Honest descriptive statistics, on an audited analysis engine
The only product in its category that publishes its own negative result — and the only one that can demonstrate what it does not claim.
The demand is real and demonstrated. The supply is, almost entirely, undemonstrable.
Tip and “system” sites dominate Romanian search results for lottery statistics. None publishes a causal evaluation of its own predictions. Not because something is being hidden — but because, measured correctly, there is nothing to show.
A product that implies the odds of winning can be improved pushes people to spend more. That is precisely the profile app stores and regulators treat most harshly.
Any predictive claim is, by construction, a promise that cannot be supported. It is a compliance and reputation liability that grows with every user acquired.
Demand for data exploration — frequencies, streaks, distributions, visual patterns — exists independently of the promise of prediction. It is the legitimate half of the market, and right now nobody serves it cleanly. An honest product does not lose that demand: it wins whatever remains once the false promise is removed.
Two and a half years, roughly sixty model versions, three games. The primary result is negative — and it is published.
We do not sell prediction, because we rigorously measured that it does not exist. That position is not a marketing weakness — it is the one thing in this category a competitor who does sell prediction cannot copy. They cannot publish their audit. We already published ours.
The full study — methodology, code, failures, our own documented errors — is on kort-x.com. Any technical due diligence can reproduce every figure from the validated databases.
In a pari-mutuel game the players are half of the outcome — and their behaviour is measurable.
+20.2%
additional expected value versus a birthday-style ticket, at the same price and the same probability of winning — obtained purely by avoiding crowded combinations.
It does not increase the probability of winning, which remains 1 in 13,983,816 for every combination. The total return on the optimised ticket stays negative (−15.4%). This is exactly how the product will communicate it to the user. Bootstrap adds a second limit: the direction is solid (34 of 49 coefficients exclude zero) but the exact six-number ticket reappeared in only 1 of 400 resamples. The strategy replicates; a specific combination does not.
After sixty versions of fooling ourselves, the burden of proof was set deliberately high.
| Check | Objection it answers | Result |
|---|---|---|
| Rolling-origin cross-validation, held-out Poisson deviance | Does it generalise out of sample? | −38.97% |
| Permutation test, numbers shuffled against counts | Is it signal, or model flexibility? | p ≤ 0,002 |
| Temporal split, 2018–2022 vs 2022–2026 | Is it stable enough to build a product on? | r = 0,890 |
When the association between numbers and counts is destroyed, the fit collapses well below the null. The improvement is not the model bending to fit noise — which is what every earlier “win” in the project turned out to be.
An eight-year-stable human habit, not a moving target. A product built on it needs no constant retraining and does not decay between releases.
An Android app. A data-exploration instrument, not a prediction tool.
These constraints are the product rule, not a footnote. They are also the strongest defence we have with the app stores.
Components built, tested and reproducible. This is the real intellectual property.
Seventy heterogeneous tests corrected as a single Holm family. It runs on any lottery and answers the question the rest of the category avoids.
845 draws with per-category winner counts and prize values, harvested and validated. This is a data asset, not just code — nobody else in the category holds it structured.
Ridge-penalised Poisson over player preference, plus exact combinatorics via the elementary symmetric polynomial e₆(w) — not simulation. Verified: e₆(1,…,1) = 13,983,816.
A selector that refuses to ship a statistically unsupported result, plus an append-only ledger that freezes any claim before the event. Honesty enforced by infrastructure.
The whole stack runs on CPU. The final analysis — 2,565 draws, seventy tests, exact enumeration over 13,983,816 combinations — executes on a laptop in minutes. There is no GPU infrastructure cost.
Four barriers, two of which cannot be bought.
Eight years of prize data, harvested and validated. A competitor can redo the harvest, but starts from zero on validation and on history.
The preference effect is validated on three independent axes. Replicating it demands statistical discipline the category, by definition, does not have.
A competitor selling prediction cannot publish an audit without destroying their own product. The “honest and verifiable” position is occupied once.
Seven Android apps already in the KORT-X portfolio. The build, publish and maintain chain exists — it is not being created for this product.
Patent / trade-secret status and third-party licensing of the expected-value engine: to define
Demand is demonstrated by the existence and ranking of the competition. Sizing remains to be quantified.
| Active players, Romania | TAM |
| Segment interested in statistics | SAM |
| Realistic year-1 target | SOM |
| Competitor monthly traffic | — |
| Category benchmark ARPU | — |
These figures must be measured, not estimated. A real investor checks them first, and any invented approximation destroys the credibility of everything else in the deck.
The structure is clear. Price points remain to be set against market data.
Complete draw archive, ticket checker, basic frequencies. Role: acquisition and trust.
0
Configurable windows, advanced distributions, anti-crowd and expected-value calculator, data export, no ads.
price / month
The randomness audit and expected-value engine, as a service for operators, regulators or academic research.
B2B
Designed from the start as an analysis instrument, not a gambling product.
Honesty here is not decorative morality. It is the cheapest insurance available in a category where the main risk is not competition but removal from the store.
Nothing on this page has been estimated. A real investor checks these lines first.
| Amount sought | — |
| Instrument | — |
| Valuation / cap | — |
| Runway covered | — |
| Capital invested to date | — |
| Product development | % |
| Data and infrastructure | % |
| User acquisition | % |
| Legal and compliance | % |
| Operating reserve | % |
Verifiable facts available now: seven Android apps in the portfolio, the research study published, a working and reproducible analysis engine, 845 draws of prize data harvested.
Loto Analytics
In a category where everyone claims to predict, the product that can show exactly where knowledge stops is the only one that survives being checked.
The lottery is a game of chance. Nothing in this material increases the probability of winning, and the expected return on any ticket remains negative. This presentation describes a statistical analysis instrument, not a betting system.