Iteration 57 — the DRO Cousins Both Pass EXAM PASS ×2 · GROUPED CENSUS NEXT

In plain words: after the map-drawing school went zero-for-three, the queue turned to a completely different question: not "what drives what," but "how much worse is the worst regime than the average?" Two cousin instruments ask it two ways — one over our named market regimes (the worst-regime gap), one over the worst FRACTION of moments regardless of labels (the tail-risk excess, with a tunable tail size). Their joint exam was spotless. Handed the model that is perfect by construction, both read exactly zero — no worst regime, no bad tail, precisely as the ground truth demands. Handed a deliberately wrong model, both lit up: the worst regime was 2.4 loss-units worse than average, and the worst 10% of moments carried nearly 8 units of excess pain. The tail-size dial passed its built-in mathematics check (deeper tail must read worse — it did, 7.9 → 2.7 → 0.9). Next: the field test as a pair, where the named question is whether the two cousins are secretly one instrument.