{
  "id": "zbench-ivf-1.2",
  "set": "ivf-v2",
  "files": {
    "public.jsonl": {
      "bytes": 19612520,
      "lines": 4139,
      "sha256": "c63e29e21d3245eefeb9aba9ab4a0f0589212cd74fe26ebcbc4ad6c92e132262"
    },
    "neighbours.jsonl": {
      "bytes": 83416670,
      "lines": 4139,
      "sha256": "5f6d4982dc034ac18c57e102a12b6f1c653ed0f4d33318f72480aa213344074b"
    },
    "heldout-questions.jsonl": {
      "bytes": 572232,
      "lines": 1005,
      "sha256": "3f2849cf44c92841c1f3ae98349ff9edfd19b1689abab6c4fe2be9667311298c"
    }
  },
  "counts": {
    "items": 5144,
    "public": 4139,
    "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,
      "7": 78,
      "9": 36
    },
    "by_probe": {
      "encoding: no cycles": 540,
      "encoding: true zero": 410,
      "encoding: zero fraction": 540,
      "retrieval: printed rate": 540,
      "encoding: suppressed numerator": 469,
      "encoding: both parts suppressed": 71,
      "encoding: small-sample fraction": 435,
      "evidence: dataset and row locator": 135,
      "temporal: form change across years": 180,
      "temporal: value change across years": 90,
      "encoding: 0/* (suppressed denominator)": 540,
      "reconciliation: national trap on no cycles": 115,
      "graph: cycle totals and band rate across years": 6,
      "comparison: small sample against a large clinic": 120,
      "comparison: suppressed cell against a large clinic": 420,
      "reconciliation: national trap on suppressed numerator": 278,
      "reconciliation: national trap on both parts suppressed": 50,
      "graph: success rate × patient and cycle characteristics": 18,
      "graph: clinic cell against the national cell, gap in points": 18,
      "reconciliation: national trap on 0/* (suppressed denominator)": 97,
      "graph: profile flags (SART, lab accreditation) with the band rate": 18,
      "graph: a count the record prints (fraction) beside the profile total": 9,
      "economic: a budget against the printed average retrievals per delivery": 12,
      "economic: cost per live birth from the printed average × a premise price": 12,
      "graph: a count the record does not print (rate cell) beside the profile total": 9,
      "economic: no cost where the record prints no average (a dollar figure is invented)": 12
    },
    "held_out": 1005,
    "by_template": {
      "L1.rate": 540,
      "L6.no-rank": 540,
      "L2.encoding": 3005,
      "L3.evidence": 135,
      "L7.derivable": 18,
      "L5.trajectory": 270,
      "L7.profile-gate": 18,
      "L7.rate-x-cycle": 18,
      "L4.national-trap": 540,
      "L9.budget-covers": 12,
      "L9.cost-per-birth": 12,
      "L9.cost-undefined": 12,
      "L7.trend-x-profile": 6,
      "L7.clinic-vs-national": 18
    }
  },
  "layers": {
    "1": "retrieval",
    "2": "encoding",
    "3": "evidence",
    "4": "reconciliation",
    "5": "temporal",
    "6": "comparison",
    "7": "graph joins",
    "9": "economic"
  },
  "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": "2026-10-13T19:30:50.660Z",
  "grader_version": "5i"
}