Probabilistic calibration assessment of GalaxiesML (top) and TransferZ (bottom) predicted pdfs by deterministic NNs after creating prediction intervals using split conformal prediction. The probability integral transform (PIT) is a qualitative assessment of pdfs, with good calibration indicated by a uniform distribution of U[0, 1]. The GalaxiesML PIT distributions are center-peaked, indicating overdispersed pdfs, while the PIT distributions are U-shaped, indicating underdispersed pdfs. The presence of peaks at the edges indicates the presence of outliers.