Fairness proof: 100,000 live spins, chi-square tested, in your own browser

How the test works

Each spin in this test calls our production randomWeightedIndex(weights) function — the very same code that runs when you spin the wheel on the home page. We feed it the same weight array, then bucket every result into a histogram. Once 100,000 results are in (about 100 ms on a modern laptop) we compute the chi-square statistic Σ((observed − expected)² / expected) and convert it to a two-sided p-value.

What a passing result means

• P-value between 0.05 and 0.95 — distribution is uniform, no detectable bias.
• P-value below 0.05 — the observed counts are more uneven than 95% of fair runs would be. (Statistically possible by chance — re-run a few times.)
• P-value above 0.95 — the counts are suspiciously too even, which can indicate a non-random source. (Also possible by chance.)

Why we use a CSPRNG, not Math.random()

Math.random() is a pseudo-random number generator (PRNG). The browser seeds it once when the page loads, and from that point on the entire sequence is deterministic. In short runs, it can also produce subtle bias depending on the implementation. Cryptographically-secure RNGs draw on hardware entropy (mouse movements, keyboard timings, OS-level entropy pool) and are designed to resist prediction. We use crypto.getRandomValues() exclusively, with rejection sampling for the integer step so there's no modulo bias.