The precession and obliquity frequencies of the Earth’s rotational motion are functions of the dynamic ellipticity of the Earth’s gravitational figure, and this connection has provided a novel bridge between studies of palaeoclimate and geodynamics. In particular, analyses of tuned climate proxy records have yielded bounds on the mean relative perturbation in dynamic ellipticity over both the last 3 Myr and 25 Myr that are less than ∼3 per cent of the non-hydrostatic component of the ellipticity. We demonstrate that this apparent consistency actually defines an important geophysical enigma. Over the last 3 Myr, changes in the Earth’s figure are likely dominated by ice age forcings—in this case, a small perturbation to dynamic ellipticity implies significant isostatic compensation of the ice-ocean surface mass loads and, hence, a relatively low mantle viscosity. In contrast, over the last 25 Myr, changes in the Earth’s long-wavelength gravitational form are likely dominated by mantle convective flow, and in this case, the small perturbation to dynamic ellipticity implies sluggish convection and a relatively high mantle viscosity. There are at least four possible routes to resolving this enigma: The viscosity in the Earth’s mantle is transient (i.e. dependent on the timescale of the applied forcing), tidal dissipation changed in a manner between the last 3 Myr and 25 Myr that was sufficient to resolve the issue, the observationally inferred bounds are unrealistically restrictive, or earth models exist in which the ice age and convection effects approximately cancel leading to no net perturbation. In this paper, we compute a suite of numerical predictions of ice age and convection-induced perturbations to the dynamic ellipticity to illustrate the enigma described above.