{
  "publisher": "Praevideo",
  "template_version": "1.0",
  "artifact_type": "task-design-template",
  "usage": "Adapt with domain experts before evaluation. Contains no cases, model scores, or executable verifiers.",
  "url": "https://praevideogroup.com/benchmarks/#field-operations",
  "title": "Recover a disrupted service schedule",
  "inputs": [
    "Open jobs and priority",
    "Technician qualifications and availability",
    "Service windows",
    "Travel times",
    "Required parts and availability",
    "Overtime policy"
  ],
  "output": [
    "Proposed assignments",
    "Unresolved jobs with reasons",
    "Escalations requiring operator approval"
  ],
  "hard_failures": [
    "Double-booked technician",
    "Missing qualification",
    "Unavailable required parts",
    "Unapproved overtime",
    "Fabricated assignment when constraints cannot be met"
  ],
  "quality_measures": [
    "Jobs completed within service windows",
    "Travel time among feasible plans",
    "Correctly escalated unresolved jobs"
  ],
  "edge_cases": [
    "No qualified technician available",
    "Conflicting service windows",
    "Missing travel information",
    "Parts delivery delayed"
  ],
  "action_boundary": "Propose outputs in a test environment. No production writes.",
  "review_before_use": [
    "Confirm task relevance and authorized source material",
    "Define output schema and allowed tools",
    "Implement outcome checks and validate them with experts",
    "Separate development, training, and held-out evaluation cases",
    "Check split boundaries for duplicate customers, assets, incidents, and templates",
    "Record model, prompt, tools, environment, retries, latency, cost, and human review",
    "Distinguish model failures from evaluation infrastructure failures"
  ]
}