Question Answering on LME-S
73.7Overall AccuracyC6-draw (dual-trace)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| C6-draw (dual-trace)Condition=C6-draw (dual-trace)2026.04 | 73.7 | 75 | 50 | 80 | 65 | 100 | |
| C7-control (fact-only)Condition=C7-control (fact-only)2026.04 | 53.5 | 75 | 20 | 55 | 25 | 95 | |
| C5-FS (+index)Condition=C5-FS (+index)2026.04 | 48.5 | 60 | 20 | 45 | 25 | 95 | |
| C3 (baseline)Condition=C3 (baseline)2026.04 | 47.5 | 55 | 5 | 55 | 30 | 95 | |
| C4 (evidence-scored)Condition=C4 (evidence-scored)2026.04 | 47.5 | 60 | 10 | 45 | 30 | 95 | |
| C1 (broken dual-trace)Condition=C1 (broken dual-trace)2026.04 | 40.4 | 50 | 0 | 35 | 20 | 100 | |
| C2 (fact-only)Condition=C2 (fact-only)2026.04 | 38.4 | 40 | 15 | 30 | 20 | 85 | |
| Vanilla LettaCondition=Vanilla Letta2026.04 | 20 | 0 | 0 | 0 | 0 | 100 |