The full history of ice sheet and climate interactions is recorded in the vertical profiles of geochemical tracers in polar ice sheets. Numerical simulations of these archives promise great advances both in the interpretation of these reconstructions and the validation of the models themselves. However, fundamental mathematical shortcomings of existing models subject tracers to spurious diffusion, thwarting straightforward solutions. Here, I propose a new vertical discretization for ice-sheet models that eliminates numerical diffusion entirely. Vertical motion through the model mesh is avoided by mimicking the real-world flow of ice as a thinning of underlying layers. A new layer is added to the surface at equidistant time intervals, isochronally, thus identifying each layer uniquely by its time of deposition and age. This new approach is implemented for a two-dimensional section through the summit of the Greenland ice sheet. The ability to directly compare simulations of vertical ice cores with reconstructed data is used to find optimal model parameters from a large ensemble of simulations. It is shown that because this tuning method uses information from all times included in the ice core, it constrains ice-sheet sensitivity more robustly than a realistic reproduction of the modern ice-sheet surface.