{
  "task_id": 3,
  "name": "Full Relationship Management",
  "difficulty": "hard",
  "description": "The agent cannot finish all 30 emails within 360 minutes — it must decide which to skip. Time-trap emails (up to 180 min) appear even from VIPs, meaning blind VIP-always-respond is punished. Angry follow-ups cascade from repeat VIP ignores. A sunset penalty deducts reward for every high-priority email left unfinished. The agent must balance time efficiency, relationship health, AND priority accuracy simultaneously.",
  "num_emails": 30,
  "vip_count": 8,
  "normal_count": 14,
  "spam_count": 8,
  "time_budget": 360,
  "episode_length": 30,
  "avg_minutes_per_email": 12,
  "features": {
    "action_cost_asymmetry": true,
    "dynamic_followups": true,
    "time_pressure": "severe",
    "sunset_penalty": true,
    "time_traps": true,
    "repeat_ignore_penalty": true
  },
  "what_agent_must_learn": [
    "Cannot respond to all emails — must prioritise strategically",
    "Time-trap emails (120-180 min) from VIPs may not be worth responding to if time is low",
    "Ignoring a VIP twice triggers a second follow-up with even higher urgency + extra -10 relationship health",
    "Sunset penalty = sum of (base_priority × relationship_health/100) for all unfinished emails",
    "Time efficiency compares actual reward to a perfect greedy agent benchmark",
    "Optimal policy: respond to short high-priority emails, ignore long low-priority ones regardless of sender tier"
  ],
  "scoring": {
    "formula": "0.3 * time_efficiency + 0.4 * relationship_health + 0.3 * priority_accuracy",
    "time_efficiency_weight": 0.3,
    "relationship_health_weight": 0.4,
    "priority_accuracy_weight": 0.3,
    "time_efficiency_benchmark": "Perfect greedy agent — picks highest priority emails until time runs out",
    "sunset_penalty_formula": "-sum(base_priority * relationship_health/100) for all remaining emails",
    "notes": "Relationship health still has the highest weight (0.4) — even under time pressure, destroying VIP relationships is costly. Time efficiency rewards staying within budget while maximising reward."
  }
}