Best LLMs for Engagement Opportunity Triage

Output schema

Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.

{
  "description": "Per-post triage: should we engage, how risky, and the angle.",
  "properties": {
    "angle": {
      "default": "",
      "description": "Short suggested angle for a helpful reply (empty if ignore).",
      "title": "Angle",
      "type": "string"
    },
    "claim_subjects": {
      "description": "Up to 6 task or model slugs, COPIED VERBATIM from the coverage index you were given, that this post is actually about. These select which measured facts the reply is allowed to draw on, so an invented or approximate slug is worse than an empty list. Empty if ignoring, or if the post is not about a listed subject.",
      "items": {
        "type": "string"
      },
      "title": "Claim Subjects",
      "type": "array"
    },
    "decision": {
      "description": "Whether to engage with this post.",
      "enum": [
        "engage",
        "ignore"
      ],
      "title": "Decision",
      "type": "string"
    },
    "reasoning": {
      "default": "",
      "description": "Brief justification for the decision.",
      "title": "Reasoning",
      "type": "string"
    },
    "relevance": {
      "description": "0-1 relevance to the benchmark's themes (cost/eval/model-selection).",
      "maximum": 1,
      "minimum": 0,
      "title": "Relevance",
      "type": "number"
    },
    "risk_tier": {
      "description": "low=factual/supportive answer; medium=opinion/promotional; high=contentious/critical/sensitive.",
      "enum": [
        "low",
        "medium",
        "high"
      ],
      "title": "Risk Tier",
      "type": "string"
    }
  },
  "required": [
    "decision",
    "relevance",
    "risk_tier"
  ],
  "title": "EngagementTriageDecision",
  "type": "object"
}