{
  "schemaVersion": "1.0",
  "publisher": "The Amateur Limited",
  "report": "independent-field-evaluation",
  "product": "Jev",
  "vendor": "TypeSafe AI",
  "modelVersion": "jev-1.13.0",
  "testDate": "2026-09-19",
  "published": "2026-09-23",
  "sampleSize": 30,
  "overall": {"correct": 29, "accuracy": 0.9667, "meanProbabilityOnSettledAnswer": 0.895},
  "semanticFrontier": {"cases": 13, "correct": 13, "meanProbabilityOnSettledAnswer": 0.957},
  "alreadyDeterministic": {"cases": 7, "correct": 6, "meanProbabilityOnSettledAnswer": 0.634},
  "typedAfterIntake": {"cases": 7, "correct": 7, "meanProbabilityOnSettledAnswer": 1.0},
  "humanAuthorityClassification": {"cases": 3, "correct": 3, "meanProbabilityOnSettledAnswer": 0.987},
  "latencyMs": {"mean": 641, "median": 637, "min": 573, "max": 764, "approxP95": 711},
  "usageTokens": {"input": 13029, "output": 1422},
  "highConfidenceWrongAnswers": 0,
  "bestFit": ["bounded closed-choice semantic classification","unstructured intake to typed route","advisory triage with fallback"],
  "avoid": ["deterministic predicates","exact arithmetic or state checks","open-ended generation","authority-bearing human judgement"],
  "privacyNote": "Task structure and aggregate measurements are preserved; identifying payloads and internal project names are omitted.",
  "canonical": "https://theamateur.co.uk/research/jev/1-13-0/"
}