Reactivating Test-Time Scaling for Plane Geometry Problem Solving

Multi-Trace Synthesis (MTS) improves test-time scaling for plane geometry by converting rigid symbolic programs into heterogeneous reasoning traces, including executable Python scripts and CoT-augmented variants. Perception-Augmented (PA) training further enables model performance by parsing diagrams into structured semantic clauses to improve visual grounding before symbolic deduction.