Quantitative results from regional climate models (RCMs) run over ice sheets are frequently used to make projections of surface melt, ice-shelf stability, and subsequently sea-level rise. However, modelled relevant mass fluxes need to be evaluated first before using future output data for projections. This study makes the case for a two-step framework when evaluating RCMs. Firstly, the reliability of the RCM when forced with reanalysis data must be assessed through comparison with historical observations. Secondly, the accuracy of using a non-observationally constrained Earth System Model as forcing must be assessed through comparison with the reanalysis forced run during the same historical period. Simulating surface melt in Antarctica with the RCM RACMO2.3p2 is given as an example. Applying this two-step procedure we show that RACMO2.3p2 respectively forced with ERA5 and CESM2 is robust for modelling contemporary and future surface melt in Antarctica. Building on this conclusion, we briefly discuss an application, i.e. three future SSP realizations of melt-over-accumulation across the Antarctic ice sheet until 2100 are presented, providing insights into the future sensitivity to meltwater ponding of major Antarctic ice shelves.