The Verification Venue · pointed at a thing everyone gets wrong
Chance Keeps No Ledger
A roulette wheel has no memory. It does not know it just landed on black ten times, it is not saving up a red for you, and nothing about the past is owed to the future. After any losing streak, your next spin is exactly the odds it always was.
Verdict: No In a memoryless game (roulette, coins, slots, most lotteries) you are never "due." European roulette red is 18/37 = 48.65% on every single spin, before a streak and after one. American is 18/38 = 47.37%.
The feeling that a win is "owed" after a run of losses is the gambler's fallacy. Its most famous case: Monte Carlo Casino, 18 August 1913, when the ball landed on black 26 times in a row. On a fair single-zero wheel a specified color repeating 26 times is (18/37)^26 ~ 1 in 137 million (either color 26 times, 2×(18/37)^26 ~ 1 in 68.4 million). Gamblers poured francs onto red, sure it was "due," and lost millions. Yet the conditional probability of red on spin 27 was still exactly 18/37. Below, spin the wheel yourself: condition on any losing streak you like, spin it thousands of times, and watch the next-spin win rate refuse to budge.
Base rate (red, per spin)
48.65%
European wheel, 18/37
Win rate AFTER ≥ 6 losses
–
press spin
Drift vs base
–
a streak buys you nothing
The simulator finds every spin that was preceded by a run of this many reds-missed in a row, then reports how often the very next spin was red.
The single green zero (European) or two greens (American) are the house edge. They lower the base rate but do not give the wheel a memory.
Not yet spun. The dots you will see are the win rate conditioned on every streak length from 1 to 12: they all cling to the flat base line. No upward drift, ever.
Run it again and the dots reshuffle within sampling noise, but the shape never changes: a flat line at the base rate. Longer streaks are rarer, so their dots wobble more (fewer samples), but they wobble around the base rate, never above it. That is what "independent" means: the spin does not read its own history.
The "due" instinct is not stupid. It is aimed at the wrong machine.
Here is the honest part almost every explainer skips. The rule "the more it has been missing, the more it is owed" is correct arithmetic, just applied to the wrong kind of system. Whether the past matters depends on one thing: does the machine have memory?
Twist 1 (the honest exception): without replacement, you really do become due
A roulette wheel replaces every outcome: the pockets are all still there on the next spin. But a blackjack shoe, a bingo cage, a batch of scratch cards do not. Each draw removes an outcome from a finite pool, so a run of one color genuinely depletes it and the other color becomes more likely. This is the difference between a coin (Bernoulli, with replacement) and a deck (hypergeometric, without replacement). It is exactly why card counting works and why casinos shuffle.
Imagine the 37 European pockets as a shuffled deck dealt without replacement: 18 red, 18 black, 1 green. Draw blacks and watch the chance the next card is red actually climb. Compare the two machines side by side:
Blacks already drawn (b)
10
from a 37-pocket deck
Next-red, WITH replacement
48.65%
the wheel: unchanged, no memory
Next-red, WITHOUT replacement
66.67%
the deck: 18 / (37 − 10) = 18/27
The wheel line never moves. The deck line climbs, because you have literally removed black outcomes from the pool. Same instinct, and here it is right.
The rule to carry home: "being due" is a fallacy only for independent, replaced events (roulette, coin flips, slot reels, independent lottery draws). For dependent, non-replaced events (a shoe, bingo, a scratch-card batch) the same instinct is the correct hypergeometric arithmetic.
Twist 2 (the reversal of the reversal): the hot hand turned out to be partly real
For 30 years the "hot hand," the belief that a basketball player on a streak is more likely to score next, was dismissed as the gambler's fallacy's twin: a mirage the mind paints on random noise (Gilovich, Vallone & Tversky, 1985). Then Miller & Sanjurjo (2018) found a selection bias hiding in the standard test, and correcting it partly reversed the 1985 verdict.
The bias is astonishing because it lives in a fair coin. Take every possible sequence of 4 flips. In each one, look only at the flips that came right after a heads, and ask what fraction of those were also heads. You would guess 50%. Enumerated live below, the average is 40.5%, not 50%. A naive "streaks are not real" test built on this measure is rigged to understate streakiness, so when you correct it, the evidence for a real hot hand comes back.
–
This cuts the opposite way from "being due." A streak can carry real information about a skilled shooter's hidden state (fatigue, rhythm, confidence) because a person is not memoryless. It can carry zero information about a wheel, a coin, or a lottery ball, because those are. Miller & Sanjurjo does not resurrect "due" in games of pure chance. It rescues the hot hand only in skill sequences.
The complete answer: a ruler with two ends
Does the past predict the next outcome? It depends entirely on what the system is made of. Past outcomes matter when the machine has memory (a finite deck) or a hidden skill state (a hot shooter). They matter exactly zero when it has neither.
| System | Has memory? | Are you "due"? | Why |
|---|---|---|---|
| Roulette wheel | No | Never | Every pocket returns each spin (with replacement, i.i.d.) |
| Coin flip | No | Never | Independent Bernoulli trials, p fixed |
| Slot machine | No | Never | Each spin drawn fresh from the same RNG distribution |
| Independent lottery draw | No | Never | New balls, replaced each draw; past numbers carry nothing |
| Blackjack shoe | Yes | Genuinely, yes | Cards removed without replacement (hypergeometric) |
| Bingo cage / scratch-card batch | Yes | Genuinely, yes | Finite pool depletes as outcomes are drawn |
| A skilled shooter's next shot | Yes | Can be, partly | A person has a hidden state; a real hot hand exists (Miller & Sanjurjo 2018) |
The check: every number recomputed in front of you
Nothing here is a stored figure. The base rates, the 1913 odds, the without-replacement climb, and the 40.5% coin bias are all computed live in your browser from first principles:
The offline gate recomputes all of this, twice over where it can (closed form and Monte-Carlo simulation): node research/are-you-due-for-a-win-gamblers-fallacy/verify-are-you-due-for-a-win-gamblers-fallacy.mjs. Free choices & uncertainty: the streak simulator is a finite Monte-Carlo sample, so its post-streak rate lands within sampling error of the base rate, never exactly on it, and longer streaks (rarer, fewer samples) wobble more. The 1913 odds assume a fair single-zero wheel; a physically biased wheel is a real but separate phenomenon (a manufacturing defect, not a memory). The 40.5% is the size of the selection bias in 4-flip sequences, not the size of the hot-hand effect itself.
What's exactly true here, and what's a model
Exactly true (arithmetic). Independent trials have no memory: for i.i.d. Bernoulli(p) events, the conditional probability of a success after any run of failures is exactly p. European roulette red = 18/37 = 48.65%; American = 18/38 = 47.37%. The Monte Carlo 1913 odds are exact powers: (18/37)^26 for a specified color 26 times (~1 in 137 million) and 2×(18/37)^26 for either color (~1 in 68.4 million). Without replacement, the conditional next-red after b blacks drawn from a 37-pocket pool is exactly 18/(37 − b), which rises. The Miller & Sanjurjo figure is the exact rational 17/42 = 0.4048, the average heads-after-heads proportion over the 14 (of 16) four-flip sequences where it is defined.
A sample, not a proof (the simulator). The live streak simulator draws a finite number of pseudo-random spins, so the post-streak win rate is an estimate of p, correct to within Monte-Carlo error. It converges to p, it does not land on it exactly, and that is the honest picture: no drift, only noise that shrinks with more spins.
The honest firewall. Miller & Sanjurjo makes the hot hand partly real only in skill-based sequential performance (basketball shooting). It does not resurrect "being due" in games of pure chance. Roulette, coins, slots, and independent lottery draws still have exactly zero memory. And the 40.5% is the magnitude of a statistical artifact, not a claim that the hot hand is large.
Named scope. Red is 18/37 (European) versus 18/38 (American): name the wheel or the number is ambiguous. "Being due" is a fallacy for independent, replaced events only; for finite, non-replaced draws it is correct. And the 1913 numbers are for a fair wheel; wheel bias (the Jagger story) is a separate, real, physical effect.