Why Random Feels Rigged: Streaks and Repeats, Measured

The short answer

Genuine randomness repeats itself far more than people expect. On a 25-name wheel with repeats allowed, the chance that some name has come up twice passes 50% at pick 7, and in 100 flips of a fair coin 97.2% of sessions contain a streak of 5 or more.

The same pattern shows up wherever a wheel is used:

• On a 25-name wheel, the chance that the next spin repeats the last one is only 4.0%, yet a session of 50 spins contains at least one back-to-back repeat 86.5% of the time.
• In 100 flips of a fair coin, the longest run of the same side averages 6.98 flips.
• In 30 spins of a 6-slice wheel, each result's fair share is 5, but the most frequent result averages 7.98. A perfectly even 5-5-5-5-5-5 happens in about 1 session in 2,500.

The figures come from two sources: spins simulated through random.js, the file that picks every result on Wheel Cue, and exact math. The two were compared at 90 points, and all 90 agreed within four standard errors. The percentages quoted here are the exact values unless a sentence says "simulated".

A fair wheel produces all of this as a matter of course. If repeats are not acceptable, take winners off the wheel; if they are, say so before the first spin.

Back-to-back repeats are routine

On a wheel with k equal slices, the next spin matches the previous one with probability 1/k. That sounds small for a big wheel, but a session gives it many chances: the chance of at least one back-to-back repeat in s spins is 1 - (1 - 1/k)^(s - 1), since s spins have s - 1 neighboring pairs.

A teacher spinning a 25-name wheel ten times will see the same name twice in a row in almost a third of lessons (30.7%). A 6-slice game wheel spun ten times does it in 80.6% of sessions. On a coin, perfect alternation over 10 flips (HTHTHTHTHT or THTHTHTHTH) is the oddity: 0.20%, or 1 session in 512.

The wheel has no memory

A common reaction to a repeat is to expect the next spin to "correct" it. It does not. In the simulation, after a coin had come up the same way 3 or more times in a row, the next flip matched 50.01% of the time across 3,638,093 simulated cases (one long streak supplies several overlapping cases). On a 6-slice wheel, after 2 or more in a row, the next simulated spin matched 16.73% of the time, against an exact 1 in 6 (16.67%).

Expecting the other side to be "due" after a run is known as the gambler's fallacy. The fairness report tests the same property at scale: it draws pairs of consecutive results on a 6-slice wheel and checks that all 36 combinations turn up equally often.

Streaks run longer than intuition allows

A run of five or six identical results can look like a malfunction. In 100 flips of a fair coin it is the normal case, and the average longest run is close to 7. The yes or no wheel has two slices by default, so these are its numbers too.

The numbers behind that shape:

• The most likely longest streak is 6, at 26.4% of sessions. The average is 6.98.
• 54.2% of sessions have a streak of 7 or more.
• 92.8% of sessions have a longest streak between 5 and 10. Only 2.8% stay at 4 or less, and 4.4% reach 11 or more.
• Even in a short game of 20 flips, 76.8% of sessions contain a streak of 4 or more.

More slices make streaks shorter, but they still turn up more often than most people expect. Spin a 6-slice wheel 50 times and 69.4% of sessions contain the same result at least 3 times in a row; 16.8% contain 4 or more in a row.

Switching is the other half of the story

A streak ends with a switch. In 100 flips there are 99 places where the result can switch from heads to tails or back, and on average it does so at half of them: 49.5 times. 93.0% of genuinely random sequences have between 41 and 58 switches, and only 2.2% have 60 or more.

That is where hand-made "random" sequences give themselves away. When people try to write down random coin flips, they tend to switch too often and avoid long runs. Try it below.

"You already picked Sam!"

A teacher who spins a 25-name wheel with repeats allowed is more likely than not to hear this by pick 7, for the same reason that a room of just 23 people more likely than not contains a shared birthday. The question is not whether one particular student comes up again, but whether any of the 25 does, and the number of possible pairs grows with every pick.

With 25 names, the chance that some name has come up twice is 47.8% after 6 picks and 60.3% after 7. By 10 picks it is 87.6%. That is far higher than the 30.7% chance of the same name twice in a row from the table above: a repeat anywhere in the lesson has many more ways to happen than a repeat on consecutive spins. By 15 picks it is 99.54%, and on average the first repeat arrives at pick 6.95.

Meanwhile, the chance that one particular student has been picked twice is only 5.8% after 10 picks, and 26.4% after 25. That gap is the whole illusion: each repeat is unlikely for any given name, so it feels like a coincidence, but by 13 picks some repeat is 97.83% likely.

Class size changes the timing less than you might think. Some repeat passes 50% at pick 5 with 10 names, pick 6 with 20 and pick 7 with 30.

For what this means over a whole week of calls, see how many spins it takes until everyone gets a turn.

Long waits and lopsided counts

The flip side of repeats is the student who never seems to get picked. On a wheel with k slices, the average wait before a given entry comes up again is exactly k spins. The typical wait is shorter than that, and the long waits are longer than people expect.

On a 25-name wheel, about half of all waits end within 17 spins, but 13.0% last more than 50 spins and 4.7% more than 75. Here a wait counts the spins until the entry reappears, so a back-to-back repeat is a wait of 1. The same odds apply from any moment, not just from the last pick: a student who has waited 30 spins is no more likely to come up next than anyone else. As the wheel grows, the three right-hand shares climb toward 36.8%, 13.5% and 5.0%.

Even counts are the exception

Totals are lumpy too. Spin a 6-slice wheel 30 times and the fair share for each result is 5, but the result that comes up most often lands on 7 in 31.0% of sessions and on 8 in 30.7%.

In 29.4% of sessions at least one result comes up 9 or more times. At the other end, the least frequent result averages 2.37, and 53.0% of sessions have some result at 2 or fewer. A perfectly even split happens 0.040% of the time by exact math, in about 1 session in 2,500. If a wheel produced even counts every time, that would be the suspicious result.

Seeing a pattern in a clump like this is sometimes called the clustering illusion: random data is full of clumps, and the eye reads them as a cause.

Why fair output looks rigged

Most people judge randomness by how a sequence looks, and their picture of "random" tends to be too smooth: frequent switches, no long runs, every option getting its turn. Real randomness is independent from one spin to the next, and independence produces exactly the repeats, streaks and clumps measured above.

The gap is real enough that a music service built a feature around it. In "Shuffle: Making Random Feel More Human" (Spotify Engineering, 13 November 2025), Ludvig Borgne and Chidem Sahiner described listeners complaining that shuffle felt repetitive, even though, by the post's account, it was genuinely random. Spotify introduced a "Fewer Repeats" mode that generated several random orders, scored each one by how recently its tracks had been played, and played the freshest. As the post put it, "statistical randomness doesn't always translate into perceived randomness." The post also said Fewer Repeats had become the default for Premium users.

That trade is reasonable for a playlist. It is the wrong one for a classroom pick or a prize draw, where the promise is equal odds on every spin. A wheel that quietly steers away from recent winners is no longer giving equal odds, and no one watching can tell. When repeats are a problem, the honest fix is a rule everyone can see.

What to do before someone says "rigged"

The numbers above suggest a simple habit: decide how repeats are handled before the first spin, and say it out loud. Each wheel tool has an option that takes the winner off after each spin; its name and its default depend on the tool.

In a classroom

• Pick the rule first. If every student should get one turn, use the random student picker, where Take each student off the wheel once picked is on by default. On the name picker, turn on Remove the winner after each spin. Reset puts back every name taken off since the list was last typed in or loaded, so a round that starts and ends in one open tab comes back whole; editing or loading the list starts a fresh round. The guide to selection without replacement explains what removal changes, and how many spins it takes until everyone gets a turn covers rounds that run across several lessons.
• If repeats are allowed, say so, with the number. A line to read out before the first spin: "With 25 names and repeats allowed, someone will probably be picked twice by spin 7. That is the wheel working." The exact figure is 60.3%, and it turns "you already picked Sam" into a prediction that came true.
• Expect long waits without removal. On a 25-name wheel, 13.0% of waits last more than 50 spins. If every student needs a turn, removal is the tool, not more spinning.

Before a giveaway draw

• State the rules before you spin. Say whether one person can win twice, how many prizes there are, and that the first result stands. The giveaway winner picker has Remove each winner before the next draw on by default, so a name entered once cannot win twice, and the raffle wheel has Retire each ticket once it wins, so no ticket wins twice. Someone holding several raffle tickets can still win more than one prize, so say before the draw whether that is allowed.
• If someone may win more than once, read out the odds. For example: "Every spin is a fresh draw, so the same name can win twice in a row. With 6 names on the wheel, any spin has a 1 in 6 chance of matching the one before."
• Use weights, not duplicate lines. Write Sam *3 with Weighted choices on, so the result window shows the probability each winner actually had and removing the winner takes all of Sam's tickets off at once (weighted picking).
• Let everyone see the spin. On a projector or shared screen, Fullscreen fills it with the wheel, and the result window opens over it as each result lands.
• Publish the whole log. Copy in the results log puts every spin on your clipboard with its number, result, probability, spin ID and time; the panel itself shows only the latest 40. The log lives only in the open page, so copy it before you close the tab.
• Never re-spin a result that looks wrong. A repeat or an early streak is ordinary. Spinning again because a result looked suspicious is what actually biases a draw.

The fairness report finds no sign of bias in the code across millions of draws. It cannot show that a host did not re-spin or edit the list, and clearing the results log starts the spin numbers again at 1, so a log alone does not prove that no draws were dropped. That is why the rules come first, the draw happens on a live stream or a recording, and the log is published in full. The giveaway guide has a full checklist for a public draw.

When someone asks anyway

Point to the strip under the wheel: with equal entries and repeats allowed, it shows every name at the same 1 in n, before and after a repeat.

For the spin itself, open How this result was picked before you close the result window, and take a screenshot or keep a screen recording. It shows the raw random number, the rejection limit, how many draws were discarded and the arithmetic that turned the number into a slice, so anyone can recheck that spin by hand. The copied log has the spin ID and probability but not the raw number, so this is the only moment the working is visible.

How to tell rigged from random

One short session rarely shows that a wheel is biased, because the patterns that look suspicious are ones fair randomness produces regularly. A streak of 10 or more in 100 flips happens in 8.7% of sessions, a repeated name within 7 picks of 25 in 60.3%, and one result coming up 9 or more times in 30 spins of a 6-slice wheel in 29.4%.

Evidence of bias looks different: one result coming up more often than its share over thousands of spins, or one result making another more likely next. Those are the things to measure, and they are measured:

• The fairness report runs millions of draws through the same random.js and tests equal slices, weights, shuffles and consecutive results.
• How randomness works explains the method step by step and lets you run a fairness test in your own browser.
• The simulation script and raw results for this article are linked at the top of the page. Save the script and random.js in one folder and run node why-random-feels-rigged.mjs random.js results.json to re-run the numbers with Node.js. The simulated values will differ in the last decimals; the exact ones will not.