Methodology
We don't predict the future. Lottery draws are independent random events — no model can beat pure chance over the long run. What we do is compute honest statistics on real historical draw data and expose them, along with clearly-labeled AI-generated commentary, so you can play more thoughtfully.
1. Data sources
Historical draw data is pulled from official lottery operators and public results feeds, then normalized into our draws table. Every draw stores the date, main numbers, bonus numbers and jackpot amount where available. You can inspect the raw data on any results page.
2. Frequency, hot & cold
For each lottery we count how many times every number in the pool has appeared across the last N draws. "Hot" numbers are the top decile by frequency; "cold" numbers are the bottom decile. Frequency is a descriptive statistic — it does not imply a number is more or less likely on the next draw.
3. Overdue / gap analysis
For every number we compute the number of draws since it last appeared ("gap"). A large gap is often called "overdue" — again, this is descriptive. In a truly uniform draw, past absence tells you nothing about future presence (the gambler's fallacy).
4. Odd / even and sum distributions
Winning combinations across most lotteries cluster in a predictable range for their sum and odd/even split (because there are more combinations in the middle of the distribution). We show the observed histogram and the theoretical expectation so you can see how they compare.
5. Pair frequency
We compute how often each pair of main numbers has been drawn together. This is useful for wheeling and for spotting artifacts in older draw sets, not for prediction.
6. AI picks & rationale
Our AI pick generator uses the Lovable AI Gateway (Google Gemini family). It receives the lottery's pool configuration and the observed hot / overdue / pair-frequency arrays as context, and is asked to produce a valid pick plus a short rationale. The rationale is explanatory, not predictive — it describes which statistical signals influenced the pick, not why it "will" win.
We do not train custom models on user data. Every request is stateless.
7. Confidence score
The confidence score shown on each pick is a synthetic 0-100 measure that combines how "typical" the pick is versus the historical distribution (sum, odd/even, hot coverage). A high score means the pick looks statistically ordinary; it does not mean it is more likely to win. All valid tickets have identical mathematical odds.
8. Backtesting
Our backtest tool replays a strategy against real historical draws and reports match counts and hypothetical prize tiers. Backtests are for educational curiosity — past performance is not predictive of future draws.
9. What we will never do
- Claim a pick is "guaranteed" or "likely" to win.
- Sell tickets on your behalf.
- Use engineered "insider" data — there is none.
- Hide our reasoning behind a black box. If you want the source stats behind any pick, they're on the analysis page.
10. Play responsibly
Lotteries have negative expected value. Never play with money you can't afford to lose. See responsible gaming.