{
  "meta": {
    "scenario": "Mei's revenue-cycle model: five CFO questions before and after Prep for AI",
    "chapter": "Chapter 11 - Prep for AI and Data Questions",
    "book_figure": "Figure 11.2",
    "book_figure_svg": "assets/figures/11-before-after-prep.svg",
    "domain": "Healthcare-network BI engineering",
    "synthetic": true,
    "narrative_purpose": {
      "before": "Five CFO-shaped questions asked in one Copilot session against a healthy hand-built semantic model. Two work cleanly; one hits definition ambiguity (which flavour of average); one confidently returns a wrong number; one refuses even though the data is clearly there. Three of five broken on a model Mei was confident was high-quality.",
      "after": "Same five questions re-asked in a fresh session after Prep for AI (schema simplification + verified answers + AI instructions). Four work; one still fails but for the right reason (causal-question shape - out of chart scope)."
    }
  },
  "model": "revenue-cycle",
  "questions": [
    {
      "id": 1,
      "question": "What was total net patient revenue last quarter?",
      "before": {
        "verdict": "worked",
        "notes": "Copilot picked the right measure by construction of the model"
      },
      "after": {
        "verdict": "worked",
        "notes": "Continues to work; verified-answer pinning ensures stability"
      },
      "prep_component": null
    },
    {
      "id": 2,
      "question": "How does that compare to last quarter the year before?",
      "before": {
        "verdict": "worked",
        "notes": "YoY comparison; numbers match Mei's hand-built check"
      },
      "after": {
        "verdict": "worked",
        "notes": "Continues to work"
      },
      "prep_component": null
    },
    {
      "id": 3,
      "question": "What is our average reimbursement per encounter?",
      "before": {
        "verdict": "ambiguous",
        "notes": "Returned a number; Mei cannot tell which definition of average was used"
      },
      "after": {
        "verdict": "worked",
        "notes": "AI instructions declare 'average = mean weighted by encounter complexity, inpatient-only cohort' explicitly"
      },
      "prep_component": "ai_instructions"
    },
    {
      "id": 4,
      "question": "Which payers had the largest denial-rate change month over month?",
      "before": {
        "verdict": "wrong",
        "notes": "Confidently returned a payer ranking based on absolute denial count (not rate); ranked the largest payers first regardless of rate change"
      },
      "after": {
        "verdict": "worked",
        "notes": "Verified-answer pinning maps 'denial-rate change' to the correct measure; schema simplification hid the count-based confusable field"
      },
      "prep_component": "verified_answers + ai_data_schema"
    },
    {
      "id": 5,
      "question": "Why did denials rise in March?",
      "before": {
        "verdict": "wrong",
        "notes": "Returned a paragraph naming three 'likely causes' Mei cannot defend from the data"
      },
      "after": {
        "verdict": "refused",
        "notes": "AI instructions include a causal-question refusal directive; Copilot now says 'I cannot answer causal questions from this model' and points to the descriptive comparison alternatives"
      },
      "prep_component": "ai_instructions"
    }
  ],
  "prep_components": {
    "ai_data_schema": {
      "kind": "restrict",
      "tables_deselected": 3,
      "columns_deselected": 27
    },
    "verified_answers": {
      "kind": "trigger_phrase_map",
      "pairs": 5
    },
    "ai_instructions": {
      "kind": "brief",
      "lines": 6,
      "includes_causal_refusal": true
    }
  },
  "before_after_summary": {
    "before_worked": 2,
    "before_ambiguous_or_wrong": 3,
    "after_worked": 4,
    "after_refused_but_correct_shape": 1,
    "after_still_broken": 0
  },
  "source_sha256": "90ca5662f040bbd3340764a4f2794419c6deb0675865899c003d08872cf48d2b"
}