The flaw in the casino game betting strategy

The question

Suppose a casino game gives a 50 percent chance of winning and a 50 percent chance of losing on each round. You start by betting one unit, and every time you lose you double the next bet. Since the first win recovers all previous losses and adds one unit of profit, it is tempting to conclude that unlimited capital turns the strategy into a guaranteed gain. If each individual round has expectation zero, where exactly does the reasoning fail?

Where the intuition goes wrong

The seductive step is the idea that because every game is fair, the whole strategy should also be fair while still somehow guaranteeing a positive result. That combination is already a warning sign. The deeper issue is that the strategy implicitly creates an infinite random sum of gains and losses, and infinite sums are not automatically governed by the same expectation rules that work perfectly in finite settings.

In particular, linearity of expectation is safe for finitely many random variables, and for certain infinite settings when convergence behaves well. Here, however, the stake sizes keep growing without bound, and the relevant infinite series is not absolutely convergent. The rare cases in which you lose many times in a row become astronomically expensive, and that tail risk cannot be ignored just because it is unlikely.

The real flaw

The strategy only appears to convert a fair game into a guaranteed profit because it hides its cost inside an unbounded bankroll. Once you allow arbitrarily large losses before the eventual win, the expected-value reasoning is no longer as simple as adding up a neat collection of zero expectation bets. The process relies on an infinite escalation, and that destroys the clean conclusion.

Put differently, the statement "you eventually win one unit" is true only if you are allowed to survive an arbitrarily long losing streak. The mathematical price of that assumption is exactly what breaks the naive expectation argument.