Randomness methodology

Random number source

Every pick uses window.crypto.getRandomValues(new Uint32Array(1)), exposed by all modern browsers. This function taps the operating system\'s cryptographic RNG — the same source used to generate keys, tokens, and nonces. Its output is:

• Unpredictable — you cannot infer future values from past ones.
• Uniformly distributed — every value in the range has equal probability.
• Fast — a single call takes microseconds.

For pick-one selection, we take a random Uint32 and modulo it by the list length. To eliminate the modulo bias that appears when the list length doesn\'t divide 2³² cleanly, we discard values above Math.floor(2^32 / listLength) × listLength and re-sample. In practice this rejection almost never fires, but it guarantees perfect uniformity.

Wheel-of-names spin animation

The wheel is a display layer, not a fairness layer. We pick first, then compute a target angle such that the pointer lands on the picked wedge, then animate the wheel via CSS transforms. The animation timing curve is cubic-bezier(0.22, 0.11, 0.15, 1) — a natural deceleration curve. The result is that the wheel feels physical without being biased.

Without-replacement mode

When enabled, the picker maintains an internal list of "still-in-the-hat" items. Each pick removes the chosen item until the hat is empty, then resets. Both the current hat state and the reset event are handled deterministically — no hidden bias.

Weighted picks

When weights are provided, we build a cumulative-probability array and sample from it using the same crypto RNG. Weights are floating-point and normalised, so a weight of 2.0 vs 1.0 doubles the probability, exactly.

Distribution testing

We run automated tests that spin each picker 10,000 times and confirm the empirical distribution is within 2σ of expected uniform. Any dataset that fails triggers an audit. As of this writing, no picker has ever failed distribution testing.

Reproducibility for high-stakes draws

Live draws (e.g. school raffles, streamer giveaways) benefit from documented reproducibility. We recommend:

1. Record the draw on video (screen-share works).
2. Have a witness present.
3. Save the URL — our shared results are permalinkable.
4. Publish the T&Cs of the draw before running it.

See our How To Run A Fair Giveaway guide for the full playbook.

Third-party verification

Our picker code is not open source, but the crypto.getRandomValues path is a well-documented browser standard. Third-party libraries like random.org use TRNG (true random number generators) sourced from atmospheric noise; that is arguably more random but not more fair — for our use cases, cryptographic PRNG is indistinguishable.

What we don\'t do

• We do not weight results toward affiliates or paid partners. There is no A/B test on the pick itself.
• We do not use ML-driven "engagement optimisation" on the result. What you see is what the picker picked.
• We do not remember your previous picks unless you enable "without replacement" mode explicitly.