-sample validation.
- Engineering — Python for the computation chain, TypeScript and Next.js for the site, Node for the API.
- Writing — telling what a model says without betraying it, in French or in one of the site’s five other languages.
None of these is an open role with a spec and a posted salary: we do not hire continuously. Apply anyway — we read everything, and we reply, including to say no.
No cover letter. Tell us instead what you have built and what was wrong with it: that teaches us more.
Going further
A figure, an error, a collaboration
Press, clubs, researchers, or simply a reader who spotted an inconsistency: we reply. Reporting an error helps us; it is not a criticism.