Our risk setting wasn't setting our risk. At 40%, 22–26% of entries hit the broker's 1000-lot ceiling. At 20% it barely binds. Which explains the number that made no sense. 50% risk gave a shallower drawdown than 40%: 23.72% vs 24.84%. The cap truncates exactly the positions that deepen a hole. Same book, six start dates: 2025 → 12.6% 2021 → 15.6% 2024 → 18.7% 2022 → 20.0% 2020 → 21.6% 2023 → 31.1% PF held 1.74 to 1.84. The edge was stable, the path wasn't. We'd sized against 21.6%, the second-deepest of the six. Fine, by luck. Leave-one-out said our breakout leg bought profit at 11.5% per drawdown point, against 43.2 for engine two. Obvious move: cut it. So we moved that leg alone. Up: ratio 1.14 → 1.47. Down: → 0.91. It had been charged for compounding. Not an attribution. What survived: 25/70/40. $1.45M → $1.74M PF 2.05 → 2.21 DD 19.65% → 14.62% Cost: engine two is now 56% of profit, on 342 trades and 8 days of scrutiny. Top-3 trades of engine three: 46% of its own profit. And 26 of 26 quarters in the sample are positive. That's the window, not us.