How the lab works
The short version: lottery draws are designed, certified and audited to be random, and the lab checks that claim with the same tools a statistician would. Here is exactly what it does.
The tests
Every number equally likely. A chi-square test on how often each number was drawn. Because numbers can't repeat within one draw, the counts vary less than the textbook test assumes, so the statistic is scaled by (N−1)/(N−K) to stay exact.
Shape tests. The sum of each draw, its odd/even split, its low/high split, neighboring numbers and numbers repeated from the previous draw, each compared with its exact distribution (worked out by counting combinations, not simulated), except the neighbors test, which uses 200,000 seeded simulated draws.
Memory tests. A Wald–Wolfowitz runs test on high and low totals, and the correlation between one draw's total and the next.
Reading them. A p-value below 0.05 means "a fair machine does this less than one time in twenty". With about ten tests per game, a flag or two somewhere is expected. A real problem would show up in the same test, draw after draw.
The generators
Hot, cold and overdue strategies weight numbers by recent counts or current absence. The Bayesian Oracle uses the posterior mean of a Dirichlet prior over all draws. The Bell Curve keeps tickets to common totals and splits.
The Lab Pick draws a few hundred random tickets and keeps the one other players are least likely to have chosen. Its popularity model follows published studies of hand-picked numbers: birthdays (1–31) and months (1–12), 7, straight lines and sequences, repeated endings, multiples of five, and past winning numbers. The weights are an informed estimate, not measured sales data.
The proving ground walks forward through history. Before each draw, each strategy buys ten tickets using only the draws before it, with fixed seeds so the result can be reproduced.
The value calculator
Each prize tier's exact probability times its prize. The jackpot is shared: with λ other matching tickets expected (tickets sold × popularity ÷ combinations), your expected share is (1 − e−λ) ÷ λ. Mega Millions' built-in multiplier averages exactly 3x (2x 15/32, 3x 10/32, 4x 4/32, 5x 2/32, 10x 1/32). Fantasy 5's Doubler, on about one ticket in ten, averages 1.1x.
Taxes are simplified to the top federal rate (37%) and Arizona's flat 2.5% on prizes over $600. Ticket sales follow a rough curve fitted to public figures and can be overridden.
Where the numbers come from
- Powerball and Mega Millions history: New York State's Open Data portal (data.ny.gov), the lotteries' public result datasets.
- Latest results and jackpots for every game: the Arizona Lottery's public results service.
- Cash values: powerball.com and megamillions.com.
- Arizona games: the Arizona Lottery publishes the last 180 days, so the Arizona history here starts in May 2026 and grows with every draw.
Each draw is checked against its game's rules (count, range, no repeats, bonus ball range) before it's stored. Always confirm a win with the lottery itself.
What the lab can't do
It can't predict a draw. Nothing can: each combination has exactly the same chance, every time, and the backtest shows every strategy matching numbers at the rate pure chance predicts. What it can do is tell you how a ticket would be shared if it won, whether tonight's jackpot changes the math, and whether the machines are behaving. Have fun with it, and play only what you'd happily spend on a night out.