Pseudorandom number generator

A PRNG holds an internal state. Each call applies a fixed function to that state, returns a number derived from it, and stores the updated state for next time. Nothing in the process is random. Feed the same seed in and the same sequence comes out, every time, on every machine that implements the arithmetic identically. The output only looks random in the statistical sense: no obvious pattern, a roughly flat distribution, no short repeating cycle.

The formal name in the standards literature is a deterministic random bit generator. NIST's recommendation for them, SP 800-90A, specifies mechanisms built on hash functions or block ciphers, which is the industrial end of the same idea a two-line JavaScript generator sits at.

Daily numbers on Vibe Numbers come from a PRNG, deliberately. core/rng.ts hashes the seed string with xmur3, a small string hash that yields one 32-bit integer, and hands that integer to mulberry32, a compact 32-bit generator returning a float in the range 0 to 1 on each call. mulberry32 advances its state by adding the constant 0x6d2b79f5, then mixes the result with shifts, exclusive-ors and multiplications. Same seed, same numbers, on any device.

Reproducibility is the feature here and a liability elsewhere. A 32-bit state can be recovered by anyone who watches enough output, which rules the generator out for session tokens, keys or a real-money shuffle. Where that line falls, and what the alternatives cost, is the subject of our guide to how random is random.