{
  "passed": true,
  "numeric_fields_compared": 80,
  "evidence_references_checked": 32,
  "findings_count": 7,
  "errors": []
}
No model was invoked. Evidence meaning and error-class coverage require reading the source lines.
Content review: all six expected error classes are covered by seven findings; statistical superiority and forecast are separate findings.
All 32 evidence references support their associated claims or corrections.
Report is consistent with the input and calculated metrics. No invented mismatch causes, profit, statistical test, or future campaign results found.
Independent recalculation from the three CSV/TXT inputs confirms all 80 numeric summary fields.
