Inconsistencies between observations from long and short period seismic waves and geochemical data mean craton formation and evolution remains enigmatic. Specifically, internal layering and radial anisotropy are poorly constrained. Here, we show that these inconsistencies can be reconciled by inverting cratonic Rayleigh and Love surface wave dispersion curves for shear-wave velocity and radial anisotropy using a flexible Bayesian scheme. This approach requires no explicit vertical smoothing and only adds anisotropy to layers where required by the data. We show that all cratonic lithospheres are comprised of a positively radially anisotropic upper layer, best explained by Archean underplating, and an isotropic layer beneath, indicative of two-stage formation. Within the positively radially anisotropic upper layer, we find a variable amplitude low velocity zone within 9 of 12 cratons studied, that is well correlated with observed Mid-Lithospheric Discontinuities (MLDs). The MLD is best explained by metasomatism after craton formation.