Probability Distribution Analysis of Roulette Outcomes in 500,000-Spin Dataset

We present a comprehensive probability distribution analysis of 500,000 consecutive spins recorded across twelve European roulette wheels in three partner venues over fourteen months. Using chi-square goodness-of-fit tests, entropy analysis, and run-length encoding, we characterize the empirical distribution of outcomes and compare it against the theoretical uniform distribution over 37 pockets. Our primary finding is that the aggregate empirical distribution does not differ significantly from the theoretical expectation at any standard significance level, confirming the mechanical integrity of the sampled wheels. However, within-session analysis reveals characteristic variance signatures that are consistent across wheels and venues, suggesting that the perceptual experience of 'hot' and 'cold' pockets is a robust feature of short-sample statistics rather than any mechanical property of the wheels.

The foundational claim of roulette mathematics — that each pocket lands with probability 1/37 on a European wheel — is mathematically necessary given the physical symmetry of the instrument. Empirical confirmation of this claim requires a dataset large enough to distinguish mechanical bias from statistical noise. We argue that 500,000 spins, distributed across twelve wheels, provides sufficient statistical power to draw meaningful conclusions at the level of individual pockets.

Data collection was conducted by trained observers using standardized recording protocols. Observers were stationed at each wheel for rotating six-hour shifts. Each recorded outcome was timestamped and linked to a specific wheel identifier. After collection, we conducted independent verification by cross-referencing a 10% random sample against the partner venues' own electronic records.

Our primary statistical analysis used a chi-square goodness-of-fit test applied to the aggregate frequency distribution of 37 pockets. The test statistic was 39.1 on 36 degrees of freedom, yielding a p-value of 0.34 — well above any conventional significance threshold. We therefore fail to reject the null hypothesis that outcomes are uniformly distributed. No individual pocket exceeded the Bonferroni-corrected significance threshold of p < 0.00135.

Secondary analysis examined the distribution within sessions of 100 spins — approximately the length of a typical recreational session. At this scale, the apparent non-uniformity is dramatic: the maximum observed frequency for a single number in a 100-spin window was 12 occurrences (expected value 2.70), and windows with zero occurrences of specific numbers were common. This finding quantifies the gap between the mathematical expectation and the experienced reality for recreational players, and provides a statistical basis for the pervasive belief in 'hot' and 'cold' numbers.

We also analyzed run lengths — consecutive occurrences of the same outcome category (e.g., red/black, odd/even). The empirical run-length distribution matched the theoretical geometric distribution closely, with no evidence of clustering beyond chance. The longest observed run in our dataset was 26 consecutive red outcomes, a result consistent with the expected frequency of such events in 500,000 spins.

The practical implications of this study are twofold. For players, the absence of detectable bias across twelve wheels confirms that outcome prediction based on observed frequencies is not a viable strategy at any resolution of observation available to a recreational player. For venue operators, our methodology provides a replicable audit framework for mechanical integrity — a useful tool for the regular inspection programs that modern casinos employ.