validation.json

{
  "version": "reviewed-comparison-v1",
  "label": "Bufo count",
  "scale": "0\u201310, recognizable human Bufo labels over ten fixed attempts",
  "denominator": 10,
  "source_batch": "reviewed-comparison-v1-96f165e976d9cce5",
  "description": "A model earns one point only for an explicit human Bufo label. Close and Not Bufo earn zero. Every row uses ten fixed attempts; verified generation failures contribute zero points, while unattempted or unreviewed attempts prevent ranking.",
  "policy": "restored human recognizable-Bufo comparison; exact-hash joins with recorded precedence; no classifier fallback; no partial credit; visitor votes never change rows",
  "cost_policy": "Cost per accepted Bufo uses a complete comparable ten-attempt recorded or estimated cost divided by Bufo count. Partial costs and zero-pass rows remain undefined.",
  "label_coverage": {
    "bufo": 97,
    "close": 4,
    "conflicts": 7,
    "expected": 179,
    "not_bufo": 78,
    "records": 179
  },
  "observed_ability": "A positive count demonstrates ability in this batch only; zero is not a global inability claim."
}