RNG Tournament — Powerball
Ten random number generators — eight classical, plus ChaCha20 and Web Crypto — each guess one Powerball number per draw since the October 2015 matrix change to 1–26. On an honest draw, each RNG should hit ≈3.85% — the 1-in-26 baseline. Who matches it closest?
Standings · 1,339 draws · 1 pick each
● clean spread| Rank | RNG | Hits | Rate | z vs expected | Best hit streak | Longest miss |
|---|---|---|---|---|---|---|
| 1 | Xorshift128 | 62 | 4.63% | +1.49 | 2 | 108 |
| 2 | PCG | 62 | 4.63% | +1.49 | 2 | 87 |
| 3 | ChaCha20 | 58 | 4.33% | +0.92 | 2 | 75 |
| 4 | MSWS | 57 | 4.26% | +0.78 | 2 | 92 |
| 5 | SplitMix | 55 | 4.11% | +0.50 | 2 | 134 |
| 6 | Fibonacci | 54 | 4.03% | +0.36 | 2 | 105 |
| 7 | Web Crypto | 53 | 3.96% | +0.21 | 2 | 92 |
| 8 | Mulberry32 | 52 | 3.88% | +0.07 | 1 | 125 |
| 9 | Math.random | 47 | 3.51% | -0.64 | 2 | 104 |
| 10 | LCG | 45 | 3.36% | -0.92 | 1 | 144 |
Hits vs Binomial expectation
How it works
The matrix change. On October 7, 2015, the Powerball pool changed from 1–35 to 1–26. To keep the test apples-to-apples, this tournament uses only post-change draws. The existing whites tournament uses all of history because the white-ball pool has also been 1–69 since that date.
The game. Each RNG is reseeded once at start and then asked for one Powerball prediction per draw — a single integer in 1–26 — for all 1,339 post-change draws. A hit is an exact match with the actual Powerball.
What to expect. Under a fair lottery, hits per RNG follow Binomial(n, 1/26). That gives expected 51.5 hits with σ ≈ 7.04 — so a ±2σ range of roughly ±14 hits. If any RNG consistently finished much higher than the others, that'd be an indication the RNG is somehow aligned with the lottery's implicit pattern (or broken in a way that happens to bias toward frequent-PB values).
Unlike the whites tournament, there's no “best of N picks” here — with only 26 possibilities, 50 tries would almost always land a match and the test becomes trivial. One guess per draw keeps the Bernoulli structure clean.