Law of Large Numbers
Flip a fair coin 1000 times. The proportion of heads won't be exactly 0.5, but it will be close, and closer still with more flips. Formally, for i.i.d. with mean :
The mechanism is variance shrinkage: . The LLN is why casinos and insurers profit reliably on thin edges, and why backtests need many independent bets before the sample mean means anything. It does not make tails due after a run of heads; there is no "law of averages". Convergence comes from dilution, not correction.