My Stop Was -3.5% on Both Configs. One Lost -13.3% Anyway. Same Coin

I was cleaning up my candidate list for ARIAUSDT — eight configs from the same GPU batch, same entry family (range bottom reclaim), same rough profit factor around 1.7-1.73. I sorted by max adverse excursion out of habit, expecting a flat line since they all share the same coin and roughly the same stop-loss parameter. Instead I got two clean groups: five configs bottomed out around -13% intra-trade, three bottomed out around -5%. Same coin. Same stop level, near enough. 2.6x difference in how bad it actually got.

My first assumption was a wider stop on the -13% group. So I checked the actual parameter.

The stop barely moved

The final_invalidation_pct — the hard exit level in the strategy config — averaged -3.51% on the tame group and -3.78% on the rough group. That's a 0.27 percentage-point difference. Nothing. The take-profit target, the DCA-rescue trigger, the average hold time — I checked all of them. Tame group averaged 38.7 minutes held, rough group 40.3 minutes. Same DCA aggressiveness, same win rate spread (both groups mixed 48-89% WR depending on other unrelated knobs). None of it lines up with a 2.6x jump in worst-case pain.

Whatever produced that -13.3% low was path-dependent — some specific sequence of candles that one config's exact parameter combination walked into and the others narrowly avoided. Not a deliberate risk trade-off. Noise that happened to land on one side of a threshold.

Why this matters more than the stop number itself

If you'd only seen the config sheet — "-3.5% stop, 1600 trades, PF 1.71, 48% WR" — you'd size your position assuming your worst realistic loss is somewhere near that -3.5%. Four of my eight ARIAUSDT configs would have handed you almost 4x that instead, and the parameter that's supposed to tell you your risk gave zero warning. The stop level is where the strategy intends to get out. Max adverse excursion is what the market actually did to the trade before that exit filled or before a DCA leg moved the effective entry. Those are two different numbers, and only one of them shows up in most people's "risk" column.

I've written before about how win rate lies when DCA-rescue is doing the work — this is the same lesson from a different angle. A single summary parameter (win rate there, stop-loss level here) tells you what the strategy is aiming for. It doesn't tell you what actually happened in the worst 5% of trades, and the gap between "aimed for" and "actually happened" is exactly the part that empties an account.

What I do with this now

I stopped trusting a stop-loss percentage as a drawdown ceiling. It's an exit instruction, not a guarantee. Before I size anything now, I check the logged max MAE across every config variant on that coin, not just the one I'm about to run — a 0.27-point difference in the stop parameter told me nothing about whether I was looking at the -5% version or the -13% version. If you're doing this with real capital, run your position size against the position size calculator using the worst MAE you've actually observed, not the stop level you configured, and check what that deeper number does to your liquidation buffer in the liquidation calculator before you assume the stop protects you the way the config sheet implies.