{
  "example_kind": "original synthetic manual planning case",
  "slug": "forecast-threshold-confusion-matrix",
  "title": "Inspect false positives and false negatives at a decision threshold",
  "scenario": "Synthetic planning case: Four cases have supplied ready probabilities 0.9,0.7,0.4,0.2 and observed ready labels 1,0,1,0. The declared rule predicts ready when probability is at least 0.5.",
  "layout": "evaluation",
  "dataset": {
    "heading": "Inspect the invented case records",
    "headers": [
      "Case",
      "Ready probability",
      "Observed ready",
      "Predicted ready",
      "Category"
    ],
    "rows": [
      [
        "A",
        "0.9",
        "1",
        "1",
        "True positive"
      ],
      [
        "B",
        "0.7",
        "0",
        "1",
        "False positive"
      ],
      [
        "C",
        "0.4",
        "1",
        "0",
        "False negative"
      ],
      [
        "D",
        "0.2",
        "0",
        "0",
        "True negative"
      ]
    ]
  },
  "derivation": "At threshold 0.5, A and B predict ready while C and D predict not ready. Only A and D match observations: 2/4=50% accuracy. False-ready B and missed-ready C have different planning consequences.",
  "result": "The rule has one true-ready, one false-ready, one missed-ready and one true-not-ready case. Keep the two error types separate rather than reporting only 50% accuracy.",
  "boundary": "This maps a fixed threshold to four case-level outcome cells, distinct from probabilistic scoring or majority-class accuracy.",
  "limitations": "No real classifier, calibrated readiness forecast or automatic TeamBoostAI decision is asserted.",
  "sources": [
    "original"
  ],
  "tasks": [
    [
      "Freeze the threshold and positive meaning",
      "Evaluation owner",
      "Positive means ready; equality at 0.5 counts as ready."
    ],
    [
      "Retain each case classification",
      "Reviewer",
      "A/B/C/D map to all four named cells under the declared rule."
    ],
    [
      "Review error consequences separately",
      "Decision owner",
      "An early commitment and an unnecessary wait get distinct actions rather than an undifferentiated error count."
    ]
  ],
  "faqs": [
    [
      "Does this threshold have optimal performance?",
      "No. It is a supplied rule; choosing a threshold requires an objective and appropriate evaluation."
    ],
    [
      "Can these counts be treated as probabilities?",
      "They describe four invented cases, not future error rates."
    ]
  ],
  "method_references": [
    {
      "id": "original",
      "title": "Original worked-case definitions",
      "scope": "Definitions, policy choices, records and calculations are authored for this worksheet. No external standard, statistical validation or live product measurement is claimed."
    }
  ]
}