Methodology
We tested 11 improvements and 10 made it worse
Trailing stops, breakeven moves, partial exits. We pre-registered a battery of eleven popular strategy “improvements” and ran them against a real, live edge. Ten degraded it. The survivor’s profit gain was plausibly noise (bootstrap p = 0.23).
Cost forensicsThe bad years were a toll booth
We decomposed a strategy’s losing years line by line. 91% of the losses were spread costs at collapsed stop geometry. In 2013, the cost-loaded breakeven win rate sat above the strategy’s lifetime win rate. Losing was arithmetic, not regime.
Case studyThe clock that lied for ten years
A profitable live bot. A broker server clock that silently never followed US daylight-saving time. 436 of 867 backtest trades executed a full hour later in the session than anyone believed.