Nuclear metamodels - phenomenological parametrizations of the energy of nuclear matter - are convenient tools to explore the space of realistic neutron star configurations constrained by astrophysical and nuclear data. While much recent work has focused on composition-agnostic barotropic models, the metamodel approach is designed to describe the composition dependence of the relevant thermodynamic potential. We revise a previously proposed non-relativistic metamodel by introducing a more controlled high-density behavior, improving both its causal properties and its accuracy in reproducing the pressure and the $\beta$-equilibrium composition of microscopically motivated equations of state. Since asymptotic causality is enforced by construction, the fraction of discarded models due to superluminal sound speeds is substantially reduced, facilitating metamodel-based explorations of equilibrium neutron star configurations. We further assess our framework by performing a Bayesian inference of neutron star properties beyond standard observables such as masses and radii, exploiting the metamodel’s ability to probe composition-dependent quantities including the dUrca threshold and the Ledoux criterion for g-mode stability.