Fig. 2: Model-data evaluation.

Each model simulation is assessed for consistency with the available geologic data across marine sectors and terrestrial constraints (see Supplementary Table S1 for specific geologic criteria). ASE Amundsen Sea Embayment, RSE Ross Sea Embayment, WSE Weddell Sea Embayment, WSB Wilkes Subglacial Basin, PB Prydz Bay, ASB Aurora Subglacial Basin, A1B ANDRILL-1B provenance, Nun. interior nunataks, PH Pirrit Hills. Cell color reflects the evaluation of model-data consistency, and cell width indicates the confidence of this evaluation; a ‘least confident’ classification may result from the equivocal nature of geologic evidence, or inherent difficulties with comparing model results with that particular kind of geologic data (e.g., spatial resolution). Model scores are calculated by multiplying the model-data agreement score for each criterion (1, 0, or −1) by the confidence weight (10, 5, or 1), summed across columns. Model member naming convention reflects the parameter combination (ocean temperature sensitivity OC 2,3,4—hydrofracture parameterization HF off,low,medlo,medhi,max—climate matrix scaling approach area,vol). Total sea-level amplitude “SL ampl.” reports the largest difference in sea-level equivalent (m SLE) between maximum and minimum ice sheet configurations, while “Max masl” reports the maximum sea-level equivalent contribution above present. Geologic data from Wilkes Subglacial Basin is split into two categories: datasets constraining glacial advance and retreat across the continental shelf, versus datasets constraining the inland extent of grounding line retreat. Asterisks denote the two best-fit model runs, identified by weighting simulations based on model-data comparison confidence.