Monte Carlo on trade lists — which drawdown percentile do you size for?

After reshuffling a backtest trade list 10k times, median max DD is often ugly.

Question: which percentile do you size to — 90th, 95th, 99th — and why?
I use QuantMogul MC reports sometimes, sometimes a 40-line Python script. Same math either way. Looking for sizing rules tied to account psychology, not software brands.