✦ THE OBSERVATION DECK
Fooled by Randomness.
Many wheels. Many chances. Watch extraordinary luck happen together.
Fair wheels. Unfair conclusions.
Extraordinary luck.
Ordinary people.
Players who have won at least 5 times in a row. We remember their best streak, not just their latest spin.
Experiment settings Wheel wear, pro players & evidence
A little luck. Or a wonky wheel?
FAIR MODE · ALL 37 POCKETS EQUALLY LIKELY
Off by default. Switching modes resets the run with fresh wheels. In wear mode, each table gradually favors one fixed, randomly chosen pocket.
Casual players guess red or black. Pros can study past outcomes below. A biased number is not automatically a winning strategy—and the planets still do nothing.
Real wonky wheels? The evidence & our model
The García-Pelayo family, Spain, 1990s. They recorded roulette outcomes and exploited unequal number frequencies. The Guardian (2004) ↗ describes imbalance, uneven floors, and irregular pocket sizes. El País (2012) ↗ reports manufacturing and leveling defects.
Hibbs & Walford, Reno, 1947. Adam Kucharski’s The Perfect Bet (2016), chapter excerpt ↗ recounts their search for biased wheels and explains that wheels can gradually wear with use. It does not establish a measured wear curve for their particular wheels.
Our accelerated toy model: each pocket starts with 200 probability tickets. Over 200 spins, one fixed pocket per table gains one extra ticket per spin, capped at 400. Its chance rises from 1/37 (2.70%) to 2/38 (5.26%); the other 36 each fall to 1/38. Zero can also be favored. These sources document physical bias, not this wear curve or rate. This is not a calibrated model of real wheel aging.
Look at the evidence, not just the streak. Click a table to compare its model probabilities with all its recorded pocket counts. Finite samples fluctuate; a hot number alone does not prove bias. Modern maintenance can correct defects; these wheels intentionally never get serviced.
THE EVIDENCE CLUB
Lucky guesses vs. learned edges.
Two of every ten players study their own table’s past spins. They cannot see the hidden bias. Switching this setting resets the run.
Watching. Counting. Not consulting Jupiter.
How pros learn (and sometimes fool themselves)
Until a signal emerges, pros bet red/black like everyone else. Every 100 observations, they inspect the most frequent number at their own table. One starts after 400 spins, the more cautious one after 800. They require a frequency above the 1/36 break-even rate and a standardized excess over fair-wheel frequency of at least 3.5 or 4, respectively. A qualifying signal means a one-chip straight-up bet on that number until reassessment; otherwise they return to red/black. No future spins, hidden target, or astrology enters their decision.
This is an illustrative learning heuristic, not a calibrated statistical test: scanning 37 numbers repeatedly can produce false alarms. A real edge can be missed, a wrong number can look convincing, and even a good bet can lose. Straight-up wins pay +35 chips net; losses cost 1. Fewer wins can still mean more profit. The scoreboard compares chips per 100 one-chip bets, including each pro’s learning period, not just their successful bets.
Use 200× speed to collect thousands of observations. With wear on, the fully favored number’s true expected net is about +0.89 chips per straight-up bet; a fair wheel remains −0.027 chips per such bet. These model expectations are not guaranteed realized profits. Table details disclose number-bet results and a separate hidden-truth audit.
The maths Why more chances mean more “miracles”
A 7-win streak has a 0.64% chance in a fixed 7-spin window for one player. Give chance enough chances.