- dataset: id: rdesai2/swe-marathon task_id: swe_marathon_v1_1 value: 45.0 source: url: https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/swe-marathon-v1.1 name: VeriLoop E2 evaluation evidence user: tsinghua-sigs-robot-lab notes: "20 tasks × 8 trials; 72/160 passing trials; 45.0% frozen release score."