parse-peru-wiki.mjs

/**
 * Parser determinístico das tabelas de pesquisas da Wikipedia "Opinion polling for the
 * 2026 Peruvian general election" (HTML renderizado via API). Materializa o grid respeitando
 * rowspan/colspan, mapeia colunas->candidatos pelo cabeçalho e extrai 1º turno (14 candidatos)
 * + 2º turno (Fujimori × Sánchez). Verifica contra valores conhecidos antes de gravar.
 *
 * Entrada: ../AFOS-Analitica-2026/.cache/peru-wiki.html  (string HTML)
 * Saída: polls/peru-first-round-polls.csv (long), polls/peru-runoff-polls.csv, polls/peru-polls.json
 */
import { readFileSync, writeFileSync, mkdirSync } from 'fs'
import { join } from 'path'

const HTML = readFileSync(join(process.cwd(), '..', '..', 'AFOS-Analitica-2026', '.cache', 'peru-wiki.html'), 'utf-8')
const OUT = join(process.cwd(), 'polls'); mkdirSync(OUT, { recursive: true })
const csv = (rows) => rows.map((r) => r.map((v) => { const s = String(v ?? ''); return /[",\n]/.test(s) ? `"${s.replace(/"/g, '""')}"` : s }).join(',')).join('\n') + '\n'

// candidatos conhecidos: surname-chave -> {name, party}
const CAND = {
  'Fujimori': { name: 'Keiko Fujimori', party: 'Fuerza Popular' },
  'Sánchez': { name: 'Roberto Sánchez', party: 'Juntos por el Perú' },
  'López Aliaga': { name: 'Rafael López Aliaga', party: 'Renovación Popular' },
  'Nieto': { name: 'Jorge Nieto', party: 'Partido de Buen Gobierno' },
  'Belmont': { name: 'Ricardo Belmont', party: 'OBRAS' },
  'Álvarez': { name: 'Carlos Álvarez', party: 'País para Todos' },
  'López Chau': { name: 'Alfonso López Chau', party: 'Ahora Nación' },
  'Pérez Tello': { name: 'Marisol Pérez Tello', party: 'Demócrata Verde' },
  'Espá': { name: 'Carlos Espá', party: '' },
  'Olivera': { name: 'Fernando Olivera', party: 'Frente de la Esperanza' },
  'Luna': { name: 'José Luna', party: 'Podemos Perú' },
  'Lescano': { name: 'Yonhy Lescano', party: 'Cooperación Popular' },
  'Acuña': { name: 'César Acuña', party: 'Alianza para el Progreso' },
  'Valderrama': { name: 'Enrique Valderrama', party: 'APRA' },
}
const SURNAMES = Object.keys(CAND)

const decode = (s) => s
  .replace(/<sup[^>]*class="[^"]*reference[^"]*"[^>]*>[\s\S]*?<\/sup>/gi, '') // refs
  .replace(/<style[\s\S]*?<\/style>/gi, '')
  .replace(/<[^>]+>/g, '') // demais tags
  .replace(/&#(\d+);/g, (_, n) => String.fromCharCode(+n)) // entidades numéricas (&#32; espaço, &#8211; ndash)
  .replace(/&nbsp;/g, ' ').replace(/&ndash;/g, '–').replace(/&amp;/g, '&').replace(/&[a-z]+;/gi, ' ')
  .replace(/ /g, ' ')
  .replace(/\s+/g, ' ').trim()

// extrai todas as tabelas wikitable
function tables(html) {
  const out = []
  const re = /<table[^>]*class="[^"]*wikitable[^"]*"[\s\S]*?<\/table>/gi
  let m; while ((m = re.exec(html))) out.push(m[0])
  return out
}

// materializa o grid de uma tabela respeitando rowspan/colspan
function grid(tableHtml) {
  const trs = tableHtml.split(/<tr[^>]*>/i).slice(1).map((x) => x.split(/<\/tr>/i)[0])
  const g = []; const pending = [] // {col, span, rows, text}
  for (const tr of trs) {
    const cells = []
    const cre = /<(t[hd])([^>]*)>([\s\S]*?)<\/\1>/gi
    let cm
    while ((cm = cre.exec(tr))) {
      const attrs = cm[2]
      const rs = parseInt((attrs.match(/rowspan="?(\d+)/i) || [])[1] || '1', 10)
      const cs = parseInt((attrs.match(/colspan="?(\d+)/i) || [])[1] || '1', 10)
      cells.push({ text: decode(cm[3]), rs, cs })
    }
    const row = []; let col = 0; let ci = 0
    const place = (text) => { row[col] = text; col++ }
    while (ci < cells.length || pending.some((p) => p.rows > 0)) {
      const p = pending.find((x) => x.col === col && x.rows > 0)
      if (p) { for (let k = 0; k < p.span; k++) place(p.text); p.rows--; continue }
      if (ci >= cells.length) break
      const c = cells[ci++]
      for (let k = 0; k < c.cs; k++) { const cc = col; place(c.text); if (c.rs > 1) pending.push({ col: cc, span: 1, rows: c.rs - 1, text: c.text }) }
    }
    for (const p of pending) if (p.rows > 0 && p.col >= row.length) { /* trailing spans */ }
    g.push(row)
  }
  return g
}

// rótulo de cada coluna a partir das linhas de cabeçalho
function labelColumns(g) {
  const headerRows = g.slice(0, 6)
  const ncol = Math.max(...g.map((r) => r.length))
  const labels = []
  for (let c = 0; c < ncol; c++) {
    const texts = headerRows.map((r) => r[c] || '')
    let label = null
    for (const sn of SURNAMES) if (texts.some((t) => t === sn || t.includes(sn))) { label = `cand:${sn}`; break }
    if (!label) {
      const joined = texts.join(' ').toLowerCase()
      if (/pollster/.test(joined)) label = 'pollster'
      else if (/sample/.test(joined)) label = 'sample'
      else if (/margin/.test(joined)) label = 'margin'
      else if (/date/.test(joined)) label = 'date'
      else if (/other/.test(joined)) label = 'other'
      else if (/blank|none/.test(joined)) label = 'blank'
      else if (/undecided/.test(joined)) label = 'undecided'
      else if (/lead/.test(joined)) label = 'lead'
    }
    labels.push(label)
  }
  return labels
}

const MONTHS = { jan: '01', feb: '02', mar: '03', apr: '04', may: '05', jun: '06', jul: '07', aug: '08', sep: '09', oct: '10', nov: '11', dec: '12' }
// "3–4 Apr 2026" / "28 Feb–5 Mar 2026" / "6 Jun 2026" -> ISO da data FINAL. fy = ano fallback p/ datas sem ano.
function endDate(s, fy) {
  if (!s) return ''
  const last = s.split(/[–-]/).pop().trim() // após o último traço
  const ym = s.match(/([A-Za-z]{3})[a-z]*\s+(\d{4})/) // mês+ano do todo (fallback)
  let day, mon, year
  const mm = last.match(/(\d{1,2})\s+([A-Za-z]{3})[a-z]*\s+(\d{4})/)
  if (mm) { day = mm[1]; mon = MONTHS[mm[2].toLowerCase()]; year = mm[3] }
  else {
    const d = last.match(/(\d{1,2})/); const monM = last.match(/([A-Za-z]{3})/) || s.match(/([A-Za-z]{3})/)
    day = d ? d[1] : null; mon = monM ? MONTHS[monM[1].toLowerCase()] : null; year = ym ? ym[2] : fy
  }
  if (!day || !mon || !year) return ''
  return `${year}-${mon}-${String(day).padStart(2, '0')}`
}
const num = (s) => { const m = String(s).replace(/,/g, '').match(/-?\d+(?:\.\d+)?/); return m ? parseFloat(m[0]) : null }
const isData = (txt) => txt && !/^pollster|^date|^sample|^margin|^other|^blank|^lead|results?$/i.test(txt) && txt.length > 1
// canonicaliza grafias de instituto (mesmo pollster fragmentado / typo de diacrítico)
const normPollster = (p) => p.replace('Ipsos Peru/Péru 21', 'Ipsos Perú/Perú 21').replace('Ipsos Perú/Perú21', 'Ipsos Perú/Perú 21')

// processa todas as tabelas
const frPolls = []; const ruPolls = []
const allTables = tables(HTML)
allTables.forEach((t, ti) => {
  const g = grid(t)
  const labels = labelColumns(g)
  const candCols = labels.map((l, i) => ({ l, i })).filter((x) => x.l && x.l.startsWith('cand:'))
  if (candCols.length < 2) return
  const colOf = (name) => labels.indexOf(name)
  const pc = colOf('pollster'), dc = colOf('date'), sc = colOf('sample')
  if (pc < 0) return
  const isRunoff = candCols.length === 2 && candCols.some((x) => x.l === 'cand:Fujimori') && candCols.some((x) => x.l === 'cand:Sánchez')
  // ano predominante da tabela (p/ datas sem ano, ex. seções "2025")
  const yc = {}; for (const row of g) { const m = String(row[dc] || '').match(/\b(20\d\d)\b/); if (m) yc[m[1]] = (yc[m[1]] || 0) + 1 }
  const tableYear = Object.keys(yc).sort((a, b) => yc[b] - yc[a])[0] || null
  let kept = 0
  for (const row of g) {
    const pollster = row[pc]
    if (!isData(pollster)) continue
    const date = dc >= 0 ? row[dc] : ''
    const iso = endDate(date, tableYear)
    if (process.env.DBG && !isRunoff) console.error(`    [#${ti}] "${pollster.slice(0, 18)}" date="${date}" -> ${iso || 'FAIL'}`)
    if (!iso) continue // linhas sem data válida (separadores, resultado sem data parseável) são puladas
    if (iso < '2026-01-01' || iso > '2026-07-01') continue // janela do ciclo (ajustar floor ao país)
    if (pollster.length > 35) continue // pollster real é curto; eventos/contagens têm texto longo
    if (/death|deadline|withdraw|election|result|tribunal|count|finali|decision of vote|preliminary|rapid|candidate/i.test(pollster) || /\b(19|20)\d\d\b/.test(pollster)) continue
    const sample = sc >= 0 ? num(row[sc]) : null
    if (sample && sample > 50000) continue // contagem oficial de votos, não pesquisa
    const results = []
    for (const { l, i } of candCols) {
      const sn = l.slice(5); const v = num(row[i])
      if (v != null && v <= 100) results.push({ surname: sn, candidate: CAND[sn].name, party: CAND[sn].party, percent: v })
    }
    // 1º turno real lista vários candidatos; eventos/anotações têm 0-1 → exige ≥4 (runoff ≥2)
    if (results.length < (isRunoff ? 2 : 4)) continue
    if (!isRunoff && results.every((r) => r.percent === results[0].percent)) continue // 1º turno: todos iguais = artefato (runoff PODE empatar legitimamente, ex. 38–38)
    const rec = { poll_date: iso, fieldwork: date, pollster: normPollster(pollster.replace(/\s*\(.*/, '').trim()), sample, results }
    if (isRunoff) ruPolls.push(rec); else frPolls.push(rec)
    kept++
  }
  if (kept > 0) console.error(`  tabela #${ti}: ${candCols.length} candidatos, ${isRunoff ? 'RUNOFF' : '1º turno'}, ${kept} linhas 2026`)
})

// dedup por (pollster, poll_date, round) mantendo o mais completo
const dedup = (arr) => {
  const m = new Map()
  for (const p of arr) { const k = `${p.pollster}|${p.poll_date}`; const prev = m.get(k); if (!prev || p.results.length > prev.results.length) m.set(k, p) }
  return [...m.values()].sort((a, b) => a.poll_date.localeCompare(b.poll_date))
}
const FR = dedup(frPolls); const RU = dedup(ruPolls)

// CSVs long
const frRows = [['poll_date', 'fieldwork', 'pollster', 'sample', 'candidate', 'party', 'percent']]
for (const p of FR) for (const r of p.results) frRows.push([p.poll_date, p.fieldwork, p.pollster, p.sample ?? '', r.candidate, r.party, r.percent])
writeFileSync(join(OUT, 'peru-first-round-polls.csv'), csv(frRows))

const ruRows = [['poll_date', 'fieldwork', 'pollster', 'sample', 'fujimori_pct', 'sanchez_pct', 'lead_pp']]
for (const p of RU) {
  const f = p.results.find((r) => r.surname === 'Fujimori')?.percent
  const s = p.results.find((r) => r.surname === 'Sánchez')?.percent
  if (f != null && s != null) ruRows.push([p.poll_date, p.fieldwork, p.pollster, p.sample ?? '', f, s, Math.round((f - s) * 100) / 100])
}
writeFileSync(join(OUT, 'peru-runoff-polls.csv'), csv(ruRows))

writeFileSync(join(OUT, 'peru-polls.json'), JSON.stringify({
  description: 'National opinion polls for the 2026 Peruvian general election — all candidates, first round + runoff. Parsed deterministically from the Wikipedia aggregation (rowspan/colspan-aware).',
  source: 'Wikipedia "Opinion polling for the 2026 Peruvian general election" (rendered table HTML)',
  election: { first_round: '2026-04-12', runoff: '2026-06-07', runoff_matchup: 'Keiko Fujimori (Fuerza Popular) vs Roberto Sánchez (Juntos por el Perú)' },
  counts: { first_round_polls: FR.length, first_round_rows: frRows.length - 1, runoff_polls: RU.length },
  first_round: FR, runoff: RU,
}, null, 2))

// --- verificação contra valores conhecidos ---
console.log(`1º turno: ${FR.length} pesquisas (${frRows.length - 1} linhas), 2º turno: ${RU.length} pesquisas`)
const cands = new Set(); for (const p of FR) for (const r of p.results) cands.add(r.surname)
console.log(`candidatos distintos no 1º turno: ${cands.size} (${[...cands].join(', ')})`)
const jun6 = RU.find((p) => p.poll_date === '2026-06-06' && /Ipsos/i.test(p.pollster))
console.log(`CHECK 2T Ipsos 06/Jun (esperado ~Fujimori 44.1 / Sánchez 43.7): ${jun6 ? JSON.stringify(jun6.results.map((r) => [r.surname, r.percent])) : 'NÃO ACHADO'}`)
const apr = FR.filter((p) => p.poll_date.startsWith('2026-04')).slice(-1)[0]
console.log(`CHECK 1T pesquisa mais recente: ${apr ? apr.poll_date + ' ' + apr.pollster + ' ' + JSON.stringify(apr.results.slice(0, 3).map((r) => [r.surname, r.percent])) : 'n/a'}`)