{
  "id": "zbench-ivf-1.0",
  "set": "ivf-v2",
  "files": {
    "public.jsonl": {
      "bytes": 18824144,
      "lines": 4050,
      "sha256": "cb04d0198ce3b3fb9747afd0cd9248ca9630f687c540638cb4d5114eb943bce3"
    },
    "neighbours.jsonl": {
      "bytes": 81497247,
      "lines": 4050,
      "sha256": "8309c219d576e98d7d11f2a1f949b933c16f9b470985e657bf75eb7fbd8560bc"
    },
    "heldout-questions.jsonl": {
      "bytes": 558104,
      "lines": 980,
      "sha256": "89a3882a56cecafbed88c9a2c9bdae6e565a58360fa9e2b1a6289d105d04e96f"
    }
  },
  "counts": {
    "items": 5030,
    "public": 4050,
    "by_form": {
      "number": 540,
      "fraction": 435,
      "no_cycles": 540,
      "true_zero": 410,
      "zero_fraction": 540,
      "suppressed_part": 540,
      "zero_of_suppressed": 540
    },
    "by_layer": {
      "1": 540,
      "2": 3005,
      "3": 135,
      "4": 540,
      "5": 270,
      "6": 540
    },
    "held_out": 980,
    "by_template": {
      "L1.rate": 540,
      "L6.no-rank": 540,
      "L2.encoding": 3005,
      "L3.evidence": 135,
      "L5.trajectory": 270,
      "L4.national-trap": 540
    }
  },
  "layers": {
    "1": "retrieval",
    "2": "encoding",
    "3": "evidence",
    "4": "reconciliation",
    "5": "temporal",
    "6": "comparison"
  },
  "source": {
    "datasets": {
      "2020": {
        "portal": "data.cdc.gov",
        "success_rates": "3x54-3thk",
        "services_profiles": "92ri-yjps"
      },
      "2021": {
        "portal": "data.cdc.gov",
        "success_rates": "ey8b-ejrf",
        "services_profiles": "ui6g-vumy"
      },
      "2022": {
        "portal": "data.cdc.gov",
        "success_rates": "cchw-gdwa",
        "services_profiles": "ix4g-rt8v"
      }
    },
    "publisher": "CDC, National ART Surveillance System",
    "row_locator": "verify.row on every public item names ClinicId, type, filter, question and breakout"
  },
  "grading": {
    "note": "rules are regular expressions applied case-insensitively to the answer with markdown escapes removed; substitution labels come from the neighbour cells",
    "order": "fail_hard → clarifying question / abstention → Pass (every pass_all) → Half (any half_any; a Half that carries another row's figure is a Fail) → Fail (any fail_any, or another row's figure) → Review",
    "verdicts": [
      "Pass",
      "Half",
      "Fail",
      "Abstain",
      "Review"
    ]
  },
  "license": "zBench IVF is released for model evaluation under the zReports HealthCare Data Lab terms; redistribution of the item set, use as training data, or publication of held-out results requires a written license from zReports.",
  "created_at": "2026-10-12T16:51:39.624Z",
  "grader_version": "5f"
}