{
  "dataset_name": "KoEVD",
  "label_policy": "Public JSONL records use paper-facing descriptive labels. Legacy internal codes are not duplicated in public records.",
  "severity_band": [
    "Benign",
    "Low",
    "Moderate",
    "High",
    "Extreme",
    "HD/borderline"
  ],
  "explicitness": [
    "Neutral",
    "Implicit",
    "Explicit",
    "Not Applicable",
    "Hold/Borderline"
  ],
  "utterance_type": [
    "Exploring Opinions",
    "Expressing Opinion",
    "Providing Information",
    "Proverb/Term Meaning",
    "Biased Qs",
    "Prohibited Acts",
    "Controversial Requests",
    "Copyright",
    "Personal Info",
    "Specialized Advice",
    "Future Predictions"
  ],
  "topic": [
    "Origin",
    "Religion",
    "Age & Generation",
    "Taste",
    "Disability",
    "Physical Appearance",
    "Gender",
    "Sexuality",
    "Politics",
    "Socio-economic Status",
    "Marital Status",
    "Parenting",
    "Family Form",
    "Criminal/Abusive Behaviors",
    "Education",
    "Personal Information",
    "National Security",
    "Intellectual Property",
    "Specialized Advice",
    "Future Prediction",
    "Other Sensitive",
    "Null/Benign"
  ],
  "response_tier": [
    "Accepted",
    "Rejected",
    "Strongly Rejected",
    "Chosen"
  ],
  "response_strategy": [
    "Evaluate Utterance",
    "Set a Boundary",
    "Define Terms",
    "Detouring Question",
    "Sanitized Reiteration",
    "Counter-speech",
    "Conditional Empathy",
    "Neutral Framing",
    "Reasoning",
    "Safe Information",
    "Educating / Safer Alternative",
    "Summation",
    "Social Expression",
    "Redirecting",
    "Humour"
  ],
  "task5_risk_category": [
    "Cyber/Illegal Harm",
    "Privacy",
    "Copyright",
    "Harmful Instruction",
    "Harmful Publication",
    "High-Stakes Action",
    "Safe High-Stakes Information",
    "Safe Information",
    "Ambiguous"
  ]
}