The DeepSeek Problem Is Structural Not Anomalous

DeepSeek-R1 achieving frontier-equivalent capability
at dramatically lower compute is not an outlier it is the expected
direction of progress. The history of computing is a history of efficiency
improvements that expand access to capabilities previously requiring
expensive hardware. Fixed FLOP thresholds set at the current frontier
will systematically become more permissive as the efficiency frontier
advances. This is not a calibration problem that better threshold-setting
can solve. It is an architecture problem that requires adding capability
evaluation as a second layer.