Spin ices, frustrated magnetic materials analogous to common water ice, are exemplars of high frustration in three dimensions. Recent experimental studies of the low-temperature properties of the paradigmatic Dy$_2$Ti$_2$O$_7$ spin ice material, in particular whether the predicted transition to long-range order occurs, raise questions as per the currently accepted microscopic model of this system. In this work, we combine Monte Carlo simulations and mean-field theory calculations to analyze data from magnetization, elastic neutron scattering and specific heat measurements on Dy$_2$Ti$_2$O$_7$. We also reconsider the possible importance of the nuclear specific heat, $C_{rm nuc}$, in Dy$_2$Ti$_2$O$_7$. We find that $C_{rm nuc}$ is not entirely negligible below a temperature $sim 0.5$ K and must be taken into account in a quantitative analysis of the calorimetric data of this compound below that temperature. We find that small effective exchange interactions compete with the magnetostatic dipolar interaction responsible for the main spin ice phenomenology. This causes an unexpected refrustration of the long-range order that would be expected from the incompletely self-screened dipolar interaction and which positions the material at the boundary between two competing classical long-range ordered ground states. This allows for the manifestation of new physical low-temperature phenomena in Dy$_2$Ti$_2$O$_7$, as exposed by recent specific heat measurements. We show that among the four most likely causes for the observed upturn of the specific heat at low temperature -- an exchange-induced transition to long-range order, quantum non-Ising (transverse) terms in the effective spin Hamiltonian, the nuclear hyperfine contribution and random disorder -- only the last appears to be reasonably able to explain the calorimetric data.