#!/usr/bin/env python3
"""Genera reports/fresh/review_alta.html desde reports/fresh/CANDIDATES_alta.json.
Visor de revisión: tarjetas con diff (rojo=D1 actual / verde=escaneo original),
filtro por libro, marcar/desmarcar visibles y export JSON de aprobados.
Borde ámbar = libro con divergencia densa (densidad de candidatos alta por 1k
palabras D1 >= UMBRAL) -> probable revisión de edición completa, no edits sueltos.
"""
import json, os, html, re
BASE = os.path.dirname(os.path.abspath(__file__))
UMBRAL_DENS = 2.0 # candidatos alta / 1000 palabras D1 -> marca ámbar
cands = json.load(open(os.path.join(BASE, 'reports/fresh/CANDIDATES_alta.json'), encoding='utf-8'))
# --- densidad por grupo (mismo cálculo que triage_fresh.py) ---
ALIAS = {'rvidya': 'rv', 'ejop': 'ej'}
d1w = {}
for f in ('/tmp/fp_small.json', '/tmp/fp_cc.json', '/tmp/fp_sb.json'):
for r in json.load(open(f))[0]['results']:
b = r['book']
n = len(((r.get('translation') or '') + ' ' + (r.get('purport') or '')).split())
if b == 'sb': g = 'sb' + r['ref'].split('/')[1]
elif b == 'cc': g = 'cc_' + r['ref'].split('/')[1]
else: g = b
d1w[g] = d1w.get(g, 0) + n
from collections import Counter
per_grp = Counter(c['group'] for c in cands)
dens = {}
for g, k in per_grp.items():
w = d1w.get(ALIAS.get(g, g), 0)
dens[g] = round(1000 * k / max(w, 1), 1)
amber = {g for g, dv in dens.items() if dv >= UMBRAL_DENS}
groups = sorted(per_grp)
e = html.escape
# --- tarjetas ---
cards = []
for i, c in enumerate(cands):
cls = 'card ed' if c['group'] in amber else 'card'
chg = (f"…{e(c.get('cb',''))} {e(c['veda'])} "
f"{e(c['gold'])} {e(c.get('ca',''))}…")
cards.append(
f'
'
f''
f'
{chg}
')
opts = ''.join(f'' for g in groups)
DATA = json.dumps([{k: c.get(k, '') for k in ('ref','col','veda','gold','cb','group')}
for c in cands], ensure_ascii=False)
doc = f"""Pasada fresca scan↔D1 — candidatos alta
Pasada fresca escaneo↔D1 — {len(cands)} candidatos de confianza alta
rojo=D1 actual · verde=escaneo original. Borde ámbar = libro con divergencia densa (dens≥{UMBRAL_DENS}/1k; probable revisión de edición completa, no edits sueltos). Reescrituras largas aparecen fragmentadas en diffs contiguos del mismo ref.
Libro:
{''.join(cards)}
"""
out = os.path.join(BASE, 'reports/fresh/review_alta.html')
open(out, 'w', encoding='utf-8').write(doc)
print(f"escrito {out}: {len(cands)} tarjetas, {len(groups)} libros")
print("ámbar (densos):", ' '.join(f'{g}({dens[g]})' for g in sorted(amber, key=lambda x:-dens[x])) or '—')