Right Decisions from Wrong Predictions: A Mechanism Design Alternative to Individual Calibration