Probabilistic calibration assessment of GalaxiesML (top) and TransferZ (bottom) predicted pdfs by BNNs after applying split conformal prediction. The probability integral transform (PIT) is a qualitative assessment of pdfs with good calibration indicated by a uniform distribution U[0,1]. Most GalaxiesML PIT distributions are uniform with BNN-2 showing a U-shaped distribution, an indication of underdispersed pdfs. On the other hand, the TransferZ PIT distributions are right-skewed, indicating the presence of a systematic bias. All PIT distributions have spikes at the edges caused by outliers; the height of these spikes is model and data dependent.