Process Reward Models (PRMs) trained using Single-Pass Annotation with Reference-Guided Evaluation (SPARE) methodology proposed in our AAAI-2026 paper