This is the deepest idea in the group and the least known, and it can be shown with one bet.

You start with $100. A coin is flipped. Heads, your money increases by 50 percent. Tails, it decreases by 40 percent. You may play as often as you like.

The expected value is positive. Half the time you multiply by 1.5, half by 0.6, so the average outcome of one flip is 1.05 times your money — a 5 percent gain per flip. Standard analysis says play, and play as much as possible.

Now play a hundred times in sequence. Almost everyone goes broke.

Why both are true

They are computing different things.

The ensemble average runs the bet across many people simultaneously and averages the results. That number is 1.05 per flip, and it is correct: if a thousand people each play once, the total pot grows.

The time average is what happens to one person playing repeatedly. That is a product of the outcomes, not a sum. A heads and a tails, in either order, gives 1.5 × 0.6 = 0.9. You have lost 10 percent on an even split, and with enough flips an even split is what you get.

When these two averages differ, the process is non-ergodic, and the ensemble average is not a guide to any individual's experience.

Why this matters more than it sounds

Most consequential decisions are non-ergodic, because they are sequential and because there is a floor you cannot come back from.

You cannot rerun. The average across possible versions of your career is not available to you. You get one path, and the order of events matters because each one changes the resources available to the next.

Ruin is absorbing. A path that hits zero stops. Any calculation that averages over outcomes including "goes bankrupt and then recovers" is averaging over a world that does not exist. This is the mathematical form of never risk what you cannot afford to lose, and it explains why that advice survives being called conservative.

Multiplicative processes punish variance. Where outcomes multiply rather than add — which is how wealth, reputation and most compounding does behave — volatility itself reduces the time-average return, even with an unchanged expected value. A steadier path with the same average beats a volatile one, and the gap is not small.

What follows for decisions

Ask what happens if you do this repeatedly, not what the average outcome is. The sequence, not the distribution.

Size the bet so that the bad case leaves you able to continue. This is why professionals size positions rather than picking them, and it is the entire content of risk management once ruin is on the table.

Treat survival as a precondition rather than a preference. Compounding requires staying in the game, and an interruption does not merely pause it — the path is the outcome.

The reason this is the sharpest of these ideas is that it is not a bias or a heuristic. It is an arithmetic fact about which average applies, and most of standard decision analysis quietly uses the wrong one.