Is a Yes-or-No Wheel Truly Random?
A yes-or-no wheel is truly fair for a binary choice when P(Yes) = 1/2 and P(No) = 1/2 for every spin under the stated rule.
A worked example with numbers
Suppose a wheel has two equal labels, Yes and No. A fair bit has two possible values, so mapping 0 to No and 1 to Yes gives each label 1/2 = 50%. If the wheel instead has four equal sectors reading Yes, Yes, No, No, the labels are still balanced because each label owns two of four sectors. If it has three Yes sectors and one No sector, Yes has 3/4 = 75%, regardless of whether the coloured slices look similar on screen.
How to set the rule before the result
Inspect the outcome rule before spinning. Count label occurrences or sector weights, freeze the labels, and decide whether the result is one harmless prompt or a binding allocation. The yes-or-no wheel should show exactly two equal alternatives; the coin flip is an independent two-outcome check. For a browser implementation, choose the final label from a uniform integer rather than deriving it from the apparent landing position of an animation.
Common mistakes that change the odds or the process
Do not infer 50/50 from ten spins: eight Yes results in ten independent fair trials has probability 45/1024, about 4.39%, so it can occur without a biased wheel. Do not add duplicate labels as decoration. Do not click again because a result feels inconvenient; that changes the real procedure into repeated sampling. Do not confuse an equal graphic with an equal mapping if the program assigns different hidden weights to sectors.
Where this method stops being appropriate
A binary wheel can resolve a low-stakes tie between two already acceptable actions, such as which finished game to start first. It cannot decide safety, consent, healthcare, legal rights, employment, education eligibility, or a financial commitment. Equal probabilities only describe how the displayed answer is selected; they do not make the two real-world consequences equally acceptable.
How the random source fits into the rule
A browser can generate a cryptographically strong integer value with the method described by MDN’s Crypto.getRandomValues reference. That source helps make the next bit difficult to predict, but it does not prove that a page has not altered labels or saved a result. NIST’s binomial-distribution reference gives the probability model for a count of successes in independent two-outcome trials; it explains why a short observed run cannot certify a wheel.
How to inspect a 50/50 wheel setup
Start with the labels rather than the pointer. If the setup screen contains Yes and No exactly once, an equal-choice implementation has two target entries. If it contains eight decorative slices, count the labels: four Yes and four No still gives 4/8 for each label, whereas five Yes and three No gives 62.5% versus 37.5%. A wheel with unequal visible angles can still be fair if its software samples labels equally, but that rule must be stated because the picture alone cannot establish it.
Why a short spin history is weak evidence
In 16 fair spins, exactly eight Yes results have probability C(16,8)/2^16, about 19.64%. Counts of seven or nine Yes are also ordinary. A sequence such as Yes, Yes, Yes may feel informative, but its probability is 1/8 and the equally long sequence No, Yes, No has the same probability. Decide a sample size and a test before observing outputs if you are investigating software; never use a handful of hand-picked spins as certification.
Separate randomness from a decision rule
A wheel may fairly select a label and still be a poor decision process. “Yes” should name a concrete, low-risk action, and “No” should name its alternative. For example, either finished film can be played first; choosing a doctor, a contract, or a person’s access to help is not made fair by equal sectors. The valid claim is narrow: the tool selected one of two labels under the configured mapping.
What to retain after a shared spin
For a casual group choice, retain the two labels, the count of each label, the announced time when edits closed, and the first result. That record makes a later question answerable: was the output generated from the agreed 50/50 setup? It does not need private accounts or a public history. If someone finds an error in the labels before the spin, correct it; if the error is found only afterwards, follow the replacement rule agreed before the result.
Test the claim you actually need
For a home choice, count the two labels and make one result; this is usually enough. For a software review, inspect the code path from random integer to label and verify that no label is skipped. For a regulated or security-sensitive process, a wheel is the wrong evidence source: the required controls may include identity checks, a retained log, independent oversight, and a procedure set outside the page. “Random” is not a single certification grade.
A useful statement after the spin is “the wheel contained one Yes and one No and selected No once.” It reports observable facts. “The wheel is perfectly random” claims a property of all past and future operation that a single result cannot establish.
Count labels at the decision level
A wheel can use two large sectors, four small sectors, or a list behind an animation. The relevant count is how many equally likely source results map to Yes and how many map to No. Equal display angles are useful only when they match that source mapping; a checker should inspect both.