Published TREC evaluation
Official results
Explore one metric at a time. Complete result tables remain available below for verification.
Higher is better. Values are official percentages reported in the TREC 2025 proceedings paper.
Show complete official tables
Table 3
Answer quality
| Run | Precision | Recall | Completeness | Correctness |
|---|---|---|---|---|
| ABaseline | 88.31 | 32.34 | 77.63 | 66.90 |
| BSparse (narrow) | 89.04 | 35.15 | 67.70 | 64.19 |
| CDense (narrow) | 92.28 | 36.41 | 69.53 | 65.74 |
| DDense (wide) | 93.83 | 36.73 | 72.56 | 67.33 |
| ESparse (wide) | 90.99 | 35.80 | 66.77 | 64.56 |
Table 4
Citation quality
| Run | Coverage | Support rate | Contradiction rate |
|---|---|---|---|
| ABaseline | 74.87 | 97.38 | 0.44 |
| BSparse (narrow) | 95.45 | 91.12 | 0.39 |
| CDense (narrow) | 94.90 | 90.37 | 0.46 |
| DDense (wide) | 92.21 | 87.46 | 1.39 |
| ESparse (wide) | 95.83 | 87.32 | 0.72 |