Fair random choice guides — page 3 of 4

Duplicate entries and integer weights give the same stated odds only when every duplicate is intentional and each row is selected uniformly. Three copies of A and two copies of B create 5 equally likely rows, so A has 3/5 = 60% and B has 2/5 = 40%; a visible weight table makes that rule auditable.

Check a prize draw before selecting a winner by freezing the eligibility list, identifying duplicate or weighted entries, publishing the selection and redraw rule, and recording the first result. A list of 40 eligible entries gives each one-entry participant 1/40 = 2.5% only when every entry appears once and the stated draw selects rows uniformly.

Random lottery numbers can choose a valid combination, but they cannot improve the probability of that combination in a fair draw. If a game has 292,201,338 possible jackpot combinations, one correctly entered combination has probability 1/292,201,338, about 0.000000342%; a random generator and a personally chosen line have the same draw probability.

To generate random numbers in a range without repeats, define the inclusive range, confirm that the requested count does not exceed its size, then sample without replacement. The inclusive range 1 through 20 contains 20 integers; requesting 5 unique values leaves 15 unused values after the fifth selection and gives no value a second position.

Generate an integer for countable labels, a decimal for a value on a stated numeric interval, and a percentage when the output must be read as a share out of 100. The integer range 1–6 has 6 discrete outcomes; a decimal from 0.00 to 1.00 at 0.01 steps has 101 stated values if both endpoints are included; 37% means 37 per 100, or 0.37.

Use random words for charades by fixing a suitable category, audience, and difficulty before drawing a prompt, then remove unsuitable terms before the game begins rather than rejecting played results. A 30-word approved list gives every approved prompt probability 1/30 = 3.33% for one equal draw; a wider unreviewed dictionary cannot promise the same suitability.

Use a random question generator by separating question selection from person selection: choose from a pre-approved prompt set, state whether answering is voluntary, and give a pass or think-time rule before the prompt appears. With 24 approved questions, one equal draw gives each question probability 1/24 = 4.17%; it does not give any named person an obligation to answer.

Generate a random date or time within a valid range by stating both inclusive endpoints, the time zone, the unit being sampled, and any excluded times before drawing. From 2026-04-01 through 2026-04-10 inclusive there are 10 calendar dates, so each date has 1/10 = 10% probability in an equal date draw; a time range needs a separate precision such as whole minutes.

A UUID is a 128-bit identifier format, not a mathematical guarantee that no collision can ever occur. A version-4 UUID has 122 random bits after fixed version and variant bits, giving 2^122 possible random values; the collision risk is extremely small for ordinary volumes but it remains probabilistic, so systems still need uniqueness checks where a collision would be harmful.

Generate UUID v4 when a random 122-bit identifier with no embedded timestamp is the useful property; generate UUID v7 when sortable creation-time order is useful and exposing millisecond time is acceptable. RFC 9562 gives v7 a 48-bit Unix-epoch millisecond timestamp and 74 remaining random or monotonicity bits, while v4 uses 122 random bits after its required format bits.

A hand-flipped coin has a small, measured bias toward landing the way it started; a digital coin flip works on a completely different, purely algorithmic principle.

The expected value of a fair die is (highest face + 1) ÷ 2 — a long-run average that no single roll, weighted die, or short session can ever show directly.

NdM+K notation means roll N dice with M sides, sum them, then add K once — and 2d6 is a very different distribution from a single d12.

A coin flip embedded on a page needs exactly one uniformly chosen bit; the animation is cosmetic, but the source of that bit decides whether it is really 50/50.

Google’s and Bing’s built-in coin and dice widgets answer one quick question inside the search page, but they offer no shared screen, history, or documented method.

Statistical test suites like NIST SP 800-22 check a generator’s long output for detectable patterns; they cannot certify that any one past coin flip or roll was fair.

A true-random API draws from a physical process like atmospheric noise; a browser’s crypto API draws from an algorithm reseeded with hardware entropy — both can be fair.

The chance of an all-heads run halves with every extra coin flip while the number of possible sequences doubles — a reference table from 1 to 10 flips.

Two summed dice concentrate outcomes near the middle; a single die spreads them evenly — board game designers choose deliberately between the two shapes.

The NFL’s visiting captain calls the coin toss by rule; cricket’s visiting captain calls by tradition — and a fair coin does not guarantee a fair procedural advantage.