A simple model shows the logic. Suppose each trade is an even bet that you win with probability 55% and lose otherwise, risking one unit of capital each time. With 10 units of capital the chance of eventually being wiped out is (0.45 ÷ 0.55) to the power 10, about 13%. With 20 units it is about 1.8%. Same edge, smaller bets, far lower risk.
Real trading is messier: payoffs vary, results are not independent and the true win probability is unknown. The model is therefore a way to see the direction of the effect, not a forecast. What it teaches is robust: ruin depends heavily on how large each bet is relative to capital.
That is why a positive expectancy is not a licence to size aggressively. A good method traded at too large a size can still end an account during an ordinary bad run, and one that survives has simply not yet met the unlucky sequence.
In practice, risk of ruin is managed by fixing a small risk per trade, setting a drawdown limit at which size is cut, and keeping a margin for estimates that turn out too optimistic.
If you cannot say what loss would stop you trading, you have not yet measured your risk of ruin.