Random Number Generator

Raffles and giveaways are the most obvious case. Number every entry from 1 to N, turn duplicates off, and draw as many winners as prize slots. The result is repeatable enough to defend — screenshot the range, count, and output — and unpredictable enough that no participant can game it.

Teachers use it to cold-call students without bias, assign presentation order, split classes into groups, and generate practice problem sets. Researchers use it to draw random samples from a numbered population, assign participants to control and treatment groups, or shuffle survey question order to control for ordering effects.

Game designers and tabletop players use it as a browser-based replacement for physical dice. A d20 is just a random integer from 1 to 20; a percentile die is 1 to 100. Dungeon Masters can pre-roll a batch of encounter checks before a session so play doesn't stall.

Software engineers reach for it when they need seed values for tests, quick sample data, or a random port number during local development. Marketers use it for A/B bucket assignment when they don't want the overhead of a full experimentation platform.

The generator uses the Web Crypto API's crypto.getRandomValues() when available. That's the same source browsers expose for generating cryptographic keys, session tokens, and one-time passcodes, and it draws from the operating system's cryptographically secure random number generator (CSPRNG) — /dev/urandom on Linux, BCryptGenRandom on Windows, SecRandomCopyBytes on macOS.

For each draw, the generator asks the CSPRNG for a 32-bit unsigned integer, then maps it into your requested range using modulo arithmetic. The 32-bit output has about 4.3 billion possible values, so for any range up to a few million, the modulo bias is negligible — well under one part in a thousand.

The fallback path uses Math.random(), which is a pseudo-random number generator seeded from an unspecified source (typically clock jitter and process state). It's fine for games and sampling but is not appropriate for anything a determined adversary could try to predict — cryptographic key material, session tokens, or high-stakes drawings.

'Without replacement' mode uses a Set to track drawn values and rejection sampling to skip duplicates. When the requested count approaches the size of the range, this can slow slightly, but for typical use (drawing 10 winners from 500 entries, say) the extra work is invisible.

Fix your parameters before you draw. Screenshot the range, count, and duplicates setting alongside the result, so participants can see exactly what was rolled. Changing the range after the fact — even by one — invalidates the fairness of the draw in the eyes of your audience.

For contests with legal weight (sweepstakes, promotions bound by state law), use a certified third-party RNG service or a physical drawing witnessed by a notary. The Federal Trade Commission's endorsement guides and most state promotion laws require documented, tamper-evident randomization for prizes above certain thresholds.

For classroom or team use, generate all rolls at the start of the session in one batch and share the list. That prevents anyone from re-rolling until they get a favorable result — the number-one way informal random draws stop feeling random.

If you need to seed something like a Monte Carlo simulation and want the same result later, this tool won't help — it's intentionally unseeded. Use a language-level RNG (numpy.random.default_rng(seed=42), for example) so runs are reproducible.