Dice Statistics Companion
Paste or upload a provably-fair dice history and get a plain-English fairness report: a letter grade, expected-vs-observed band table, quantiles, streak structure, randomness tests, and a statistical-power meter that tells you how much to trust the result. The Statistics Companion measures how a finished roll history compares to a fair uniform model. It cannot predict future rolls, change the house edge, or prove a casino is rigged — an unusual sample is a reason to verify, not a verdict.
Frequently asked questions
How many dice rolls do I need for the statistics to mean anything?
A few hundred rolls show only large effects. Aim for 1,000 or more for a dependable read, and 5,000+ for strong randomness tests. Below ~30 rolls the numbers are noise.
Can this tool predict my next roll?
No. Provably-fair dice outcomes are independent once the seed is set. The companion only describes a finished history; it cannot forecast future rolls or beat the house edge.
What does a failing chi-square or KS test prove?
On its own, nothing definitive. It means the sample looks unusual under a fair model, which is a reason to re-verify your export, seeds, and algorithm mapping — not proof of cheating.
Why is my high-side rate not exactly 50%?
Random samples wobble. The companion shows a 95% confidence interval around your rate; if 50% sits inside that interval, your result is consistent with fair dice.
Is the fairness grade an official rating?
No. It is a convenience score blending several rough tests so you can skim results. Always read the individual test verdicts and verify the seed commitment for anything important.