Random Number Generator

crypto.getRandomValues · uniform · in-browser

Why this is “true random”

Browser Math.random() is fast but uses a deterministic pseudo-random algorithm seeded from process entropy — fine for graphics, not fine for picking a winner or drawing a lottery. This page calls crypto.getRandomValues instead, which is required by the Web Crypto specification to be a cryptographically strong RNG. On every modern browser that means values are pulled from the operating system’s entropy pool (/dev/urandom on Linux/macOS, BCryptGenRandom on Windows), the same source TLS keys come from.

Avoiding modulo bias

Naively writing bytes[0] % 6 to roll a die is slightly biased: 256 is not a multiple of 6, so values 0–3 occur 43 times each and values 4–5 only 42 times each in a perfectly uniform sample of 256 bytes. For a die that’s ~0.5% bias — usually invisible, but it grows quickly with larger ranges. This page uses rejection sampling: it generates uint32 values and discards any that fall in the “remainder” window above the largest multiple of the range that fits. The expected number of rejections is below one, so it’s effectively as fast as the biased version, just correct.

Decimal mode

Decimal values are sampled by drawing 53 random bits (matching IEEE-754 double precision) to form a uniform float in [0, 1), then affinely mapping it to your chosen [min, max] range and rounding to the requested decimal places. This is slightly different from the integer path because the underlying values aren’t all distinct in floating-point — near-zero values have more representable doubles than near-one values. For typical use (random angles, percentages, weights, coordinates) the result is indistinguishable from a continuous uniform distribution.

Unique mode

With No duplicates, the count is capped at the size of the range (you can’t pick 10 unique values out of 5). For small ranges the page uses a partial Fisher–Yates shuffle of the full range; for huge ranges it samples with rejection into a Set until the count is reached. Either way every k-tuple is equally likely, which is what you want for a fair lottery draw.

Privacy

Random number generation is purely arithmetic over locally-sourced bytes — the page never makes a network request to a randomness API. There is no fetch() for entropy, no third-party RNG service, no analytics that watch what you generate. The numbers you draw stay in this tab.