We Measure What Others Narrate
Editorial — for press and research
For years, the argument about online gambling was conducted in stories: operators narrated their games, method sellers narrated their winnings, and almost nobody measured anything. That era is ending. On one side, regulators are finally publishing real data, at scale. On the other, prediction sellers have industrialised the manufacture of their own proof. Our line has not changed since day one: measure, deposit the data, and let anyone check the work for themselves.
- 38,000 people and 157.2 million transactions: the Gambling Commission's open-banking study shows licensed and illegal gambling fed by the same bank accounts[1].
- One million rounds simulated on the documented crash-game model, with the full corpus published: anyone can redo our calculations before citing them[2].
- Methodology in the open: every measurement is deposited with its model and its data — signed studies on Zenodo, a verifiable researcher identity (ORCID 0009-0000-7586-9788).
- Two methods over one corpus — distribution measurement, then Kelly growth-rate evaluation — produce conclusions neither method yields alone.
- The programme: the open-banking study, the Aviator pair, Crash Gambling Anatomy Part 2, the San Diego panel piece, and Series 2 — every piece deposited with its data.
Our studies are deposited, and the data behind them is attached, free to download.
Regulators now publish the real data
The UK Gambling Commission has published an open-banking study tracking 38,000 people and 157.2 million transactions[1]. These are not statements, and not players' memories: they are bank records. And the headline result fits in a single image: the same bank account feeds both licensed sites and illegal ones. The two worlds the public debate carefully separates turn out, in the reality of the accounts, to sit on the same statement[1].
This is a change of era. When a regulator measures real gambling instead of commenting on it, unsupported claims lose their place. Narrating is no longer enough: you have to show. We welcome the shift for what it is — confirmation that measurement, not narrative, is becoming the norm of the gambling debate.
Prediction sellers now manufacture their own proof
The opposite movement exists, and it too has industrialised. In late October 2026, the sellers of Aviator predictions — the products of Sellers A and B and their variants — no longer merely promise. They stage their results: screenshots presented as proof of winnings, published on Facebook and Instagram, and Telegram funnels where the next prediction is sold[3].
The mechanism deserves describing for what it is. A screenshot costs nothing to manufacture. It shows neither the total staked, nor the previous days' losses, nor how many predictions missed before the one that "worked". What is sold in a VIP group is not information: it is staged certainty, in a place where the game's documented model contains none. The player does not pay to know; he pays to stop doubting.
Our answer: measure, deposit, reproduce
Our work rests on one simple choice: assert nothing we have not measured, and deposit every measurement where anyone can redo it. We simulated one million rounds of the documented crash-game model, at the published 98% return-to-player setting, and published the full corpus: the simulation study (DOI 10.5281/zenodo.23597144)[2] and the category anatomy study (DOI 10.5281/zenodo.23761515)[4] are deposited, signed, and attached to a verifiable researcher identity.
Neither the regulator nor the seller offers that combination. The regulator publishes a study; it does not deposit a corpus anyone can reanalyse. The seller publishes screenshots; he deposits nothing at all. We deposit. That is the entire difference between a claim and a result.
Several methods over the same data
A dataset does not contain a single conclusion. It contains as many as the methods applied to it — provided the data is deposited, and therefore reusable. That is the commitment we make here, and it structures the programme below.
Method 1 — distribution measurementStandard deviation of one round's payout, by cash-out target (units staked)
Expectation does not move: -0.02 units per unit staked, at every target[2] — on expectation, the targets are tied. What grows with the target is the spread of one round's outcomes. The full study of this method: the real cost of low cash-out.
Method 2 — Kelly growth-rate evaluationRisk of a halved bankroll over 1,000 rounds — fixed stake of 2% of bankroll each round
The growth-optimal Kelly stake is 0 at every target[2]: no target grows a bankroll. At a fixed stake, each target becomes a measured speed of ruin, and the targets rank — where expectation left them tied.
Our first method was distribution measurement: what happens, round after round, when a player always cashes out at the same multiplier? The measured answer: the cash-out target chooses the session's variance, not its expectation. The second method, Kelly growth-rate evaluation, rereads exactly the same data and sees something else: read as a bettor sizing stakes to grow a bankroll, every cash-out target becomes a measurable speed of ruin. The same corpus, two methods, two conclusions neither could have produced alone. And the same Kelly grid can evaluate the predictions sold in VIP groups: treated as a bettor, their performance reads in bankroll figures, not in screenshots.
Our analysis
We distinguish three things in what follows: the findings of the sources cited on this page — the Gambling Commission study, the San Diego panel and the law of the documented model — credited to them and cited; our calculations on the deposited one-million-round corpus, flagged as ours, with their margins; and our interpretations, labelled as such.
On expectation, the targets are perfectly tied: -0.02 units per unit staked at every target, from 1.2× to 10× — the first method therefore produces no ranking of targets.[2] Read through the Kelly growth rate, the same corpus produces a strict ranking: at a fixed 2% stake, the probability of a bankroll halved within 1,000 rounds rises at every target — 20% at 1.2×, 35% at 1.5×, 44% at 2×, 55% at 3×, 64% at 5×, 79% at 10× — and the growth-optimal stake is 0 everywhere.[2] Neither reading contains the other: they bear on the same million rounds.
Our reading: this contrast is the practical demonstration of our line. A deposited corpus yields not one conclusion but as many as the legitimate methods applied to it — and the second reading exists only because the data was reusable. Without deposit, all that remains is the narrative of whoever holds the figures.
What this changes for players
These are not statisticians' curiosities. They are real-budget problems. A panel from the Rady School at UC San Diego tracked 717,724 online consumers between 2019 and 2023[5]: only 4% withdrew more than they spent[5]. In other words, over five years, 96% of online gamblers were net losers. Meanwhile, prediction-group fees leave the same budgets as the rent — and illegal sites, the British study shows, draw from the same account as the licensed ones[1].
That is why we measure. Behind every distribution sits a household deciding without the figures. Our job is to publish the figures.
The programme, piece by piece
The research programme, piece by piece:
- The open-banking article — the same-account finding, in full: how licensed and illegal gambling share the same bank statements.
- The Aviator pair — one piece for the press and one for our readers: the fear prediction sellers exploit, and how to avoid it; their claims are evaluated as a Kelly bettor, in bankroll figures.
- Crash Gambling Anatomy, Part 2 — the derivation of the same mechanics published this year by Scott, Sher and Daigle in UNLV's Gaming Research & Review Journal[6], read against our measurement.
- The San Diego panel piece — 717,724 online consumers tracked from 2019 to 2023, 4% of them net winners: the psychological and economic conclusions of five years of real behaviour, in the US frame.
- Series 2 — the one-million-round corpus reread through Kelly growth-rate evaluation: method 2 of the figure above, applied to already-public data.
Every study we publish is deposited with its corpus open to anyone who wants to verify our figures before citing them. That is how we intend to continue: measuring what others narrate.
If you are a journalist or a researcher and you want the data, an extract or a comment: contact the newsroom via the Press page. This page is our statement of method; every piece of the programme is deposited before it is pitched.
Frequently asked questions
Is your data free to reuse?
Yes. The studies are deposited on Zenodo under a CC BY 4.0 licence, with the corpora behind them; cite the source and the DOI, and redo the calculations if you wish — that is what they are public for.
Why apply several methods to the same data?
Because a corpus does not contain a single conclusion. Distribution measurement reads a session's variance; Kelly growth-rate evaluation reads a speed of ruin. Both readings are true, neither contains the other — and the data must be deposited for the second reading to be possible at all.
How do you evaluate the predictions sold in paid groups?
As a bettor, not as a screenshot: the Kelly grid turns their claims into bankroll figures — measurable growth or ruin. That is the evaluation the Aviator pair applies, in the programme above.
I am a journalist: how do I receive the programme pieces?
Each piece is deposited, with its data attached, before it goes to the press. Write to the newsroom via the Press page.
References
- UK Gambling Commission — open-banking study of gambling transactions, October 2026 (38,000 people, 157.2 million transactions). ↑
- Dagor, Andrew — Crash-game bust multipliers: a one-million-round simulation of the documented crash distribution (RTP 98%), Zenodo, DOI 10.5281/zenodo.23597144. ↑
- Chicken Road Productions — Research Radar social-media sweep, 26–29 October 2026, and predictor-claim tracker (instrument F104): the wave of Sellers A and B, the staged screenshots and the Telegram funnels. ↑
- Dagor, Andrew — Crash gambling: anatomy of an online gambling category, Zenodo, DOI 10.5281/zenodo.23761515. ↑
- UC San Diego, Rady School of Management — online consumer panel, 2019–2023 (717,724 consumers). ↑
- Scott, Sher & Daigle — Cash or Crash, UNLV Gaming Research & Review Journal, 2026. ↑