High-entropy alloy (HEA) catalysts are commonly modeled as homogeneously mixed surfaces, although thermodynamic driving forces can produce surface segregation and chemical ordering. Here, we test the validity and practical limits of this approximation by comparing random homogeneous slabs with surfaces annealed using cluster-expansion Monte Carlo (CEMC) against three published high-throughput composition-activity datasets for the oxygen reduction and hydrogen evolution reactions. Monte Carlo-based uncertainty quantification simulations propagated uncertainties in catalyst composition and predicted binding energies through 10,000 trials. CEMC generally improved agreement with experiment, but the advantage is not robust, and neither model reliably reproduces experimental composition-activity map. Structural analysis revealed surface segregation, rather than short-range ordering, is the primary origin of the difference between the models. Nevertheless, the element-conditioned binding-energy ranges of the two models largely overlapped, supporting the use of homogeneous slabs for rapid screening of elemental combinations.