Per-language English-marker density threshold. A brief fails when the measured density exceeds this value AND the absolute count exceeds MIN_ENGLISH_MARKERS.
Three per-script buckets, mirroring the per-language SERP budgets in seo/serp-budgets.ts:
Latin (sv/da/no/fi/de/fr/es/nl): 5 % — mirrors SWEDISH_DENSITY_THRESHOLD in check-analysis-language.ts.
RTL (ar/he): 2 % — RTL prose tokenises into the empty Latin word stream (Arabic / Hebrew script characters are non-Latin), so any Latin English marker is a strong signal of leaked English. The lower floor catches even a single untranslated sentence in what should otherwise be a near-zero Latin-word brief.
CJK (ja/ko/zh): 3 % — legitimate English loanwords are rarer in CJK prose than in Latin scripts, so a single untranslated paragraph stands out at lower density.
Per-language English-marker density threshold. A brief fails when the measured density exceeds this value AND the absolute count exceeds MIN_ENGLISH_MARKERS.
Three per-script buckets, mirroring the per-language SERP budgets in
seo/serp-budgets.ts:sv/da/no/fi/de/fr/es/nl): 5 % — mirrorsSWEDISH_DENSITY_THRESHOLDincheck-analysis-language.ts.ar/he): 2 % — RTL prose tokenises into the empty Latin word stream (Arabic / Hebrew script characters are non-Latin), so any Latin English marker is a strong signal of leaked English. The lower floor catches even a single untranslated sentence in what should otherwise be a near-zero Latin-word brief.ja/ko/zh): 3 % — legitimate English loanwords are rarer in CJK prose than in Latin scripts, so a single untranslated paragraph stands out at lower density.