{
  "title": "Underground coal public reporting periods for AI",
  "source_page": "https://fiund.com/industries/underground-coal-report-period-data",
  "authorship": "Original fictional Fiund editorial exercise. No actual operating records or third-party source assets are included.",
  "example": {
    "title": "Apply a category correction within its reporting period",
    "context": "A fictional public-style coal summary uses made-up categories and event counts. It is designed to evaluate reporting scope, not mining operations or safety performance.",
    "records": [
      [
        "UC1",
        "Quarter Q1 summary revision 1",
        "January 1–March 31: category A=8, category B=5; reporting coverage group G."
      ],
      [
        "UC2",
        "Authorized Q1 correction",
        "Category A corrected to 7. Category B and coverage group unchanged."
      ],
      [
        "UC3",
        "April rolling summary",
        "March 1–April 30: A=6 and B=4, coverage groups G and H combined; no underlying event list."
      ]
    ],
    "question": "What are the current category counts for Q1 group G, and can the April summary identify which event caused the correction?",
    "answer": "UC2 changes Q1 group G category A to 7 while leaving category B at 5 under UC1. UC3 covers a different time window and combines G with H, so it cannot replace the Q1 figures. Without an event list or an explicit correction explanation, none of these aggregates identifies the underlying event, location or cause responsible for the revision.",
    "failureModes": [
      "Replacing Q1 counts with the newer rolling summary despite different dates and coverage.",
      "Inventing a removed event or a cause from the one-count correction alone."
    ],
    "fields": [
      [
        "Period start and end",
        "Keeps calendar-quarter and rolling-window summaries distinct."
      ],
      [
        "Coverage population",
        "Preserves G-only versus G-plus-H scope."
      ],
      [
        "Category-specific correction",
        "Avoids changing untouched fields."
      ],
      [
        "Underlying-event availability",
        "Limits explanations to what the aggregate actually establishes."
      ]
    ],
    "holdout": "Add a correction for category B in group H during April. Require the model to decline applying it to Q1 group G, even when the category label matches.",
    "boundary": "Use fictional aggregates or appropriately cleared public summaries. Exclude maps, exact locations, private production data, workers, equipment configurations and emergency-response details."
  }
}