#!/usr/bin/env python3
"""АУДИТ ТИПОВ 36а — grader 0 (author). Rubric-based adjudication of the branch-relevant rows.
name/place = proper noun -> transliterate; title = office/rank/honorific -> translate;
term = common noun/concept/realia -> translate (or established realia translit); nickname = epithet.
BRANCH-HARM (36а) = realia/common-noun mis-typed name/place AND dst came out transliterated."""
from parse_common import parse_bankstop
bs=parse_bankstop()

# adjudicated true class for rows where I judge type WRONG (high confidence). Others assumed OK.
# format: src -> (adjudicated_class, note)
ADJ = {
 # ---- 36а BRANCH-HARM: realia typed name -> transliterated ("too Sinicized") ----
 "元石": ("term","realia: mined mineral (灵泉产出大量元石); name->юаньши = 36а harm"),
 "元海": ("term","realia: internal energy-sea (元海未开); name->Юаньхай = 36а harm; inconsistent w/ 真元海=term"),
 # ---- name mis-typed, but branch benign (terminologist translated anyway) ----
 "古月山寨": ("place","山寨=settlement; name->place; dst translated поселение"),
 "古月族长": ("title","族长=clan chief=title; name->title; dst translated глава клана"),
 # ---- place mis-typed onto common nouns/concepts (benign: translated) ----
 "地下溶洞": ("term","common noun 'underground cave'"),
 "宗祖祠堂": ("term","common noun 'ancestral hall'"),
 "山寨": ("term","common noun 'mountain stronghold'"),
 "池塘": ("term","common noun 'pond'"),
 "灵泉": ("term","common-noun realia 'spirit spring'"),
 "竹楼": ("term","common noun 'bamboo house'"),
 "蛊室": ("term","common noun 'gu chamber'"),
 "酒肆": ("term","common noun 'tavern'"),
 "光阴之河": ("term","poetic concept 'River of Time', not a geographic place"),
 "赤脉": ("term","clan sub-lineage 'Red lineage', not a place"),
 "赤之一脉": ("term","clan sub-lineage, not a place"),
 # ---- title mis-typed onto idiom/kinship ----
 "物是人非": ("term","chengyu/idiom, not a title"),
 "舅母": ("term","kinship term 'maternal aunt', not a title"),
 "舅父": ("term","kinship term 'maternal uncle', not a title"),
}
# BORDERLINE (grader-disagreement plausible; EXCLUDED from mis-type numerator, reported separately)
BORDER = {
 "人祖":  "myth 'Human Ancestor' — name vs title/term; TL waffled Прародитель<->Жэньцзу (near-harm)",
 "古月一族":"clan collective — name vs term",
 "古月师": "古月 clan + 师 — person-name vs role",
 "沈嬷嬷": "嬷嬷=matron title + surname — name vs title",
 "家主阁": "'clan-head pavilion' — place vs term",
 "花海":  "'sea of flowers' — descriptive place vs term",
 "熊家寨": "named settlement — place ok-ish",
 "白家寨": "named settlement — place ok-ish",
 "咏梅":  "poem title (work-title subtype of title)",
 "江城子": "cí tune-name; work-title; TL transliterated Цзянчэнцзы (near-harm)",
 "将敬酒": "poem title (work-title)",
 "管家":  "steward — title vs term",
 "魔道巨擘":"'titan of demonic path' — title vs nickname",
 "花酒行者":"villain epithet — nickname; TL transliterated (near-harm)",
 "采花大盗":"euphemism epithet — nickname vs term",
}

def is_translit(dst):
    core=dst.replace("«","").replace("»","").replace('"',"").strip().lower()
    RU=set("камень море река гора гу мастер уровень ранг класс клан вождь глава поселение храм точка город "
      "бронзовый бронзовое зелёный красный чёрный белый винный червь дом талант культивация океан истинной "
      "сущности серийный насильник великий титан демонического пути жизни воспеваю сливу призыв матушка "
      "камера зал апертура очистка сокровище клад чудесных редчайших десять ветвь странствующего сущность "
      "энергия дух душа первый второй третий четвёртый пятый шестой седьмой восьмой девятый старейшина "
      "наставник управляющий наложница ученик мешок церемония открытие прародитель дядя тётя брат отец мать "
      "сын дочь господин госпожа поколения разряда способности качество отверстий пустотных изначального "
      "подземная пещера крепость пруд духовный источник бамбуковый терем таверна павильон обитель пурпурная "
      "все лица изменилось непредсказуемы ветер облака".split())
    import re
    toks=re.findall(r"[а-яё]+",core)
    return bool(toks) and not any(t in RU for t in toks)

rows=list(bs.items())
N=len(rows)
mistyped=[(s,bs[s]["type"],ADJ[s][0],ADJ[s][1]) for s in ADJ]
harm=[]
for s,(cls,note) in ADJ.items():
    r=bs[s]
    if r["type"] in ("name","place") and cls in ("term","title") and is_translit(r["dst"]):
        harm.append((s,r["type"],cls,r["dst"]))

print(f"=== TYPE AUDIT (grader 0 / author) over N={N} ===")
print(f"  clear mis-types: {len(mistyped)}  ({len(mistyped)/N*100:.1f}%)")
print(f"  borderline (excluded from numerator): {len(BORDER)}")
print(f"\n  === 36а BRANCH-HARM (realia/title typed name/place AND transliterated) : {len(harm)} ===")
for s,t,c,d in harm:
    print(f"     {s}  type={t} -> true={c}  dst='{d}'")
print(f"\n  === all clear mis-types by recorded->true ===")
from collections import Counter
dirs=Counter((bs[s]['type'],ADJ[s][0]) for s in ADJ)
for (a,b),n in sorted(dirs.items(),key=lambda x:-x[1]):
    print(f"     {a:6s} -> {b:6s}: {n}")
print(f"\n  mis-type rate by class (recorded):")
for cls in ("name","place","title","term","nickname"):
    tot=sum(1 for s,r in rows if r['type']==cls)
    mis=sum(1 for s in ADJ if bs[s]['type']==cls)
    bor=sum(1 for s in BORDER if bs[s]['type']==cls)
    print(f"     {cls:8s}: {mis}/{tot} clear-mis ({mis/tot*100:.0f}%)  +{bor} borderline")