Post-shift accepted-trajectory risk estimation on Agentic QA case study
0.9Post-shift Accepted Trajectory RiskPrompt/retrieval-only
Evaluation Results
| Method | Links | |
|---|---|---|
| Prompt/retrieval-onlyCalibration Strategy=Prompt/retrieval-only2026.06 | 0.9 | |
| Fixed trajectoryCalibration Strategy=Fixed trajectory2026.06 | 1 | |
| ToolChain-CRCCalibration Strategy=Trajectory calibration2026.06 | 1 | |
| Final-onlyCalibration Strategy=Final-answer-only2026.06 | 12.2 |