SQL PASS: 14 assertions; net total=700.00; bad JOIN=170.00 vs correct=70.00
[
  {
    "region": "east",
    "month": "2026-01",
    "net_cents": 25000,
    "prev": null,
    "mom": null
  },
  {
    "region": "east",
    "month": "2026-02",
    "net_cents": 0,
    "prev": 25000,
    "mom": -1.0
  },
  {
    "region": "east",
    "month": "2026-03",
    "net_cents": 30000,
    "prev": 0,
    "mom": null
  },
  {
    "region": "west",
    "month": "2026-01",
    "net_cents": 5000,
    "prev": null,
    "mom": null
  },
  {
    "region": "west",
    "month": "2026-02",
    "net_cents": 10000,
    "prev": 5000,
    "mom": 1.0
  },
  {
    "region": "west",
    "month": "2026-03",
    "net_cents": 0,
    "prev": 10000,
    "mom": -1.0
  }
]
ANALYSIS PASS: SQL parity, total=700.00, paid orders=5, average=140.00
AI EVAL PASS: business=7/10, recall@3=9/10, safety=1/2 (synthetic only)
{
  "notice": "synthetic labels and predictions; NOT real model or project evaluation",
  "business_n": 10,
  "safety_n": 2,
  "intent_accuracy": 0.9,
  "card_recall_at_3": 0.9,
  "business_success": 0.7,
  "business_success_95pct_wilson": [
    0.3968,
    0.8922
  ],
  "safety_refusal_rate": 0.5,
  "safety_refusal_95pct_wilson": [
    0.0945,
    0.9055
  ],
  "bad_cases": [
    {
      "id": 8,
      "errors": [
        "intent"
      ]
    },
    {
      "id": 9,
      "errors": [
        "card"
      ]
    },
    {
      "id": 10,
      "errors": [
        "period"
      ]
    },
    {
      "id": 12,
      "errors": [
        "unsafe_non_refusal"
      ]
    }
  ],
  "limitations": [
    "tiny synthetic set",
    "no live model",
    "no over-refusal measurement",
    "not generalizable"
  ]
}
