Fair random choice guides — page 4 of 4

A chi-square goodness-of-fit test turns six raw face counts into one number that separates a genuinely loaded die from an ordinary uneven-looking run.

Excluding only the immediately previous result from a random pick does not leave every remaining item at its old 1/N share — it raises each one’s share to 1/(N−1).

A linear congruential generator produces every future value from one visible formula and a handful of past outputs — the exact weakness a modern cryptographic generator is built to remove.

A generator documented as inclusive on both ends can still silently under-produce its own endpoint values — and confirming it takes far more trials than a quick manual check.

Regulated lottery and gaming equipment is checked by an accredited outside laboratory against a published standard, not only by running a generator’s own output through a statistical test.

A machine that fully controls a coin’s launch can flip it with a fully predetermined outcome; software drawing from a cryptographic source cannot be predicted by any algorithm, AI included.

Converting a random byte into a pick among 30 teams instead of 32 introduces a small but measurable bias unless the code explicitly corrects for it.

Rounding a wheel’s slice boundaries to whole degrees can leave adjacent slices with different actual widths even though every slice looks the same size on screen.

With 10,000 possible 4-digit PINs, only about 119 people need to pick one before there is a better-than-even chance two of them match — far fewer than most people expect.

In a fair race to three heads, an exact tie happens often enough that neither side is truly a 50/50 favorite — each wins only 43.21% of the time.

NIST SP 800-90B requires two continuous, on-the-fly self-tests from every certified entropy source — not just a one-time lab review — and the cutoff numbers show exactly how sensitive each one is.

A fair coin flip carries exactly 1 bit of randomness and a fair six-sided die carries about 2.585 bits — the same information-theoretic yardstick that makes three coin flips and one eight-sided die numerically interchangeable.

Regulation casino dice are manufactured to within 0.0005 inches of a true cube with sharp, unrounded edges and density-matched pip cavities; an ordinary board-game die is tumbled during production, which rounds its edges and leaves a lighter side where its pips were drilled.

An AFL coin toss lets the winning captain choose only which end their team defends first, since both sides kick and defend at both ends across four quarters; a cricket toss lets the winning captain choose to bat or bowl first for the whole match, a far larger strategic asymmetry.

One person correctly calling 10 independent fair coin flips in a row has only about a 0.098% chance — but put roughly 710 independent people through the same challenge and there is better than even odds that at least one of them succeeds by chance alone, not skill.

A 26-letter wheel with equal slices gives every letter exactly 1/26 ≈ 3.85%; real English text uses E about 12.7% of the time and Z about 0.074% — over 170 times rarer — so the right choice depends on whether the wheel needs to be fair among letters or realistic as language.

A group of 12 people split into two labeled teams of 6 has exactly C(12,6) = 924 distinct possible rosters for Team A alone — or 462 if the two teams are otherwise interchangeable — numbers that show why a random-teams tool never runs short of genuinely different outcomes.

A hardware random number generator measures a genuinely unpredictable physical process directly; a software CSPRNG does not create randomness from nothing — it algorithmically stretches a much smaller seed of that same kind of physical entropy, commonly 256 bits, into an effectively unlimited stream of unpredictable output.

Bumping an excluded number up to its neighbor instead of re-drawing quietly doubles that neighbor’s odds; the correct fix is rejection sampling, and for a 1–100 range with 2 numbers blocked, that costs an average of about 1.02 draws to produce one valid result.

RANDOM.ORG’s HTTP API gives each client a starting quota of 1,000,000 bits, tops that up by 200,000 bits a day once the balance falls below 1,000,000, and returns an error once the quota runs negative — a real operational limit a browser’s built-in generator, which never leaves the device, simply does not have.