Multi-agent social interaction evaluation on KAIROS n = 3000 (test)
72.4AccuracyRAW (independent)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RAW (independent)Condition=independent, Model=gpt-4o, Temperature=0.7, Mode=reflection mode2026.04 | 72.4 | — | — | — | |
| PBRC (token-empty gate)Condition=token-empty enforcement mode, Evidence tokens=none, Model=gpt-4o, Temperature=0.7, Mode=reflection mode2026.04 | 72.4 | 0 | 0 | 0 | |
| SOCIAL_REFLECTED (peers+reflection)Condition=peers+reflection, Model=gpt-4o, Temperature=0.7, Mode=reflection mode2026.04 | 69.8 | 697 | 327 | 249 | |
| SOCIAL (peers)Condition=peers, Model=gpt-4o, Temperature=0.7, Mode=reflection mode2026.04 | 67.8 | 838 | 411 | 273 |