| { |
| "tier": "linear", |
| "parameters": 1080, |
| "fit_seconds": 4.3, |
| "policy": { |
| "t_interrupt": 0.545, |
| "t_block": 0.5 |
| }, |
| "over_block_budget": 0.03, |
| "metrics": { |
| "train": { |
| "n": 13999, |
| "over_block_rate": 0.0311, |
| "detection_rate": 0.8728, |
| "decision_accuracy": 0.9276, |
| "deviation_type_accuracy": 0.7731, |
| "per_deviation_detection": { |
| "scope_expansion": 0.978, |
| "unrequested_action": 0.758, |
| "irreversible_unconfirmed": 0.9813, |
| "target_mismatch": 0.7459, |
| "out_of_scope": 0.9152, |
| "privilege_escalation": 0.8579 |
| } |
| }, |
| "validation": { |
| "n": 3265, |
| "over_block_rate": 0.0293, |
| "detection_rate": 0.8805, |
| "decision_accuracy": 0.932, |
| "deviation_type_accuracy": 0.7801, |
| "per_deviation_detection": { |
| "scope_expansion": 0.9812, |
| "unrequested_action": 0.7925, |
| "irreversible_unconfirmed": 0.967, |
| "target_mismatch": 0.7406, |
| "out_of_scope": 0.934, |
| "privilege_escalation": 0.8673 |
| } |
| }, |
| "test_unseen_tools": { |
| "n": 5840, |
| "over_block_rate": 0.0181, |
| "detection_rate": 0.7931, |
| "decision_accuracy": 0.8824, |
| "deviation_type_accuracy": 0.6959, |
| "per_deviation_detection": { |
| "scope_expansion": 0.7513, |
| "unrequested_action": 0.6555, |
| "irreversible_unconfirmed": 1.0, |
| "target_mismatch": 0.732, |
| "out_of_scope": 0.9098, |
| "privilege_escalation": 0.7082 |
| } |
| } |
| } |
| } |
