{
  "unit": "meeting",
  "aggregation": "summed_counts",
  "headline": {
    "metric": "f1",
    "higher_is_better": true,
    "unit": "fraction"
  },
  "secondary": [
    {
      "id": "action_precision",
      "metric": "precision",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "action_recall",
      "metric": "recall",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "decision_precision",
      "metric": "precision",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "decision_recall",
      "metric": "recall",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "owner_not_speaker_recall",
      "metric": "recall",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "invented_dates",
      "metric": "count",
      "higher_is_better": false,
      "unit": "dates",
      "aggregation": "median"
    },
    {
      "id": "invented_date_rate",
      "metric": "precision",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "summed_counts"
    },
    {
      "id": "action_items_found",
      "metric": "count",
      "higher_is_better": true,
      "unit": "actions",
      "aggregation": "median"
    },
    {
      "id": "gold_action_items",
      "metric": "count",
      "higher_is_better": true,
      "unit": "actions",
      "aggregation": "median"
    }
  ],
  "repeats": 3,
  "seed": 42,
  "max_tokens": 8192,
  "reasoning_effort": "medium",
  "prompt": "You read a line-numbered meeting transcript and extract the commitments and agreements. Every action item must be an open task that a person explicitly accepted or was explicitly assigned in the transcript. Discard completed history, cancelled proposals, suggestions and repeated mentions. An owner is the final agreed owner, not automatically the speaker; when no one has accepted ownership, owner is null. A due date is only a date a person stated for that task, resolved against the stated meeting date; otherwise due_date is null. A decision is an agreement explicitly reached in the meeting, not an open question. Cite the exact line number of the commitment or decision.",
  "verdict_rule": {
    "bootstrap": "paired-items",
    "resamples": 10000,
    "confidence": 0.95,
    "prng": "mulberry32",
    "quantile": "type-7",
    "range": "absolute-difference-exceeds-wider-run-range",
    "label_tie": "json-lexicographic",
    "zero_division": 0,
    "degenerate_kappa": 1,
    "rounding": "half-away-from-zero",
    "decimals": 4
  },
  "record_shape": [
    "tp",
    "fp",
    "fn"
  ],
  "tool_choice": {
    "type": "function",
    "function": {
      "name": "record_meeting"
    }
  },
  "fixture_source": {
    "source": "Original fictional meetings from our deterministic generator",
    "license": "CC0",
    "generators": [
      "Pareto (generator code author; deterministic rendering)"
    ]
  },
  "fixture_sha256": "e46ab9e6b27863846e6e29b3b59f0a20b2962649553ebe8a73e0ba51144de5d3",
  "recipe_sha256": "9b46132dca9aadf909cebe8791a90fee5980c1a693f1713bf2311709946ff74a",
  "checker_sha256": "428f4d8ef03431907e482485fac0345a9aa15dfaee51520044569560d855fc60",
  "runner_sha256": "be8c958ccd72502099276b054078fde6c2d91e69fb8e5c309112064364fbf1f9"
}