Combinatorial optimization problems are pervasive across a range of sectors, including finance, logistics, and machine learning. When it comes to tackling large-scale problems, traditional 'von Neumann' computing architectures encounter substantial limitations. Optoelectronic Ising machines, on the other hand, leverage the dynamic evolution of physical systems to enable inherently parallel problem-solving. These machines represent a pivotal technology for hardware-accelerated large-scale combinatorial optimization. Nevertheless, a fundamental discrepancy exists between the continuous analog nature of their underlying physical carriers and the discrete binary constraints inherent in the Ising model. This disparity renders the systems vulnerable to hardware amplitude imbalances, noise buildup, and performance drift, ultimately compromising solution accuracy and constraining scalability.