provenance.json

{
 "source": {
  "dataset": "google/IFEval",
  "revision": "966cd89545d6b6acfd7638bc708b98261ca58e84",
  "items_total": 541,
  "items_excluded": 112,
  "items_released": 429
 },
 "translator": {
  "model": "google/gemini-3.1-pro-preview",
  "provider_pin": "google-vertex/global/flex",
  "reasoning": {
   "effort": "high"
  },
  "temperature": 0,
  "guideline_sha256": [
   "c090a5261f66cd366204e41634984c71657150da5e9ae054f9dd0ae656f02b53"
  ],
  "cost_usd": 3.6891
 },
 "checker": "checker/ (allganize/IFEval-Ko@54199e3 with [hanifeval] changes; v1.1 changes listed under v1.1)",
 "edits": [
  {
   "key": 1546,
   "id": "E01",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 1627,
   "id": "E02",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 2063,
   "id": "E03",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 2337,
   "id": "E04",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 2713,
   "id": "E05",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 2739,
   "id": "E06",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 3633,
   "id": "E07",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 3718,
   "id": "E08",
   "finding": "R+V",
   "prompt": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "repeat": [
    [
     "쉼표(,)",
     "쉼표"
    ]
   ],
   "kwargs": {},
   "reason": "Guideline rule 6 wording '쉼표(,)' put an ASCII comma inside prompt_to_repeat, so the required verbatim repeat always fails punctuation:no_comma (strict). '(,)' removed from prompt and prompt_to_repeat."
  },
  {
   "key": 1691,
   "id": "E09",
   "finding": "R+C",
   "prompt": [
    [
     "모저크리스탈",
     "모저"
    ],
    [
     "카를로비바리",
     "바리"
    ]
   ],
   "repeat": [],
   "kwargs": {
    "1": {
     "forbidden_words": [
      "모저",
      "유리공장",
      "프라프치체",
      "카를로비",
      "바리"
     ]
    }
   },
   "reason": "Guideline v1.1 over-corrected: '모저크리스탈' lets bare '모저' pass. Reverted to the v1.0 pilot rendering (Gemini output, work/pilot_translations.jsonl). Corpus screen: '모저' 0.00/10k, '바리' 0.31/10k eojeol, so the collision v1.1 guarded against is rare."
  },
  {
   "key": 2471,
   "id": "E10",
   "finding": "R+C",
   "prompt": [
    [
     "'미친'",
     "'광기'"
    ]
   ],
   "repeat": [],
   "kwargs": {
    "0": {
     "forbidden_words": [
      "슬픈",
      "광기",
      "스트레스"
     ]
    }
   },
   "reason": "'미친' (crazy) is a substring of '영향을 미친다', routine in the requested academic paper, so faithful answers fail. Replaced by another negative example word, '광기' (madness; 0.00/10k). The English lists the words as examples ('such as')."
  },
  {
   "key": 2549,
   "id": "E11",
   "finding": "R+C",
   "prompt": [
    [
     "'가오'",
     "'gao'"
    ]
   ],
   "repeat": [],
   "kwargs": {
    "1": {
     "keywords": [
      "gao",
      "하트"
     ]
    }
   },
   "reason": "Keyword '가오' is satisfied by '다가오는' (0.46/10k), giving free credit. The source keyword 'gao' is a nonce token; kept in Latin script, as in the English item."
  },
  {
   "key": 3369,
   "id": "E12",
   "finding": "R",
   "prompt": [],
   "repeat": [],
   "kwargs": {
    "1": {
     "keyword": "옳"
    }
   },
   "reason": "Keyword '옳다' (dictionary form) misses 옳은/옳습니다, so the '< 2' cap was near-vacuous. Stem '옳' (0.15/10k) counts all forms. Consequence: the required repeat now contains it twice (옳은, 옳다), so the item is unsatisfiable under strict scoring, exactly like the English source ('right' twice in the repeat). Tagged source_unsatisfiable_strict."
  },
  {
   "key": 1733,
   "id": "E13",
   "finding": "R",
   "prompt": [],
   "repeat": [],
   "kwargs": {
    "0": {
     "keyword": "대답했"
    }
   },
   "reason": "Keyword '대답했다' fails a fairy tale written in -습니다 style ('대답했습니다'). Stem '대답했' covers both and is a substring of the prompt's '대답했다'."
  },
  {
   "key": 2515,
   "id": "E14",
   "finding": "R",
   "prompt": [
    [
     "1개의 구역을",
     "최소 1개의 구역을"
    ]
   ],
   "repeat": [],
   "kwargs": {},
   "reason": "Source 'at least one'; kwargs num_highlights >= 1. '1개의' read as exactly one."
  },
  {
   "key": 3629,
   "id": "E15",
   "finding": "R",
   "prompt": [
    [
     "핵심 부분 1개를",
     "핵심 부분을 1개 이상"
    ]
   ],
   "repeat": [],
   "kwargs": {},
   "reason": "Source 'some key parts'; kwargs >= 1. '1개를' read as exactly one."
  },
  {
   "key": 3549,
   "id": "E16",
   "finding": "R",
   "prompt": [
    [
     "정확히 두 번",
     "두 번 이상"
    ]
   ],
   "repeat": [],
   "kwargs": {},
   "reason": "Translator added 'exactly'; source 'twice', kwargs num_highlights >= 2."
  },
  {
   "key": 3624,
   "id": "E17",
   "finding": "R",
   "prompt": [
    [
     "(예: \"\n\n\")",
     "(예: \"\\n\\n\")"
    ]
   ],
   "repeat": [],
   "kwargs": {},
   "reason": "The source's literal '\\n\\n' example became two real newlines (an empty quote). Restored the literal. Not scored."
  }
 ],
 "response_edits": [
  {
   "key": 1546,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E01"
  },
  {
   "key": 1627,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E02"
  },
  {
   "key": 2063,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E03"
  },
  {
   "key": 2337,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E04"
  },
  {
   "key": 2713,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E05"
  },
  {
   "key": 2739,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E06"
  },
  {
   "key": 3633,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E07"
  },
  {
   "key": 3718,
   "old": "쉼표(,)",
   "new": "쉼표",
   "reason": "follow E08"
  },
  {
   "key": 2549,
   "old": "가오",
   "new": "gao",
   "reason": "follow E11"
  }
 ],
 "known_issues": {
  "374": [
   "source_unsatisfiable_strict"
  ],
  "3371": [
   "source_unsatisfiable_strict"
  ],
  "3369": [
   "source_unsatisfiable_strict"
  ],
  "2028": [
   "lexical_collision"
  ],
  "1580": [
   "lexical_inflection_weak"
  ]
 },
 "v1.1": {
  "date": "2026-10-03",
  "edits": [
   {
    "key": 357,
    "id": "V11-01",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "응답은 3개의 섹션으로",
      "응답은 3개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 1127,
    "id": "V11-02",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "4개의 섹션으로 구성되어야",
      "4개 이상의 섹션으로 구성되어야"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 1131,
    "id": "V11-03",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "이력서를 4개의 섹션으로",
      "이력서를 4개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 1548,
    "id": "V11-04",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "문서는 4개의 섹션으로",
      "문서는 4개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 1730,
    "id": "V11-05",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "응답은 5개의 섹션으로",
      "응답은 5개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 2023,
    "id": "V11-06",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "시는 4개의 섹션으로",
      "시는 4개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 2889,
    "id": "V11-07",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "라는 두 개의 섹션을 포함하세요",
      "를 포함해 섹션을 두 개 이상 작성하세요"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 2925,
    "id": "V11-08",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "농담을 3개의 섹션으로",
      "농담을 3개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 3324,
    "id": "V11-09",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "각 버전은 7개의 섹션으로",
      "각 버전은 7개 이상의 섹션으로"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 340,
    "id": "V11-10",
    "finding": "review-1, review-2",
    "prompt": [
     [
      "섹션을 1개 포함해 주세요",
      "섹션을 1개 이상 포함해 주세요"
     ]
    ],
    "kwargs": {},
    "reason": "Checker scores this count as >= N (as in Google's IFEval); the prompt said N. Now states 'N개 이상'."
   },
   {
    "key": 1342,
    "id": "V11-11",
    "finding": "review-2 F01; corpus screen (사진 10.86/10k)",
    "prompt": [
     [
      "\"사진\"이라는 단어는 언급하지 마세요.",
      "\"사진\"이라는 단어는 언급하지 마세요(\"청사진\"처럼 \"사진\"이 들어간 다른 단어도 쓰지 마세요)."
     ]
    ],
    "kwargs": {},
    "reason": "keywords:frequency < 1 on '사진' is a substring rule, so '청사진' (a common business word) fails. The prompt now states the rule the checker applies, as item 3114 does ('글자')."
   },
   {
    "key": 1466,
    "id": "V11-12",
    "finding": "review-1, review-2 F11",
    "prompt": [
     [
      "'테일러', '스위프트', '투게더'라는 단어는",
      "'테일러', '스위프트', '투게더'라는 단어(영문 'Taylor', 'Swift', 'Together' 포함)는"
     ]
    ],
    "kwargs": {
     "1": {
      "forbidden_words": [
       "테일러",
       "스위프트",
       "투게더",
       "Taylor",
       "Swift",
       "Together"
      ]
     }
    },
    "reason": "The prompt keeps the English song title, so the Latin forms escaped the Hangul-only forbidden list; the English item forbids them. Latin forms added to prompt and kwargs (matching is case-insensitive)."
   },
   {
    "key": 1733,
    "id": "V11-13",
    "finding": "review-1, review-2 F04",
    "prompt": [
     [
      "'대답했다'라는 단어가 최소 2번 이상 포함되어야 합니다.",
      "'대답했다'라는 표현이 최소 2번 이상 포함되어야 합니다('대답했습니다'처럼 어미가 바뀐 형태도 셉니다)."
     ]
    ],
    "kwargs": {},
    "reason": "Since v1 (E13) the checker counts the stem '대답했'; the prompt now says that inflected forms count."
   },
   {
    "key": 3311,
    "id": "V11-14",
    "finding": "review-2 F05 (inherited from google/IFEval)",
    "prompt": [],
    "kwargs": {
     "2": {
      "keywords": [
       "지표",
       "목표",
       "관리"
      ]
     }
    },
    "reason": "The prompt requires three keywords; the source kwargs check two ('objective' missing). v1.1 fixes inherited prompt/kwargs mismatches on scored instructions."
   },
   {
    "key": 30,
    "id": "V11-15",
    "finding": "review-2 F15",
    "prompt": [
     [
      "소문자로 답변해 주세요. ",
      ""
     ]
    ],
    "kwargs": {},
    "reason": "'Answer in lower case' has no meaning for Hangul and is not scored (the change_case checker is not in this item). Removed; adaptation = task."
   },
   {
    "key": 2807,
    "id": "V11-16",
    "finding": "review-2 F15",
    "prompt": [
     [
      " 오직 소문자만 사용하세요.",
      ""
     ]
    ],
    "kwargs": {},
    "reason": "Same as V11-15."
   }
  ],
  "response_edits": [],
  "known_issues": {
   "374": [
    "source_unsatisfiable_strict"
   ],
   "3371": [
    "source_unsatisfiable_strict"
   ],
   "3369": [
    "source_unsatisfiable_strict"
   ],
   "2028": [
    "lexical_collision"
   ],
   "1580": [
    "lexical_inflection_weak"
   ],
   "2859": [
    "source_checker_artifact"
   ],
   "3305": [
    "source_contradiction"
   ],
   "2078": [
    "source_contradiction"
   ]
  },
  "adaptation": {
   "30": "task",
   "127": "lexical",
   "301": "lexical",
   "1242": "lexical",
   "1342": "lexical",
   "1476": "lexical",
   "1580": "lexical",
   "1629": "lexical",
   "1733": "lexical",
   "1936": "lexical",
   "2028": "lexical",
   "2041": "lexical",
   "2207": "task",
   "2395": "lexical",
   "2471": "lexical",
   "2577": "lexical",
   "2716": "task",
   "2807": "task",
   "2997": "lexical",
   "3081": "lexical",
   "3114": "lexical",
   "3345": "lexical",
   "3369": "lexical",
   "3386": "lexical"
  },
  "answer_language_overrides": {
   "1137": [
    "fr"
   ],
   "1675": [
    "de"
   ],
   "2534": [
    "de",
    "ko"
   ]
  },
  "checker_changes": [
   "sentence splitter: list markers only at line start (review-2 F02, F12)",
   "response_language: undetectable text fails (review-2 F13)",
   "kwargs strings NFC-normalised (review-1)"
  ]
 },
 "files": {
  "release/hanifeval_v1.jsonl": "46d85943714ed283787e5e5575181f837e1fdb46cdf4e6175537fdc050881fcc",
  "release/hanifeval_v1.1.jsonl": "9d8f5792fa9a055a0193c2b830d738723a3ab639a613c21ed372a846c87663de",
  "work/translations.jsonl": "a8f364b8a5feb65b400e8c77fb9c81d33a1c234d7597641bf42a2465a2b9ada7",
  "translation_guideline.md": "c090a5261f66cd366204e41634984c71657150da5e9ae054f9dd0ae656f02b53"
 }
}