Random number generator

How to use it

1. Set the minimum and maximum. Both ends are included, so 1 to 6 can produce a 1 or a 6.
2. Choose how many numbers you want — up to ten thousand at a time.
3. Turn on “no repeats” for a raffle or a lottery draw. Leave it off for dice, simulations and anything where each draw must be independent.
4. Sort the output if you need it in order, then copy it as a list or a comma-separated line, or download it as a text file.
5. The Dice and Coin tabs cover the two cases people ask for most often, with the same generator behind them.

Where the randomness comes from

Every value here is drawn from crypto.getRandomValues, the browser’s cryptographically secure pseudo-random generator. It is seeded from operating-system entropy — timing jitter, hardware sources, interrupt noise — and is the same facility used to generate encryption keys and session tokens.

That is a meaningful step up from Math.random(), which most online generators use. Math.random() is a fast pseudo-random algorithm (xorshift128+ in V8) with an internal state that is not secret and is not reseeded. It is perfectly good for shuffling a carousel and completely unsuitable for a prize draw, because someone who observes enough output can reconstruct the state and predict the rest.

The modulo bias, and why it matters

The obvious way to get a number in a range is to take a random integer and apply the remainder operator. It is also subtly wrong. Suppose you have a generator producing 0–255 and you want 0–99:

value % 100

value 0–55   →  0–55 can each be produced 3 ways  (v, v+100, v+200)
value 56–99  →  56–99 can each be produced 2 ways (v, v+100)

Low results are 50% more likely than high ones.

The fix is rejection sampling: discard any value in the unbalanced tail and draw again. This tool computes the largest multiple of the range that fits in 2³², rejects anything at or above it, and only then takes the remainder. The cost is an occasional extra draw; the benefit is that every number in your range is exactly as likely as every other. Decimals use a full 53-bit mantissa built from two 32-bit words, rather than the 32 bits most implementations settle for.

With or without repeats

When the request is dense — say 40 different numbers from 1 to 50 — rejection sampling would spend most of its time rediscovering numbers it already has. The tool switches to a partial Fisher–Yates shuffle over the whole range in that case, which produces the draw in a single pass regardless of how close to saturation it gets.

Common ranges

Running a draw people will trust

The generator is rarely the part anyone disputes — the process around it is. A few things that help:

• Fix the rules before you draw. Publish the range, the number of winners and the tie-break rule in advance.
• Number the entrants first, in a list you can show afterwards, then draw index numbers rather than names.
• Record the draw. A screen recording of the range being set and the button being pressed answers almost every objection.
• Decide the redraw rule beforehand — what happens if a winner is ineligible, or if the same number comes up twice.
• Keep the output. Nothing is stored here, so download the text file at the time.

Dice and coins

The Dice tab rolls any number of dice from d4 to d100 with an optional modifier, shows each die individually and keeps the last twenty rolls. Because every die is an independent draw from the same secure generator, it is fairer than a physical die — injection-moulded plastic dice have measurable bias from the hollowed-out pips and from mould draft.

The Coin tab flips up to a thousand times and reports the head and tail counts alongside the longest run. That last figure surprises people: in a hundred fair flips a run of six or seven identical results is entirely normal, and its absence would be the suspicious result. Humans asked to fake a random sequence almost never write a run that long, which is precisely how such fakes are detected.