Information flow management in multi-party professional scenarios on ConfAIde Tier 4 (test)
95MS-E Scorebase
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| baseBackbone=Mistral-7B, Zero-shot transfer=true2026.02 | 95 | 95 | |
| baseBackbone=Llama-3B, Zero-shot transfer=true2026.02 | 85 | 100 | |
| PrivAct (Refiner)Backbone=Mistral-7B, Zero-shot transfer=true2026.02 | 80 | 90 | |
| baseBackbone=Llama-8B, Zero-shot transfer=true2026.02 | 75 | 95 | |
| PrivAct (Refiner)Backbone=Llama-3B, Zero-shot transfer=true2026.02 | 75 | 100 | |
| PrivAct (Verifier)Backbone=Llama-3B, Zero-shot transfer=true2026.02 | 75 | 100 | |
| PrivAct (Refiner)Backbone=Llama-8B, Zero-shot transfer=true2026.02 | 70 | 100 | |
| PrivAct (Verifier)Backbone=Llama-8B, Zero-shot transfer=true2026.02 | 70 | 88 | |
| PrivAct (Verifier)Backbone=Mistral-7B, Zero-shot transfer=true2026.02 | 68 | 100 | |
| baseBackbone=Qwen-4B, Zero-shot transfer=true2026.02 | 55 | 80 | |
| PrivAct (Verifier)Backbone=Qwen-4B, Zero-shot transfer=true2026.02 | 54.5 | 85 | |
| PrivAct (Refiner)Backbone=Qwen-4B, Zero-shot transfer=true2026.02 | 45 | 80 |