The Gambler Was Right: A Reproducible Challenge to Strategic Indifference in Fair Coin Sequences

Preprint. Submitted for peer review, July 2026In a memoryless process, each event is independent of every prior event; it seems to follow, as a logical entailment, that the timing of an observer’s engagement cannot affect outcome distributions. We term this conclusion strategic indifference and test it directly in computationally generated fair coin flip sequences. It fails. Two players observe the same sequences of 80 binary outcomes across 500,000 sequences (40 million flips), make the same number of 9-bet anti-continuation wagers, and face identical per-entry loss odds of (1/2)^9. They differ only in entry criterion: the selective player enters after observing T consecutive same-side outcomes (T ∈ {3, 4, 5, 6, 7}); the Markov player enters at each direction change. Under equalized comparison, the selective player experiences 10.6% to 67.7% fewer catastrophic losses, scaling monotonically with observation depth, with zero gradient reversals across 25 data points and combined Z-scores from −7.16 to −22.60 — decisive after conservative correction for within-sequence correlation. Yet per-entry loss rates confirm independence: both players match (1/2)^9 at every threshold. The effect is positional, not predictive. The mechanism is path narrowing: prior observations consume positions within the streak structure, raising the effective catastrophe threshold without altering any individual flip probability. The advantage scales inversely with sequence length, converging with Miller and Sanjurjo’s (2018) finite-sequence streak bias and demonstrating that their retrospective finding operates prospectively. The result decomposes the Gambler’s Fallacy into a per-event prediction (false) and a strategic entry prescription (true); full code is provided for independent reproduction.Code and data: https://doi.org/10.5281/zenodo.21434675