This generator pulls bytes from window.crypto.getRandomValues() when available - the browser's CSPRNG, suitable for security-sensitive use - with rejection sampling so the modulo doesn't bias the character distribution. It falls back to Math.random() only if crypto is unavailable.

Modern V8/SpiderMonkey actually implement Math.random() with xorshift128+, so the output is high-quality statistically - but it's not cryptographically secure: the internal state can be reconstructed from a handful of outputs. Don't use it for tokens, secrets, or anything an attacker would benefit from predicting.

Fisher-Yates shuffle backed by a uniform integer generator on top of crypto.getRandomValues() (with rejection sampling so each permutation is equally likely). Takes a comma- or space-separated list and returns a fair shuffle.

Plot of points $(x, y)$ where both coordinates come from Math.random(). The plot runs in $O\left(\mathrm{n}\right)$ per call - constant work per sample.

With $n = 250$ points the four quadrants of $[0,1]^2$ won't have exactly the same count even from a perfect uniform RNG; that's a property of random sampling, not a flaw of the generator. The Law of Large Numbers makes the imbalance shrink as $n$ grows.

Same plot, but using a small xorshift generator (Marsaglia 2003) seeded once from crypto.getRandomValues(). It's the x ^= x << 13; x ^= x >> 17; x ^= x << 5; recurrence - fast, statistically decent at short lengths, and a useful baseline for comparison.

For full statistical evaluation, use the NIST-style testing kit.