The Hidden Math of Game Night: Why Random Is Hard

Game night runs on chance. The dice decide whether the dragon lands a hit, the shuffle decides which card you draw, the app decides whatever the app decides. We accept all of it without much thought, which is fair enough — arguing about probability is not why anyone sits down at the table.

But almost none of it is as random as it looks. Your dice are probably slightly biased, your shuffle almost certainly is not thorough enough, and the app rolling on your behalf cannot prove a thing about what it did. The one corner of the internet that actually solved the verification problem is an unlikely one — the provably fair systems documented at this dice resource — and the technique behind it is genuinely clever.

Your dice are a little bit wrong

Geometrically, a twenty-sided die is beyond reproach. It is one of the five Platonic solids, and every face is identical to every other. A perfect one is perfectly fair.

Nobody sells you a perfect one.

What the factory does

Polyhedral dice are injection-moulded in two halves, which leaves a seam and a bit of excess plastic where the mould closes. Manufacturers clean this up by tumbling the dice in an abrasive medium.

It works, but tumbling is not selective. It wears down whatever it touches, and it does not touch twenty faces equally.

Then there is the ink. Numbers sit in engraved recesses, and a 20 has noticeably more material removed than a 1. On a light plastic die, that adds up.

How to check yours

Dissolve salt into warm water until a die floats — about six tablespoons per cup. Spin it and note which face surfaces. Do it twenty or thirty times.

A balanced die will surface differently each time. A weighted one keeps presenting the same face, because the heavy side settles down. It takes ten minutes and it will settle at least one long-running argument in your group.

Seven shuffles, and not one fewer

Card randomness is worse, because everyone thinks they are doing it properly.

In 1992, mathematicians Dave Bayer and Persi Diaconis worked out how many riffle shuffles it takes to genuinely randomise a standard 52-card deck. The answer is seven. Before that, the deck retains recoverable structure from its previous order.

Which means the three quick shuffles most of us do at the table are not producing a random deck. They are producing a mildly rearranged version of the last one — and in a deckbuilder where you just fed your discard pile back in, that matters more than people assume.

Overhand shuffling, the fold-and-drop method most casual players actually use, is far worse. It takes thousands of repetitions to reach the same result.

The app on the table

Digital rolls introduce a different problem entirely.

An app does not roll anything. It calls a pseudorandom number generator — an algorithm that takes a starting value, the seed, and produces a sequence that looks statistically random but is completely deterministic. Feed it the same seed and it produces the same sequence, forever.

That is not automatically bad. A well-seeded generator produces output no human could distinguish from true chance.

The problem is that you cannot check. You do not see the seed. You cannot recompute the roll. If the software had a bug — and dice apps have shipped with genuinely broken distributions before — the output would still look completely normal to you.

A physical die can at least be inspected and floated in salt water. A digital roll offers you nothing but the developer’s word.

The trick that makes a roll checkable

Which brings us back to the surprisingly elegant solution mentioned at the top.

Before anything is rolled, the server generates a secret seed and publishes a cryptographic hash of it — a fingerprint that reveals nothing about the seed but also cannot be quietly changed later.

You supply a second seed of your own. Every result comes from combining both, plus a counter, through a cryptographic function.

When you are done, the server reveals its original seed. You hash it yourself, confirm it matches the fingerprint you were given at the start, and recompute every result independently.

If anything had been altered mid-session, the hashes would not line up. The person doing the checking is the person with the most reason to catch cheating — which is a much better arrangement than trusting a company to audit itself. There is no technical reason a virtual tabletop could not do exactly this.

Does any of it change game night?

Not really, and that is worth saying plainly. A slightly biased d20 is nothing next to a house rule, a generous DM, or the fact that somebody always forgets their modifier.

Seven shuffles is a genuinely useful thing to know. The float test is a fun ten minutes. Everything else here is mostly the pleasure of understanding the machinery underneath something you already enjoy.

What is interesting is the superstition. Nobody at the table actually believes the die can hear them. Everybody talks to it anyway, keeps a lucky one, retires a cursed one, rolls into a tower because it feels more legitimate.

That is not a failure of mathematical literacy. It is that human beings are deeply uncomfortable with outcomes nobody controls, and blaming a small plastic object is far more satisfying than accepting that the universe simply did not go your way this time.