SCORING_PROFILE.json

{
  "name": "Zarn Meeting to Actions",
  "benchmark_track": "priority-gold-track",
  "required_prediction_fields": [
    "example_id",
    "generated_summary",
    "generated_decisions",
    "generated_action_items",
    "generated_followup"
  ],
  "automatic_metrics": [
    "action_item_f1",
    "summary_faithfulness",
    "owner_accuracy",
    "followup_consistency",
    "blocker_preservation"
  ],
  "human_review_checks": [
    "The summary preserves both the decision and what is still unresolved.",
    "Action items are attributable, dated, and dependency-aware.",
    "Follow-up artifacts match the tone and caution level from the meeting."
  ],
  "slice_reporting": {
    "required": [
      "difficulty-easy",
      "difficulty-medium",
      "difficulty-hard",
      "decision-tension",
      "blocker-preservation",
      "owner-date-extraction",
      "tone-sensitive-followup",
      "scope-change"
    ],
    "critical_failure_buckets": [
      "mixed-consensus",
      "owner-and-deadline-coupling",
      "soft-pushback-in-transcript",
      "deferred-scope"
    ]
  },
  "critical_failures": [
    "Polishes away unresolved blockers in the summary.",
    "Drops owners, due dates, or dependencies from action items.",
    "Writes a follow-up that sounds more resolved than the meeting actually was."
  ],
  "multi_reference_support": true
}