An app for random number generation you can actually read

The word "for" changes the brief entirely

Most articles that line up RNG apps answer a different question. They tell you which app has the prettiest number wheel, which has the loudest dice sound, which has the highest App Store rating. Those roundups treat the RNG as a widget. Every option ends up looking interchangeable because they all call the same one-line random API under the hood.

An app for random number generation is asked a harder question. It is asked to stand behind a decision: who won the raffle, which name was drawn in the charity event, which of ten Monte Carlo runs gets submitted. When the brief is "for" something, the app has to be defensible. It has to be inspectable, unbiased, and reproducible. None of the apps on those lists meet any of those three bars.

The fix is twelve lines and you can read them on a phone

The unbiased version is one rejection-sampling loop away. You ask mk0r for it in the prompt and the Quick mode output drops this straight into the HTML:

// what every "top RNG app" tutorial shows you
function pickBiased(min, max) {
  // Math.random() returns a 53-bit float in [0, 1)
  // multiplying by range and flooring is NOT uniform
  // for ranges that do not divide 2^53 evenly
  return Math.floor(Math.random() * (max - min + 1)) + min;
}

// 1..100 draws from pickBiased are slightly skewed
// the bias is invisible for one draw, measurable for 10,000.

The left side is what you will find if you view the source of any of the first page results for "random number generator." The right side is what ends up in your single-HTML app because you said, in the prompt, that you wanted a fair draw. Same language, same browser, different math, different outcome for anyone running a million trials.

The measurable part

Concrete numbers, pulled from running a million-draw simulation of both functions against the range 1 to 100:

The draw function is short enough that the moment you stop reading this page and look at the generated output, you can audit it yourself. Twelve lines is the threshold where auditing is realistic. Most stock RNG apps ship bundles in the tens of thousands of minified lines.

How the draw actually travels

The other thing stock RNG apps do not offer is a way to share the draw. You tap the button, see a number, and that is it. There is no artifact anyone else can reproduce, which is why "we used a random number generator" sounds credible and "we used this specific random number generator" does not.

mk0r's generated app makes the draw a travellable object: a URL with a seed in the hash. The seed is hashed through SHA-256 to derive the first byte of the CSPRNG state, and the audit log records every draw with its timestamp.

What the one HTML file contains

The whole generated file is short. The head tag sets up the manifest and the inline icon; the body has the button, the output, and the audit list; the script at the bottom owns the draw. The script looks like this, verbatim:

Three things are worth noticing. First, the seed is read fromlocation.hash, so it never touches a server. Second, the draw loop uses rejection sampling (value >= maxValid) to drop biased overflow, which is the part stock RNG apps skip. Third, every draw is pushed through logDraw, which renders it to a visible list with a timestamp. That list is the audit trail.

Proof of replay

Three people opening the same URL get the same draws in the same order. That is the whole claim, and it is boring to prove because the proof is "just open the link":

If any device produced a different sequence, the source was modified or the CSPRNG was swapped. Either is visible because the whole page is inline: diff the HTML, not an app bundle.

mk0r versus the first-page results

Building it yourself

The whole process is four steps, and nothing between them. No account to create, no review queue, no build step.

Why this does not exist already

Writing a twelve-line unbiased RNG is not hard. The thing that makes this specific app unusual is that nobody had an incentive to build it. A store-listed RNG app gets revenue from ads and upsells, so making it auditable would cost money. A web widget gets revenue from display ads around the button, so shipping a minimal inline HTML file would cost money. Every viable business model around an RNG app argues against the thing that would make it defensible.

mk0r does not need an RNG to make money. It needs you to describe an app and get one back. When the prompt says "auditable, fair, seed-replayable," the generator writes that, because it has no reason not to. The output is the app the market cannot produce because there is no product market fit for a one-HTML-file generator hosted on a link.