Reasoning Faithfulness on DE driving data (held-out)
61.4Overall ConsistencyPinocchio
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PinocchioModel=Ours2026.07 | 61.4 | 69.2 | 86.4 | 97.3 | 82.5 | 71.1 | |
| VLM-JudgeModel=VLM-Judge2026.07 | 57.5 | 76.9 | 89.9 | 97.4 | 78.3 | 64.8 | |
| ADE-SwapModel=ADE-Swap2026.07 | 43.9 | 55.6 | 75.5 | 95.6 | 73.4 | 58.3 | |
| ADEModel=ADE2026.07 | 43.4 | 51.7 | 72.7 | 95.5 | 73.6 | 59.1 | |
| SFTModel=SFT2026.07 | 27.7 | 48 | 68.5 | 95.1 | 60.1 | 43.2 | |
| ADE-ReasonModel=ADE-Reason2026.07 | 25 | 45.7 | 67.5 | 95.2 | 51.6 | 38 |