Abundance-prediction residuals as a function of S/N. Residuals [X̃]−[X], defined as the difference between the abundance predicted via linear regression from the latent features (z_M, z_α, z_C) and the true values, shown as a function of S/N. From left to right: [Fe/H],[α/Fe], and [C/Fe]. Red markers indicate the mean residuals in bins of S/N (with width 10, ranging from 15 to 85), and the error bars show the standard deviation within each bin.