Gold-evidence retention on LongMemEval-S (test)
99.3Gold-evidence Retentionlearned_V
Evaluation results
| Method | Links | |
|---|---|---|
| learned_VRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 99.3 | |
| goal_onlyRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 97.9 | |
| uniform_VRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 93.9 | |
| learned_VRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 77 | |
| uniform_VRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 65.7 | |
| self_onlyRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 51.8 | |
| self_onlyRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 51.8 | |
| reliability_onlyRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 49.7 | |
| reliability_onlyRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 49.7 | |
| recency_onlyRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 36.8 | |
| randomRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 30 | |
| goal_onlyRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 28.6 | |
| emotion_onlyRegime=Oracle (peeks at Q), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 20.1 | |
| emotion_onlyRegime=Blind (realistic), Keep ratio=30%, Splits=20 resampled, Cases per split=2402026.06 | 20.1 |