When bad randomness made history

Randomness failures are invisible until they're catastrophic. Three stories make the point better than any theory.

In 1969, the US draft lottery drew birthdates from a glass jar to decide who went to Vietnam. The capsules were added to the jar month by month and stirred poorly — December's capsules, added last, stayed near the top and got drawn early. Statisticians later showed late-year birthdays were assigned dangerously low numbers far more often than chance allows. A physical shuffle done badly, with lives on the line.

The inside job and the clock

In the 2000s, the security director of a US multi-state lottery — the man in charge of the random number generator — installed code that made winning numbers predictable on specific dates, then had associates buy the tickets. The Hot Lotto fraud netted millions before a suspicious unclaimed ticket unraveled it, and it stands as the permanent argument for why 'trust the operator' isn't a randomness policy.

And in 1999, security researchers reverse-engineered an online poker site's shuffle. The 'random' deck was seeded with the server's clock time — guessable to within a second — collapsing millions of possible decks to a few hundred thousand, checkable in real time. They could read every hidden card mid-hand. The formula looked random; the seed gave it away.

The shared lesson

All three failures have the same anatomy: the randomness looked fine and was structurally broken — a bad stir, a corrupt insider, a guessable seed. The fix is also shared: use sources whose unpredictability comes from physics (modern OS entropy, which our tools tap through your browser), and processes transparent enough that no single person can quietly bend them. Randomness you can't audit is just someone's promise.