{
  "dataset": "global_frequency_index",
  "version": "v6.1",
  "build_date": "2026-08-28",
  "builder_script": "scripts/pipeline/build_frequency_v6_1.py",
  "corpus_source": {
    "path": "data/monolingual/master/v2/cleaned.parquet (interim)",
    "sentences": 5481554,
    "description": "Master v2: master v1 с RU-фильтрацией media-ветки (ru_filter_v1)"
  },
  "cleanups_vs_v6.0": {
    "ru_stopwords": {
      "forms": 65,
      "tokens": 648304,
      "list_size": 65,
      "note": "русские служебные слова внутри BA-предложений (цитаты, имена); без башкирских омонимов"
    },
    "singles": {
      "kept": [
        "ә"
      ],
      "note": "а и у удалены (доминирует русское употребление)"
    },
    "note": "склейки и merge опечаток — те же правила, что v6.0"
  },
  "cleanups_vs_v5": {
    "glues_dropped": {
      "forms": 490,
      "tokens": 910
    },
    "singles_dropped": {
      "forms": 41,
      "tokens": 1928669
    },
    "typo_merge": {
      "rule": "DL-1, ratio>=50, len>=5, count<=10",
      "forms_merged": 34919,
      "tokens_reassigned": 65774
    }
  },
  "tokenization": {
    "method": "regex lowercase word extraction",
    "pattern": "[А-ЯЁӘҒҘҠҢӨҪҮҺа-яёәғҙҡңөҫүһ]+"
  },
  "stats": {
    "total_tokens": 83274000,
    "unique_wordforms_full": 1290098,
    "unique_wordforms_public_ge3": 521415,
    "sentences_scanned": 5481554
  },
  "coverage": {
    "25%_forms": 188,
    "50%_forms": 1634,
    "80%_forms": 17559,
    "90%_forms": 53852,
    "95%_forms": 131559
  },
  "files": {
    "full": {
      "file": "production/frequency/v6.1/frequency_full.parquet",
      "rows": 1290098
    },
    "public": {
      "file": "production/frequency/v6.1/frequency_public.parquet",
      "rows": 521415,
      "cutoff": "count >= 3",
      "tokens_covered_pct": 98.83
    },
    "top_coverage_50": {
      "file": "production/frequency/v6.1/top_coverage_50.json",
      "forms": 1634,
      "covers_pct": 50.0
    }
  },
  "schema": {
    "rank": "int64 — позиция по убыванию count",
    "word": "string — словоформа в нижнем регистре",
    "count": "int64 — абсолютная частота (токенов)",
    "ipm": "float — instances per million tokens",
    "doc_freq": "int32 — число предложений, содержащих словоформу"
  },
  "sources_breakdown": {
    "media": 49403302,
    "literature_classical": 18287119,
    "wikipedia": 11413271,
    "encyclopedia": 2147807,
    "speech": 1047292,
    "literature_folklore": 513469,
    "bible": 448204,
    "wikisource": 13536
  }
}