We ran the wheel's real picking code 16,004,000 times, in a browser and in Node.js, and checked every result. Here is exactly how a winner is picked, what we tested, every number we got, and what a test like this cannot tell you.
None of the 24 tests found a bias. With 2, 6, 10 and 37 entries, every entry won its fair share to within a small fraction of a percentage point, which is the amount chance alone gives. The spin time, the order of wins and elimination mode made no difference either. This is a strong check, but it is a sample, not a proof: see what this test cannot tell you.
Everything happens in your browser, in this order:
crypto.getRandomValues, the cryptographically secure generator built into every modern browser, which draws on randomness collected by your device's operating system. It asks for 256 at a time and uses them one by one. Each is a whole number from 0 to 4,294,967,295, divided by 4,294,967,296 to give a number u from 0 up to (not including) 1. The wheel never uses Math.random, the older and weaker generator.These are the lines that do it, copied by the test script straight out of the wheel's JavaScript file:
var RB = new Uint32Array(256), RI = 256;
function rnd() { if (RI >= 256) { crypto.getRandomValues(RB); RI = 0; } return RB[RI++] / 4294967296; }
var n = it.length, a = 2 * Math.PI / n, win = Math.floor(rnd() * n);
var target = -(win * a + (0.12 + rnd() * 0.76) * a);An entry you type twice gets two slices, and twice the chance. That is the only way to weight an entry, so the weighted test below uses repeated entries.
Because u comes from one of 4,294,967,296 (232) equally likely numbers, we can count exactly how many of them pick each entry, without any sampling. With 2, 4, 8 or any power of two entries, every entry gets exactly the same count. Otherwise 4,294,967,296 does not divide evenly, so some entries get one more value than others. The difference is real but far too small to matter or to measure:
| Entries | Values per entry | Entries with one extra | Largest difference in chance |
|---|---|---|---|
| 2 | 2,147,483,648 | none | none |
| 6 | 715,827,882 or 715,827,883 | 4 of 6 | 1 in 715,827,882 of the chance |
| 10 | 429,496,729 or 429,496,730 | 6 of 10 | 1 in 429,496,729 of the chance |
| 37 | 116,080,197 or 116,080,198 | 7 of 37 | 1 in 116,080,197 of the chance |
| 100 | 42,949,672 or 42,949,673 | 96 of 100 | 1 in 42,949,672 of the chance |
| 1,000 | 4,294,967 or 4,294,968 | 296 of 1,000 | 1 in 4,294,967 of the chance |
On a 10-entry wheel, for example, 6 entries win with probability 429,496,730 / 4,294,967,296 and 4 with 429,496,729 / 4,294,967,296: a gap of about one in 430 million of a 10% chance. You would need many quintillions of spins to notice it. Even at the 1,000-entry maximum, the gap is 1 in 4,294,967 of each entry's 0.1% chance. The test script also checks the boundaries between every pair of neighbouring entries, so we know the computer's arithmetic matches this count exactly.
The real code, not a copy. The test script opens the wheel's JavaScript file, cuts out the random number buffer, the spin function and the function that reads which slice is under the pointer, and runs them unchanged. Only the page around them is replaced: no drawing, no sound, and the animation jumps straight to its final frame. To check the animation as well, 2,000 spins per run play every frame at 60 frames a second.
Two random number generators. The same test ran in headless Chromium 141.0.7390.37, using the browser's own crypto.getRandomValues, and in Node.js v22.22.0, whose crypto.getRandomValues comes from a different library (OpenSSL). Nothing is seeded, because crypto.getRandomValues cannot be seeded; every run gives new numbers.
What ran. 1,000,000 spins each on wheels of 2, 6, 10 and 37 entries; 1,000,000 spins of a wheel with one entry on 3 slices, one on 2 and one on 1; 250,000 spins of a 10-entry wheel at each spin time; 500,000 back-to-back pairs on 6 entries; and 250,000 full draws of a 4-entry wheel in elimination mode, removing each winner as the wheel does. The wheel was never reset between spins, as on the site.
The test. For each wheel we counted the wins and compared them with what a fair wheel gives, using Pearson's chi-square test. The p-value is the chance that a perfectly fair wheel would give counts at least this uneven. Fair wheels give p-values spread evenly between 0 and 1, so a single small one is expected now and then.
Decided in advance. We fixed the rule before the run: a test counts as a bias if its p-value is below 0.001 and a fresh run of that test comes back below 0.001 again. If one had, we would have fixed the wheel instead of publishing this page.
Run it yourself. The script needs Node.js and Playwright. It takes the path of the wheel's JavaScript file, finds the three functions by their text (and stops if they are not there), runs every test above in Node.js and then in headless Chromium, and writes the counts, chi-square values and p-values to a JSON file. Our run took about 2 minutes on one processor core. The chi-square p-values come from the regularized incomplete gamma function, which the script checks against textbook table values before it starts.
Run on November 13, 2026. Chromium 141.0.7390.37 (headless, Playwright); Node.js v22.22.0 (OpenSSL 3.5.4). Wheel code: assets/stw.js, SHA-256 9aa82eda9563c49d… (the file this site serves now).
Every test, both runs. A p-value of 0.001 or more means no sign of bias.
| Test | Spins | Chi-square | df | p-value | Result |
|---|---|---|---|---|---|
| 2 entries, equal slices | 1,000,000 | 0.02 | 1 | 0.88 | No sign of bias |
| 6 entries, equal slices | 1,000,000 | 1.57 | 5 | 0.90 | No sign of bias |
| 10 entries, equal slices | 1,000,000 | 7.91 | 9 | 0.54 | No sign of bias |
| 37 entries, equal slices | 1,000,000 | 30.32 | 36 | 0.74 | No sign of bias |
| Repeated entries as weights (3 : 2 : 1) | 1,000,000 | 2.71 | 2 | 0.26 | No sign of bias |
| 10 entries, 3 second spin | 250,000 | 7.42 | 9 | 0.59 | No sign of bias |
| 10 entries, 5 second spin | 250,000 | 17.30 | 9 | 0.04 | No sign of bias |
| 10 entries, 8 second spin | 250,000 | 3.73 | 9 | 0.93 | No sign of bias |
| 10 entries, 12 second spin | 250,000 | 7.71 | 9 | 0.56 | No sign of bias |
| Winner independent of spin time | 1,000,000 | 30.15 | 27 | 0.31 | No sign of bias |
| Back-to-back pairs, 6 entries | 1,000,000 | 37.05 | 35 | 0.37 | No sign of bias |
| Elimination mode: all 24 orders of 4 | 1,000,000 | 21.40 | 23 | 0.56 | No sign of bias |
| Test | Spins | Chi-square | df | p-value | Result |
|---|---|---|---|---|---|
| 2 entries, equal slices | 1,000,000 | 0.48 | 1 | 0.49 | No sign of bias |
| 6 entries, equal slices | 1,000,000 | 4.47 | 5 | 0.48 | No sign of bias |
| 10 entries, equal slices | 1,000,000 | 17.75 | 9 | 0.04 | No sign of bias |
| 37 entries, equal slices | 1,000,000 | 30.75 | 36 | 0.72 | No sign of bias |
| Repeated entries as weights (3 : 2 : 1) | 1,000,000 | 0.55 | 2 | 0.76 | No sign of bias |
| 10 entries, 3 second spin | 250,000 | 8.18 | 9 | 0.52 | No sign of bias |
| 10 entries, 5 second spin | 250,000 | 10.68 | 9 | 0.30 | No sign of bias |
| 10 entries, 8 second spin | 250,000 | 15.27 | 9 | 0.08 | No sign of bias |
| 10 entries, 12 second spin | 250,000 | 7.41 | 9 | 0.59 | No sign of bias |
| Winner independent of spin time | 1,000,000 | 36.66 | 27 | 0.10 | No sign of bias |
| Back-to-back pairs, 6 entries | 1,000,000 | 27.97 | 35 | 0.79 | No sign of bias |
| Elimination mode: all 24 orders of 4 | 1,000,000 | 21.40 | 23 | 0.56 | No sign of bias |
Across the 24 tests, the smallest p-value was 0.04 and 2 fell below 0.05, against about 1.2 expected by chance alone. "df" is degrees of freedom: one less than the number of outcomes counted.
The biggest gap between an entry's share of wins and its fair share, in the browser run:
| Entries | Fair share | Biggest gap | In standard errors | Chi-square (df) | p-value |
|---|---|---|---|---|---|
| 2 | 50% | 0.008 pts | 0.15 | 0.02 (1) | 0.88 |
| 6 | 16.667% | 0.041 pts | 1.09 | 1.57 (5) | 0.90 |
| 10 | 10% | 0.046 pts | 1.53 | 7.91 (9) | 0.54 |
| 37 | 2.703% | 0.035 pts | 2.13 | 30.32 (36) | 0.74 |
A gap is measured in standard errors because that is the size chance alone gives: with 1,000,000 spins, a fair 10-entry wheel's entries usually land within about ±0.06 points of 10%. On a 37-entry wheel, at least one of the 37 entries lands beyond 3 standard errors in about 1 run in 11 by chance. The single biggest gap in either run, entry 4 on the 10-entry wheel in the Node.js run (2.49 standard errors above), got a fresh 5,000,000-spin browser run of its own: there it was 1.02 standard errors above, and that whole wheel scored chi-square 2.81 (df 9, p = 0.97). A biased slice would have stayed off.
| Entry | Wins | Share | Gap from 1 in 2 |
|---|---|---|---|
| 1 | 499,923 | 49.992% | −0.008 pts |
| 2 | 500,077 | 50.008% | +0.008 pts |
| Entry | Wins | Share | Gap from 1 in 6 |
|---|---|---|---|
| 1 | 166,774 | 16.677% | +0.011 pts |
| 2 | 166,529 | 16.653% | −0.014 pts |
| 3 | 166,260 | 16.626% | −0.041 pts |
| 4 | 166,846 | 16.685% | +0.018 pts |
| 5 | 166,783 | 16.678% | +0.012 pts |
| 6 | 166,808 | 16.681% | +0.014 pts |
| Entry | Wins | Share | Gap from 1 in 10 |
|---|---|---|---|
| 1 | 99,995 | 10.000% | −0.001 pts |
| 2 | 99,851 | 9.985% | −0.015 pts |
| 3 | 99,743 | 9.974% | −0.026 pts |
| 4 | 100,175 | 10.018% | +0.018 pts |
| 5 | 100,100 | 10.010% | +0.010 pts |
| 6 | 100,331 | 10.033% | +0.033 pts |
| 7 | 99,540 | 9.954% | −0.046 pts |
| 8 | 100,204 | 10.020% | +0.020 pts |
| 9 | 99,645 | 9.965% | −0.036 pts |
| 10 | 100,416 | 10.042% | +0.042 pts |
| Entry | Wins | Share | Gap from 1 in 37 |
|---|---|---|---|
| 1 | 26,932 | 2.693% | −0.010 pts |
| 2 | 27,175 | 2.718% | +0.015 pts |
| 3 | 26,887 | 2.689% | −0.014 pts |
| 4 | 26,857 | 2.686% | −0.017 pts |
| 5 | 26,721 | 2.672% | −0.031 pts |
| 6 | 26,992 | 2.699% | −0.004 pts |
| 7 | 27,099 | 2.710% | +0.007 pts |
| 8 | 27,085 | 2.709% | +0.006 pts |
| 9 | 27,135 | 2.714% | +0.011 pts |
| 10 | 27,007 | 2.701% | −0.002 pts |
| 11 | 27,070 | 2.707% | +0.004 pts |
| 12 | 26,893 | 2.689% | −0.013 pts |
| 13 | 27,045 | 2.705% | +0.002 pts |
| 14 | 27,143 | 2.714% | +0.012 pts |
| 15 | 26,899 | 2.690% | −0.013 pts |
| 16 | 26,998 | 2.700% | −0.003 pts |
| 17 | 26,808 | 2.681% | −0.022 pts |
| 18 | 26,892 | 2.689% | −0.014 pts |
| 19 | 27,039 | 2.704% | +0.001 pts |
| 20 | 27,251 | 2.725% | +0.022 pts |
| 21 | 27,157 | 2.716% | +0.013 pts |
| 22 | 26,932 | 2.693% | −0.010 pts |
| 23 | 27,166 | 2.717% | +0.014 pts |
| 24 | 26,980 | 2.698% | −0.005 pts |
| 25 | 27,212 | 2.721% | +0.018 pts |
| 26 | 27,087 | 2.709% | +0.006 pts |
| 27 | 27,109 | 2.711% | +0.008 pts |
| 28 | 27,189 | 2.719% | +0.016 pts |
| 29 | 27,147 | 2.715% | +0.012 pts |
| 30 | 26,789 | 2.679% | −0.024 pts |
| 31 | 26,681 | 2.668% | −0.035 pts |
| 32 | 26,902 | 2.690% | −0.013 pts |
| 33 | 27,257 | 2.726% | +0.023 pts |
| 34 | 27,122 | 2.712% | +0.009 pts |
| 35 | 27,286 | 2.729% | +0.026 pts |
| 36 | 27,071 | 2.707% | +0.004 pts |
| 37 | 26,985 | 2.699% | −0.004 pts |
A wheel of 6 slices with A on 3, B on 2 and C on 1 should give A half the wins, B a third and C a sixth.
| Entry | Slices | Wins | Share | Expected |
|---|---|---|---|---|
| A | 3 of 6 | 500,211 | 50.021% | 50.000% |
| B | 2 of 6 | 332,662 | 33.266% | 33.333% |
| C | 1 of 6 | 167,127 | 16.713% | 16.667% |
Chi-square 2.71 with 2 degrees of freedom, p = 0.26.
The same 10-entry wheel at each Spin time setting. Each row is tested on its own, and the last line tests whether the four settings differ from each other (a chi-square test of independence on the 4 × 10 table).
| Spin time | Spins | Lowest entry | Highest entry | p-value |
|---|---|---|---|---|
| 3 seconds | 250,000 | 9.92% | 10.09% | 0.59 |
| 5 seconds | 250,000 | 9.87% | 10.13% | 0.04 |
| 8 seconds | 250,000 | 9.94% | 10.05% | 0.93 |
| 12 seconds | 250,000 | 9.91% | 10.06% | 0.56 |
Do the settings differ? Chi-square 30.15 with 27 degrees of freedom, p = 0.31. As expected: the spin time is read after the winner is picked.
If one spin could influence the next, some of the 36 possible pairs on a 6-entry wheel would come up more than others. Over 500,000 pairs, chi-square 37.05 (df 35), p = 0.37. The same entry won twice in a row 16.69% of the time, against 1 in 6 (16.67%) for a fair wheel.
Drawing all 4 entries of a wheel one by one, removing each winner, can give 24 different orders. Each should come up 1 time in 24 (4.17%). Over 250,000 full draws all 24 orders appeared, from 4.10% to 4.23% of draws; chi-square 21.40 (df 23), p = 0.56.
In all 16,000,000 quick spins and 4,000 fully animated ones, the pointer ended on the picked slice every time, between 12.00% and 88.00% of the way across it, never on a line.
This runs 100,000 spins of the same two steps (crypto.getRandomValues, then floor(u × n)) in your browser, and tests the counts the same way. It takes well under a second. Most runs give a p-value above 0.05; about 1 run in 20 gives less by chance, and that is what a fair wheel does.
Press the button to run the test.
Questions about this test? Please contact us.
Thanks, your idea was sent.