| { | |
| "run_id": "direct-r002", | |
| "base_model": { | |
| "repo_id": "AlexWortega/openjev", | |
| "subfolder": "qwen3.5-2b-nli-v5", | |
| "revision": "a20448012c213128955ca0c693e7c943865cab77" | |
| }, | |
| "input": { | |
| "scope": "SKILL.md only", | |
| "max_length": 8192, | |
| "hypothesis": "This AI agent skill is malicious.", | |
| "template": "Premise: {skill_text}\nHypothesis: {hypothesis}", | |
| "truncation": "Premise prefix plus complete fixed hypothesis" | |
| }, | |
| "label2id": { | |
| "contradiction": 0, | |
| "entailment": 1, | |
| "neutral": 2 | |
| }, | |
| "calibration": { | |
| "method": "positive_sigmoid_log_odds", | |
| "slope": 0.76127103215625, | |
| "intercept": 0.3934838733671388, | |
| "fit_partition": "validation", | |
| "formula": "p = sigmoid(slope * (entailment - logaddexp(contradiction, neutral)) + intercept)" | |
| }, | |
| "decision": { | |
| "comparator": ">=", | |
| "raw_log_odds_threshold": -2.515625536875148, | |
| "calibrated_probability_threshold": 0.17922765383653141 | |
| } | |
| } | |