We prove improved bounds for swap multicalibration, swap omniprediction, and swap agnostic learning in both the distributional and online settings.