Profile Reconstruction on PAL-Bench 50 users 1.0
60.57OFRPAL-TRACE
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PAL-TRACEPrimary Backbone=GPT-5.42026.06 | 60.57 | 47.92 | 26.84 | 83.84 | 37.64 | 32.36 | 7.44 | |
| Compute-scaled long-contextScaling=Compute-scaled2026.06 | 48.32 | 7.53 | 3.98 | 98.61 | 20.14 | 13.87 | 7.04 | |
| Text-only LLM extractorModality=Text-only2026.06 | 40.41 | 0 | 0 | — | 10.88 | 1.61 | 1 | |
| Long-context multimodal LLMContext=Long-context, Modality=Multimodal2026.06 | 40.09 | 23.36 | 16.17 | 84.46 | 19.68 | 10.75 | 1 | |
| Adapted prior/lifelog baselineType=Baseline2026.06 | 39.21 | 39 | 20.98 | 83.28 | 27.41 | 15.28 | 1 | |
| No-LLM evidence heuristicType=Heuristic2026.06 | 37.99 | 40.47 | 17.56 | 57.49 | 23.06 | 37.49 | 0 | |
| Multimodal RAGType=RAG, Modality=Multimodal2026.06 | 35.8 | 29.42 | 15.73 | 91.75 | 22.13 | 11.16 | 1 | |
| Generic tool-use agentType=Agent2026.06 | 25.6 | 33.21 | 14.05 | 83.52 | 20.39 | 16.28 | 2.1 | |
| Generic tool-use agentMode=7-call diag.2026.06 | 23.53 | 15.09 | 6.6 | 90.62 | 17.13 | 18.58 | 7.04 |