A published expected value is a property of a probability model. A short-term result is one realised sample from that model. Confusing the two produces the common belief that a sequence of losses must soon be followed by a balancing win. The model does not owe any such correction on a particular spin, draw or wager.
A coin-toss illustration
Consider ten tosses of an ideal fair coin. The expected number of heads is five, yet seeing seven heads is perfectly possible. Repeating the experiment does not require the next ten tosses to have three heads. As sample size grows, proportions tend to stabilise under appropriate assumptions; that long-run tendency does not impose a fixed timetable or force the next outcome.
Expected loss and realised loss
Imagine a hypothetical negative-EV game with an expected net result of minus five cents per dollar staked. Staking $100 in total gives an expected net loss of $5. A particular session could lose the full $100 or finish ahead, depending on the available outcomes. Neither scenario invalidates the expected value. A win does not transform the game's mathematics into a positive-EV model, and a larger loss does not prove that the published average must be false.
Why variance matters
The same expected value can occur alongside frequent small deviations or rare very large deviations. High dispersion means the average can be a poor description of any one short session. When evaluating data, the sample size, outcome distribution, possible upper and lower results and any independence assumptions should be recorded. These quantities answer different questions.
Do not turn convergence into a plan
Continuing until results approach the average is not a reliable recovery method. In a negative-EV process, more wagers usually mean more turnover exposed to the disadvantage; there is no maximum waiting time after which a positive result is guaranteed. The gambler's fallacy relies on precisely this misunderstanding.
The general expected-value guide shows the calculation itself. This page explains the gap between that calculation and an observed run, including why looking only at the most recent observations is a poor basis for predicting what happens next.