aegon-lv82 1.0.0

package/src/views/gen-ai/components/AIExport/docx-export.ts

@@ -0,0 +1,950 @@
+/**
+ * Word 導出（一期唯一底座：marked + JSZip/altChunk MHT）
+ *
+ * 版式對齊「行業調研 導出模板」：
+ * 頁眉「商密三级」藍字下劃線、頁腳 N / Total、
+ * 封面 / 小組成員 / 目錄 / N. 章節標題 / 參考文獻清單
+ */
+import { saveAs } from 'file-saver'
+import {
+  A4_CONTENT_WIDTH_PT,
+  A4_CONTENT_WIDTH_PX,
+  buildWordDocumentHtml,
+  exportHtmlToDocx,
+  prepareDomForDocx,
+  rasterizeImagesForDocx,
+  stripRefMarkersFromHtml,
+  stripWordPageBreaks,
+} from './docxExport'
+import { parseMarkdownToHtml } from './parseMarkdownHtml'
+
+/** 默認密級（與導出模板頁眉一致） */
+export const DOCX_HEADER_CLASSIFICATION = '商密三级'
+
+export type ExportReportMember = {
+  name: string
+  employeeId?: string
+  department?: string
+}
+
+export type ExportSourceRef = {
+  /** 資訊源檔名（可含副檔名） */
+  fileName: string
+  /** 頁碼（有則輸出「第 N 頁」） */
+  page?: number | string
+}
+
+export type ExportReportMeta = {
+  /** 報告標題（封面居中） */
+  reportName?: string
+  /** 年份，如 2026 */
+  year?: string
+  /** 生成人（封面「小組成員」與 Client Manager 續頁標題） */
+  members?: ExportReportMember[]
+  /** 資訊源檔名（文末參考文獻清單） */
+  sourceFileNames?: string[]
+  /** 章節「參考資料」明細（含頁碼時優先於 sourceFileNames） */
+  sourceRefs?: ExportSourceRef[]
+  /**
+   * 目錄章節標題；缺省時從正文 h1.docx-chapter-title / 章節 meta 推斷
+   */
+  tocTitles?: string[]
+}
+
+export type ExportMarkdownChapter = {
+  title?: string
+  content: string
+}
+
+export type ExportMarkdownToAiWordOptions = {
+  filename?: string
+  headerText?: string
+  showPageNumbers?: boolean
+  hideSelectors?: string[]
+  removeSelectors?: string[]
+  fallbackChapters?: ExportMarkdownChapter[]
+  reportMeta?: ExportReportMeta
+}
+
+export type ExportMarkdownToDocxWordOptions = {
+  filename?: string
+  headerText?: string
+  showPageNumbers?: boolean
+  reportMeta?: ExportReportMeta
+}
+
+const DEFAULT_HIDE_SELECTORS = ['.animate-pulse', '.report-tiptap-editor__toolbar']
+const DEFAULT_REMOVE_SELECTORS = [
+  '.report-tiptap-editor__toolbar',
+  '.no-export',
+  '.no-print',
+  '.report-chapter-review-popover',
+]
+
+/** 首頁「小組成員」超過此行數則餘下獨占續頁 */
+const COVER_MEMBER_LINES_MAX = 8
+
+function escapeHtmlText(value: string) {
+  return value
+    .replace(/&/g, '&')
+    .replace(/</g, '&lt;')
+    .replace(/>/g, '&gt;')
+    .replace(/"/g, '&quot;')
+}
+
+function normalizeFilename(filename: string) {
+  const raw = String(filename || '報告').trim() || '報告'
+  // Windows / 瀏覽器下載非法字符；保留中文與常見符號
+  const safe = raw
+    .replace(/[\\/:*?"<>|]/g, '_')
+    .replace(/[\u0000-\u001f\u007f]/g, '')
+    .replace(/\s+/g, ' ')
+    .trim()
+  const base = safe || '報告'
+  return base.toLowerCase().endsWith('.docx') ? base : `${base}.docx`
+}
+
+/** file-saver 失敗時（iframe / 部分內核）用 <a download> 兜底 */
+function downloadDocxBlob(blob: Blob, filename: string) {
+  const name = normalizeFilename(filename)
+  try {
+    saveAs(blob, name)
+    return
+  } catch (e) {
+    console.warn('file-saver saveAs failed, fallback to anchor download:', e)
+  }
+  const url = URL.createObjectURL(blob)
+  try {
+    const a = document.createElement('a')
+    a.href = url
+    a.download = name
+    a.rel = 'noopener'
+    a.style.display = 'none'
+    document.body.appendChild(a)
+    a.click()
+    a.remove()
+  } finally {
+    // 延遲 revoke，避免部分瀏覽器下載未開始就失效
+    window.setTimeout(() => URL.revokeObjectURL(url), 60_000)
+  }
+}
+
+async function assertHealthyDocxBlob(blob: Blob) {
+  if (!(blob instanceof Blob) || blob.size < 1024) {
+    throw new Error('Word 導出失敗：生成的文件異常過小，請確認已完整複製 AIExport 並安裝依賴')
+  }
+  // 校驗 ZIP 魔數，避免下載到「假 docx」
+  const head = new Uint8Array(await blob.slice(0, 4).arrayBuffer())
+  if (head[0] !== 0x50 || head[1] !== 0x4b) {
+    throw new Error('Word 導出失敗：文件不是有效的 docx（ZIP），請檢查 jszip')
+  }
+}
+
+function readableTextLength(htmlOrText: string) {
+  return htmlOrText
+    .replace(/<[^>]+>/g, ' ')
+    .replace(/&nbsp;/gi, ' ')
+    .replace(/\s+/g, '')
+    .trim().length
+}
+
+/** 導出正文去掉 [ref:…] 字面量（避免 Word 紅波浪 / 佔位下標） */
+function stripRefMarkersFromText(value: string) {
+  return stripRefMarkersFromHtml(value)
+}
+
+/**
+ * 導出不再把 [ref:…] 寫進正文 / 章節末「參考資料」
+ * （紅藍下標僅頁眉密級保留）
+ */
+function appendRefAppendixToChaptersHtml(bodyHtml: string, _meta: ExportReportMeta = {}): string {
+  return bodyHtml
+}
+
+/**
+ * Word 打開 HTML altChunk 時可靠的硬分頁。
+ * 只用單一 page-break（禁止雙重 br+p，否則 TOC 後會多一頁空白）；
+ * 禁止 section-break（否則會新建節、頁面大小不一致）。
+ */
+function pageBreakHtml() {
+  return `<p class="docx-page-break" style="page-break-before:always;mso-special-character:line-break;height:0;margin:0;padding:0;border:0;font-size:1pt;line-height:0;">&nbsp;</p>`
+}
+
+/** 離屏測量 HTML 塊高度（px），供封面/目錄實佔頁數 */
+function measureHtmlBlockHeightPx(html: string): number {
+  if (typeof document === 'undefined') return 0
+  const probe = document.createElement('div')
+  probe.style.cssText = [
+    'position:fixed',
+    'left:-100000px',
+    'top:0',
+    `width:${A4_CONTENT_WIDTH_PX}px`,
+    'visibility:hidden',
+    'pointer-events:none',
+    'box-sizing:border-box',
+    'font-family:"Times New Roman","PMingLiU","新細明體",SimSun,serif',
+    'font-size:11pt',
+    'line-height:1.5',
+    'color:#000',
+  ].join(';')
+  // 注入與導出一致的關鍵樣式，避免 class 未生效導致佔頁測偏
+  probe.innerHTML = `<style>
+    .docx-cover-spacer{margin:0;padding:0;font-size:1pt;line-height:260pt;}
+    .docx-cover-title{font-size:18pt;font-weight:bold;line-height:1.45;margin:0 0 8pt;text-align:center;}
+    .docx-cover-year{font-size:18pt;font-weight:bold;line-height:1.45;margin:0;text-align:center;}
+    .docx-toc-title{font-size:12pt;margin:12pt 0 18pt;}
+    .docx-toc-entry{font-size:12pt;line-height:1.5;margin:0 0 6pt;}
+    .docx-chapter-title{font-size:14pt;font-weight:bold;margin:14pt 0 8pt;line-height:1.45;}
+    p{font-size:11pt;line-height:1.5;margin:0 0 10pt;}
+    table{border-collapse:collapse;max-width:100%;width:100%;}
+  </style>${String(html || '').trim() || '<p>&nbsp;</p>'}`
+  document.body.appendChild(probe)
+  try {
+    return Math.max(1, Math.ceil(Math.max(probe.scrollHeight, probe.offsetHeight, 1)))
+  } finally {
+    probe.remove()
+  }
+}
+
+/** 按可印區高度把塊高換算為頁數（每段硬分頁後至少 1 頁） */
+function pageCountForHtmlBlock(html: string, pageH: number): number {
+  const h = measureHtmlBlockHeightPx(html)
+  return Math.max(1, Math.ceil(h / Math.max(1, pageH)))
+}
+
+/** 目錄條目過多時估頁（無 DOM 回退） */
+function estimateTocPageCount(titleCount: number) {
+  const n = Math.max(0, titleCount)
+  // 標題 ~36pt + 每條 ~24pt；A4 可印區約 650pt
+  const usedPt = 36 + Math.max(1, n) * 24
+  return Math.max(1, Math.ceil(usedPt / 650))
+}
+
+/** 去掉已有編號 / 方塊前綴，再套「N. 標題」 */
+export function formatChapterHeading(index: number, title: string) {
+  const bare = String(title || '')
+    .replace(/^■\s*/, '')
+    .replace(/^\d+\.\s*/, '')
+    .trim()
+  return `${index}. ${bare || `章節 ${index}`}`
+}
+
+function formatMemberLine(member: ExportReportMember) {
+  const dept = String(member.department ?? '').trim()
+  const emp = String(member.employeeId ?? '').trim()
+  const name = String(member.name ?? '').trim()
+  if (dept && emp && name) return `${dept} ${emp}: ${name}`
+  if (dept && name) return `${dept}: ${name}`
+  return name || emp || dept
+}
+
+function buildMemberLinesHtml(members: ExportReportMember[]) {
+  if (!members.length) return `<p class="docx-member-line">—</p>`
+  return members
+    .map((m) => `<p class="docx-member-line">${escapeHtmlText(formatMemberLine(m))}</p>`)
+    .join('')
+}
+
+/**
+ * 封面第 1 頁 1:1：
+ * - 上方居中標題
+ * - 中間大空白
+ * - 「小組成員：」+ 生成人落在首頁約下 1/3
+ * - 不超出首頁不新開頁；超出才續頁
+ */
+function buildFrontMatterParts(meta: ExportReportMeta): string[] {
+  const reportName = String(meta.reportName ?? '').trim() || '行業調研報告'
+  const year = String(meta.year ?? '').trim()
+  const yearLabel = year ? `(${year} 年)` : ''
+
+  // 去重生成人（同一人名只保留一行，同屬一塊）
+  const seen = new Set<string>()
+  const members: ExportReportMember[] = []
+  for (const m of meta.members ?? []) {
+    const line = formatMemberLine(m)
+    if (!line || seen.has(line)) continue
+    seen.add(line)
+    members.push(m)
+  }
+
+  // 標題分行：長標題可多行；年份單獨一行（對齊模板）
+  const titleLines = reportName
+    .split(/\n+/)
+    .map((s) => s.trim())
+    .filter(Boolean)
+  const titleHtml = titleLines
+    .map((line) => `<p class="docx-cover-title">${escapeHtmlText(line)}</p>`)
+    .join('')
+  const yearHtml = yearLabel ? `<p class="docx-cover-year">${escapeHtmlText(yearLabel)}</p>` : ''
+
+  const onCover = members.slice(0, COVER_MEMBER_LINES_MAX)
+  const overflow = members.slice(COVER_MEMBER_LINES_MAX)
+
+  // spacer 約 240pt：把「小組成員」推到首頁下 1/3（Word 對固定行高最穩）
+  const coverHtml = `
+<div class="docx-cover-page">
+  <div class="docx-cover">
+    ${titleHtml}
+    ${yearHtml}
+  </div>
+  <p class="docx-cover-spacer">&nbsp;</p>
+  <div class="docx-members-block">
+    <p class="docx-members-label"><u>小組成員</u>：</p>
+    ${buildMemberLinesHtml(onCover)}
+  </div>
+</div>
+`
+
+  const parts = [coverHtml]
+
+  if (overflow.length) {
+    parts.push(`
+<div class="docx-members-block docx-members-block--cont">
+  <p class="docx-members-label"><u>小組成員</u>：</p>
+  ${buildMemberLinesHtml(overflow)}
+</div>
+`)
+  }
+
+  return parts
+}
+
+/** 目錄頁：寫靜態實測頁碼（不用 PAGEREF——altChunk 下會失效並全部變成 1） */
+function buildTocHtml(titles: string[], pageNumbers: number[]) {
+  const entries = titles
+    .map((title, i) => {
+      const n = i + 1
+      const bare = String(title || '')
+        .replace(/^■\s*/, '')
+        .replace(/^\d+\.\s*/, '')
+        .trim()
+      const label = `${n}.  ${bare || `章節 ${n}`}`
+      const raw = Number(pageNumbers[i] ?? pageNumbers[pageNumbers.length - 1] ?? pageNumbers[0] ?? 1)
+      const page = Number.isFinite(raw) && raw >= 1 ? Math.floor(raw) : 1
+      return `<p class="docx-toc-entry" style="margin:0 0 6pt 0;padding:0;font-size:12pt;line-height:1.5;text-align:left;text-indent:0;font-family:'Times New Roman','PMingLiU','新細明體',serif;mso-ascii-font-family:'Times New Roman';mso-fareast-font-family:'新細明體';mso-hansi-font-family:'Times New Roman';tab-stops:right dotted ${A4_CONTENT_WIDTH_PT}pt;">${escapeHtmlText(label)}<span style="mso-tab-count:1">&nbsp;</span>${page}</p>`
+    })
+    .join('\n')
+
+  return `<div class="docx-toc-page-root">
+<p class="docx-toc-title" style="color:#2262E0;font-size:12pt;font-weight:normal;margin:12pt 0 18pt 0;padding:0;font-family:'FangSong','仿宋',serif;mso-ascii-font-family:'FangSong';mso-fareast-font-family:'仿宋';mso-hansi-font-family:'FangSong';text-decoration:none;border:none;">目錄</p>
+${entries || `<p class="docx-toc-entry" style="margin:0;font-size:12pt;font-family:'Times New Roman','PMingLiU','新細明體',serif;">（暫無章節）</p>`}
+</div>`
+}
+
+function buildReferencesAppendixHtml(meta: ExportReportMeta) {
+  const year =
+    String(meta.year ?? new Date().getFullYear()).trim() || String(new Date().getFullYear())
+  const names = (meta.sourceFileNames ?? []).map((n) => String(n || '').trim()).filter(Boolean)
+  const lines =
+    names.length > 0
+      ? names
+          .map(
+            (name) =>
+              `<p class="docx-ref-line" style="border:none;mso-border-alt:none;margin:0 0 4pt;">[發布機構]. (${escapeHtmlText(year)}).  《${escapeHtmlText(name)}》</p>`
+          )
+          .join('')
+      : `<p class="docx-ref-line" style="border:none;mso-border-alt:none;margin:0 0 4pt;">[發布機構]. (${escapeHtmlText(year)}).  《[上傳的文件]》</p>`
+
+  return `
+<div class="docx-refs-appendix" style="border:none;mso-border-alt:none;">
+<p class="docx-appendix-title" style="border:none;mso-border-alt:none;">参考文献清单 (References List):</p>
+${lines}
+</div>
+`
+}
+
+/** 正文章節加書籤，供目錄 PAGEREF（掛在標題上：id + name，兼容 Word HTML） */
+function annotateChapterBookmarks(bodyHtml: string): string {
+  if (typeof document === 'undefined') return bodyHtml
+  const host = document.createElement('div')
+  host.innerHTML = bodyHtml
+  const ensureBookmark = (parent: Element, index: number) => {
+    const name = `_Ch${index}`
+    const heading =
+      parent.matches?.('h1, h2')
+        ? parent
+        : parent.querySelector('h1.docx-chapter-title, h1, h2') || parent
+
+    if (heading instanceof HTMLElement) {
+      heading.setAttribute('id', name)
+    }
+
+    let anchor =
+      (heading.querySelector(`:scope > a[name="${name}"]`) as HTMLElement | null) ||
+      (heading.querySelector(`a[name="${name}"]`) as HTMLElement | null)
+    if (!anchor) {
+      anchor = document.createElement('a')
+      heading.insertBefore(anchor, heading.firstChild)
+    }
+    anchor.setAttribute('name', name)
+    anchor.removeAttribute('id')
+    anchor.removeAttribute('href')
+  }
+
+  const sections = Array.from(host.querySelectorAll('.docx-export-chapter'))
+  if (sections.length) {
+    sections.forEach((section, i) => ensureBookmark(section, i + 1))
+    return host.innerHTML
+  }
+
+  const headings = Array.from(host.querySelectorAll('h1.docx-chapter-title'))
+  headings.forEach((h, i) => ensureBookmark(h, i + 1))
+  return host.innerHTML
+}
+
+function extractChapterSectionHtmls(bodyHtml: string, chapterCount: number): string[] {
+  const host = document.createElement('div')
+  host.innerHTML = bodyHtml
+  const sections = Array.from(host.querySelectorAll('.docx-export-chapter'))
+  if (sections.length) {
+    return sections.map((el) => (el as HTMLElement).outerHTML)
+  }
+  const html = bodyHtml
+  const parts = html.split(/(?=<h1\b[^>]*class="[^"]*docx-chapter-title)/i).filter(Boolean)
+  if (parts.length >= 2) return parts
+  return Array.from({ length: Math.max(chapterCount, 1) }, (_, i) => (i === 0 ? bodyHtml : ''))
+}
+
+/**
+ * A4 可打印正文區高度（上下頁邊距各 2cm），與 @page / sectPr 一致。
+ * 勿再乘經驗係數：偏小會把目錄章頁碼系統性抬高。
+ */
+function getExportPageContentHeightPx() {
+  return Math.max(1, Math.round(((297 - 40) / 25.4) * 96)) // ≈970px @96dpi
+}
+
+/** 封面模板固定佔 1 頁；其餘硬分頁塊按實測高度折頁 */
+function pageCountForFrontPart(html: string, pageH: number): number {
+  if (/docx-cover-page/i.test(html)) return 1
+  return pageCountForHtmlBlock(html, pageH)
+}
+
+/**
+ * 按正文連續流測量每章頂部位置 → 起始頁碼（短章可同頁同碼）。
+ * 必須在表格已 rebuild 後的 HTML 上測，否則頁碼會偏大/偏小。
+ */
+function computeChapterStartPages(sectionHtmls: string[], bodyStartPage: number): number[] {
+  const start = Math.max(1, Math.floor(Number(bodyStartPage)) || 1)
+  if (!sectionHtmls.length) return [start]
+
+  const measured = measureChapterStartOffsetsPx(sectionHtmls)
+  let pages: number[]
+  if (measured && measured.length === sectionHtmls.length) {
+    const pageH = getExportPageContentHeightPx()
+    pages = measured.map((y) => start + Math.floor(Math.max(0, y) / pageH))
+  } else {
+    const CHARS_PER_PAGE = 880
+    const HEADING_OVERHEAD = 36
+    pages = []
+    let usedChars = 0
+    for (let i = 0; i < sectionHtmls.length; i++) {
+      pages.push(start + Math.floor(usedChars / CHARS_PER_PAGE))
+      const html = sectionHtmls[i] || ''
+      const textLen = readableTextLength(html)
+      const tableBonus = (html.match(/<table\b/gi) || []).length * 220
+      const imgBonus = (html.match(/<img\b/gi) || []).length * 160
+      usedChars += Math.max(HEADING_OVERHEAD, textLen + HEADING_OVERHEAD + tableBonus + imgBonus)
+    }
+  }
+
+  for (let i = 1; i < pages.length; i++) {
+    const prev = pages[i - 1] ?? start
+    const cur = pages[i] ?? prev
+    pages[i] = Math.max(cur, prev)
+  }
+  return pages.map((p) => {
+    const v = Math.floor(Number(p))
+    return Number.isFinite(v) && v >= 1 ? v : start
+  })
+}
+
+/** 離屏連續排版，返回每章相對正文頂的 offsetTop（px） */
+function measureChapterStartOffsetsPx(sectionHtmls: string[]): number[] | null {
+  if (typeof document === 'undefined') return null
+
+  const probe = document.createElement('div')
+  probe.setAttribute('data-docx-toc-measure', 'true')
+  probe.style.cssText = [
+    'position:fixed',
+    'left:-100000px',
+    'top:0',
+    `width:${A4_CONTENT_WIDTH_PX}px`,
+    'visibility:hidden',
+    'pointer-events:none',
+    'box-sizing:border-box',
+    'font-family:"Times New Roman","PMingLiU","新細明體",SimSun,serif',
+    'font-size:11pt',
+    'line-height:1.5',
+    'color:#000',
+  ].join(';')
+
+  const flow = document.createElement('div')
+  flow.className = 'docx-chapters-body'
+  flow.style.cssText = 'width:100%;box-sizing:border-box;margin:0;padding:0;'
+
+  sectionHtmls.forEach((html, i) => {
+    const wrap = document.createElement('div')
+    wrap.className = 'docx-export-chapter'
+    wrap.setAttribute('data-toc-measure-ch', String(i))
+    // sectionHtmls 可能是 outerHTML（已含 chapter 殼）或內層 HTML
+    const trimmed = String(html || '').trim()
+    if (/class=["'][^"']*docx-export-chapter/i.test(trimmed)) {
+      wrap.innerHTML = trimmed
+      // 若誤套雙層，取內層章節
+      const inner = wrap.querySelector(':scope > .docx-export-chapter') || wrap
+      if (inner !== wrap) {
+        wrap.innerHTML = inner.innerHTML
+      }
+    } else {
+      wrap.innerHTML =
+        trimmed ||
+        '<h1 class="docx-chapter-title" style="font-size:14pt;font-weight:bold;margin:14pt 0 8pt;">&nbsp;</h1><p style="margin:0 0 10pt;">&nbsp;</p>'
+    }
+    wrap.querySelectorAll('h1.docx-chapter-title, .docx-chapter-title').forEach((node) => {
+      const el = node as HTMLElement
+      el.style.fontSize = '14pt'
+      el.style.fontWeight = 'bold'
+      el.style.margin = '14pt 0 8pt'
+      el.style.lineHeight = '1.45'
+    })
+    wrap.querySelectorAll('p').forEach((node) => {
+      const el = node as HTMLElement
+      if (!el.style.margin) el.style.margin = '0 0 10pt'
+      if (!el.style.fontSize) el.style.fontSize = '11pt'
+      if (!el.style.lineHeight) el.style.lineHeight = '1.5'
+    })
+    wrap.querySelectorAll('table').forEach((node) => {
+      const el = node as HTMLElement
+      el.style.maxWidth = '100%'
+      el.style.borderCollapse = 'collapse'
+      el.style.height = 'auto'
+    })
+    flow.appendChild(wrap)
+  })
+
+  probe.appendChild(flow)
+  // 與導出正文樣式對齊，避免測頁與 Word 版式差一截
+  const style = document.createElement('style')
+  style.textContent = `
+    .docx-chapters-body .docx-chapter-title{font-size:14pt;font-weight:bold;margin:14pt 0 8pt;line-height:1.45;}
+    .docx-chapters-body p{font-size:11pt;line-height:1.5;margin:0 0 10pt;}
+    .docx-chapters-body table{border-collapse:collapse;max-width:100%;width:100%;}
+    .docx-chapters-body td,.docx-chapters-body th{border:1px solid #000;padding:4pt 6pt;font-size:11pt;vertical-align:top;}
+  `
+  probe.appendChild(style)
+  document.body.appendChild(probe)
+
+  try {
+    // 累計 offsetHeight，避免 getBoundingClientRect 全 0 導致目錄頁碼全相同
+    void probe.offsetHeight
+    const offsets: number[] = []
+    let y = 0
+    for (let i = 0; i < sectionHtmls.length; i++) {
+      const el = flow.querySelector(`[data-toc-measure-ch="${i}"]`) as HTMLElement | null
+      offsets.push(y)
+      const h = el ? Math.max(el.offsetHeight, el.scrollHeight, 0) : 0
+      y += Math.max(h, 24)
+    }
+    return offsets
+  } finally {
+    probe.remove()
+  }
+}
+
+/** 將正文章節標題統一為 N. …（h1，供 Word 導航窗格） */
+export function normalizeChapterHeadingsInHtml(bodyHtml: string, titles?: string[]): string {
+  const host = document.createElement('div')
+  host.innerHTML = bodyHtml
+
+  const sections = Array.from(host.querySelectorAll('.docx-export-chapter'))
+  if (sections.length) {
+    sections.forEach((section, i) => {
+      const heading = section.querySelector('h1, h2')
+      const fallbackTitle = titles?.[i] ?? `章節 ${i + 1}`
+      const raw = heading?.textContent?.trim() || fallbackTitle
+      const h1 = document.createElement('h1')
+      h1.className = 'docx-chapter-title'
+      h1.textContent = formatChapterHeading(i + 1, raw)
+      if (heading) heading.replaceWith(h1)
+      else section.insertBefore(h1, section.firstChild)
+    })
+    return host.innerHTML
+  }
+
+  const tops = Array.from(host.children).filter(
+    (el) => el instanceof HTMLElement && /^(H1|H2)$/.test(el.tagName)
+  ) as HTMLElement[]
+  if (tops.length) {
+    tops.forEach((el, i) => {
+      const h1 = document.createElement('h1')
+      h1.className = 'docx-chapter-title'
+      h1.textContent = formatChapterHeading(i + 1, el.textContent || titles?.[i] || '')
+      el.replaceWith(h1)
+    })
+  }
+
+  return host.innerHTML
+}
+
+function resolveTocTitles(meta: ExportReportMeta, bodyHtml: string): string[] {
+  if (meta.tocTitles?.length) {
+    return meta.tocTitles.map((t) => String(t || '').trim()).filter(Boolean)
+  }
+  const host = document.createElement('div')
+  host.innerHTML = bodyHtml
+  return Array.from(host.querySelectorAll('h1.docx-chapter-title, h1, h2'))
+    .map((el) => el.textContent?.trim() || '')
+    .filter(Boolean)
+}
+
+/** 目錄標題與正文各章順序對齊（頁碼按章測量，標題也必須同源） */
+function alignTocTitlesWithSections(sectionHtmls: string[], fallbackTitles: string[]): string[] {
+  return sectionHtmls.map((html, i) => {
+    if (typeof document !== 'undefined') {
+      const host = document.createElement('div')
+      host.innerHTML = html
+      const heading = host.querySelector('h1.docx-chapter-title, h1, h2')
+      const fromBody = heading?.textContent?.trim()
+      if (fromBody) return fromBody
+    }
+    return String(fallbackTitles[i] || '').trim() || `章節 ${i + 1}`
+  })
+}
+
+/**
+ * 拆成多個 HTML 片段，由硬分頁切開：
+ * 封面… →【目錄獨頁】→【正文連續排版，章間不分頁】→ 參考文獻
+ *
+ * 目錄頁碼：封面(1) + 目錄佔頁 + 正文連續流實測各章起始頁（靜態數字，不用 PAGEREF）。
+ */
+export function buildIndustryResearchExportParts(
+  bodyHtml: string,
+  meta: ExportReportMeta = {}
+): { parts: string[]; chapterStartPage: number } {
+  const normalizedBody = normalizeChapterHeadingsInHtml(bodyHtml, meta.tocTitles)
+  const bodyWithRefs = appendRefAppendixToChaptersHtml(normalizedBody, meta)
+  const tocTitlesFallback = resolveTocTitles(meta, bodyWithRefs)
+  const frontParts = buildFrontMatterParts(meta)
+  const pageH = getExportPageContentHeightPx()
+
+  // 先按導出規則重建表格，再連續流測頁
+  let measureHtml = bodyWithRefs
+  if (typeof document !== 'undefined') {
+    const host = document.createElement('div')
+    host.className = 'docx-page-root'
+    host.innerHTML = `<div class="docx-chapters-body">${bodyWithRefs}</div>`
+    try {
+      prepareDomForDocx(host, true)
+      measureHtml =
+        host.querySelector('.docx-chapters-body')?.innerHTML?.trim() || host.innerHTML
+    } catch {
+      measureHtml = bodyWithRefs
+    }
+  }
+
+  // 書籤可保留（不影響目錄數字）；目錄只用靜態實測頁碼
+  measureHtml = annotateChapterBookmarks(measureHtml)
+
+  const sectionHtmls = extractChapterSectionHtmls(measureHtml, tocTitlesFallback.length || 1)
+  const tocTitles = alignTocTitlesWithSections(sectionHtmls, tocTitlesFallback)
+
+  let frontPageCount = 0
+  if (typeof document !== 'undefined') {
+    frontPageCount = frontParts.reduce((sum, part) => sum + pageCountForFrontPart(part, pageH), 0)
+  } else {
+    frontPageCount = Math.max(1, frontParts.length)
+  }
+  frontPageCount = Math.max(1, frontPageCount)
+
+  let tocPageCount =
+    typeof document !== 'undefined'
+      ? pageCountForHtmlBlock(
+          buildTocHtml(
+            tocTitles,
+            tocTitles.map((_, i) => frontPageCount + estimateTocPageCount(tocTitles.length) + 1 + i)
+          ),
+          pageH
+        )
+      : estimateTocPageCount(tocTitles.length)
+  tocPageCount = Math.max(1, tocPageCount)
+
+  let bodyStartPage = frontPageCount + tocPageCount + 1
+  let pageNumbers = computeChapterStartPages(sectionHtmls, bodyStartPage)
+
+  // 目錄條目數與章節數對齊：缺的頁碼用最後一章頁碼補上
+  while (pageNumbers.length < tocTitles.length) {
+    pageNumbers.push(pageNumbers[pageNumbers.length - 1] ?? bodyStartPage)
+  }
+  pageNumbers = pageNumbers.slice(0, tocTitles.length)
+
+  let tocPart = buildTocHtml(tocTitles, pageNumbers)
+
+  if (typeof document !== 'undefined') {
+    const tocPagesFinal = pageCountForHtmlBlock(tocPart, pageH)
+    if (tocPagesFinal !== tocPageCount) {
+      tocPageCount = Math.max(1, tocPagesFinal)
+      bodyStartPage = frontPageCount + tocPageCount + 1
+      pageNumbers = computeChapterStartPages(sectionHtmls, bodyStartPage)
+      while (pageNumbers.length < tocTitles.length) {
+        pageNumbers.push(pageNumbers[pageNumbers.length - 1] ?? bodyStartPage)
+      }
+      pageNumbers = pageNumbers.slice(0, tocTitles.length)
+      tocPart = buildTocHtml(tocTitles, pageNumbers)
+    }
+  }
+
+  const bodyPart = `<div class="docx-chapters-body">${measureHtml}</div>`
+
+  return {
+    chapterStartPage: bodyStartPage,
+    parts: [...frontParts, tocPart, bodyPart, buildReferencesAppendixHtml(meta)],
+  }
+}
+
+/** @deprecated 請用 buildIndustryResearchExportParts（OOXML 硬分頁） */
+export function wrapIndustryResearchExportHtml(
+  bodyHtml: string,
+  meta: ExportReportMeta = {}
+): string {
+  return buildIndustryResearchExportParts(bodyHtml, meta).parts.join(pageBreakHtml())
+}
+
+function buildBodyHtmlFromMarkdown(
+  input: string | ExportMarkdownChapter[],
+  tocTitles?: string[],
+  meta: ExportReportMeta = {}
+): string {
+  if (typeof input === 'string') {
+    const html = parseMarkdownToHtml(input)
+    return appendRefAppendixToChaptersHtml(`<div class="docx-export-chapter">${html}</div>`, meta)
+  }
+
+  return input
+    .map((chapter, i) => {
+      const title = String(chapter.title ?? tocTitles?.[i] ?? '').trim()
+      const titleHtml = title
+        ? `<h1 class="docx-chapter-title">${escapeHtmlText(formatChapterHeading(i + 1, title))}</h1>`
+        : ''
+      const rawContent = stripRefMarkersFromText(String(chapter.content ?? ''))
+      let bodyHtml = parseMarkdownToHtml(rawContent)
+      bodyHtml = stripRefMarkersFromText(bodyHtml)
+      // 正文若已含同標題 h1/h2，去掉避免重複
+      if (title) {
+        const re = new RegExp(
+          `^\\s*<h[12][^>]*>\\s*(?:■\\s*)?(?:\\d+\\.\\s*)?${title.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}\\s*</h[12]>`,
+          'i'
+        )
+        bodyHtml = bodyHtml.replace(re, '')
+      }
+      // 導出不插入章節末 [ref:…]「參考資料」佔位
+      return `<div class="docx-export-chapter">${titleHtml}${bodyHtml}</div>`
+    })
+    .filter(Boolean)
+    .join('\n')
+}
+
+/** 優先抽取 TipTap/ProseMirror 正文 */
+function buildExportRootFromElement(element: HTMLElement): HTMLElement {
+  const proses = Array.from(
+    element.querySelectorAll('.ProseMirror, .report-tiptap-editor__prose')
+  ) as HTMLElement[]
+
+  if (proses.length) {
+    const host = document.createElement('div')
+    host.className = 'docx-page-root docx-export-from-prose'
+    for (const prose of proses) {
+      const html = prose.innerHTML.trim()
+      if (!html || readableTextLength(html) < 1) continue
+      const section = document.createElement('div')
+      section.className = 'docx-export-chapter'
+      section.innerHTML = html
+      host.appendChild(section)
+    }
+    if (host.childElementCount > 0) return host
+  }
+
+  return element.cloneNode(true) as HTMLElement
+}
+
+async function finalizeAndDownload(
+  bodyInnerHtml: string,
+  options: {
+    filename: string
+    headerText: string
+    showPageNumbers: boolean
+    reportMeta?: ExportReportMeta
+  }
+): Promise<Blob> {
+  const { parts } = buildIndustryResearchExportParts(bodyInnerHtml, options.reportMeta ?? {})
+  // 封面 / 目錄+正文 / 參考文獻：在 HTML 內分頁，再打成【單一】完整 HTML → 單一 afchunk.html
+  // 禁止多段 altChunk / 禁止 MHT（否則遷移環境易出現 --------tDocumentpart--）
+  const joinedBody = parts
+    .map((part) => String(part || '').trim())
+    .filter(Boolean)
+    .join(pageBreakHtml())
+  if (!joinedBody || readableTextLength(joinedBody) < 1) {
+    throw new Error('Word 導出失敗：組裝後無可打包內容')
+  }
+  const fullHtml = buildWordDocumentHtml(joinedBody)
+  const blob = await exportHtmlToDocx(fullHtml, {
+    headerFooter: {
+      headerText: options.headerText,
+      showPageNumbers: options.showPageNumbers,
+    },
+  })
+  await assertHealthyDocxBlob(blob)
+  downloadDocxBlob(blob, options.filename)
+  return blob
+}
+
+/**
+ * 將頁面中的報告內容 DOM 導出為 Word（.docx）
+ */
+export async function exportMarkdownToAiWord(
+  element: HTMLElement,
+  options: ExportMarkdownToAiWordOptions = {}
+): Promise<Blob> {
+  const {
+    filename = '報告.docx',
+    headerText = DOCX_HEADER_CLASSIFICATION,
+    showPageNumbers = true,
+    hideSelectors = DEFAULT_HIDE_SELECTORS,
+    removeSelectors = DEFAULT_REMOVE_SELECTORS,
+    fallbackChapters,
+    reportMeta,
+  } = options
+
+  const hidden: Array<{ el: HTMLElement; display: string }> = []
+  for (const selector of hideSelectors) {
+    element.querySelectorAll(selector).forEach((node) => {
+      const el = node as HTMLElement
+      hidden.push({ el, display: el.style.display })
+      el.style.display = 'none'
+    })
+  }
+
+  try {
+    const cloneContainer = buildExportRootFromElement(element)
+    for (const selector of removeSelectors) {
+      cloneContainer.querySelectorAll(selector).forEach((node) => node.remove())
+    }
+
+    const domTextLen = readableTextLength(cloneContainer.innerHTML)
+    const fallbackTextLen = Array.isArray(fallbackChapters)
+      ? fallbackChapters.reduce(
+          (sum, chapter) =>
+            sum +
+            readableTextLength(String(chapter.title ?? '')) +
+            readableTextLength(String(chapter.content ?? '')),
+          0
+        )
+      : 0
+
+    // 權威章節正文明顯長於 DOM（打字機未完成 / 流式佔位）→ 走 Markdown 導出
+    if (
+      fallbackChapters?.length &&
+      fallbackTextLen > 0 &&
+      (domTextLen < 1 || domTextLen < Math.max(20, Math.floor(fallbackTextLen * 0.5)))
+    ) {
+      return exportMarkdownToDocxWord(fallbackChapters, {
+        filename,
+        headerText,
+        showPageNumbers,
+        reportMeta,
+      })
+    }
+    if (domTextLen < 1) {
+      throw new Error('暫無可導出的頁面正文')
+    }
+
+    stripWordPageBreaks(cloneContainer)
+    try {
+      await rasterizeImagesForDocx(cloneContainer, element)
+    } catch (e) {
+      console.warn('rasterizeImagesForDocx soft-fail:', e)
+    }
+    await new Promise((resolve) => setTimeout(resolve, 200))
+
+    const prepared = prepareDomForDocx(cloneContainer, true)
+    if (readableTextLength(prepared.innerHTML) < 1) {
+      if (fallbackChapters?.length) {
+        return exportMarkdownToDocxWord(fallbackChapters, {
+          filename,
+          headerText,
+          showPageNumbers,
+          reportMeta,
+        })
+      }
+      throw new Error('暫無可導出的頁面正文')
+    }
+
+    return finalizeAndDownload(prepared.innerHTML, {
+      filename,
+      headerText,
+      showPageNumbers,
+      reportMeta,
+    })
+  } finally {
+    for (const { el, display } of hidden) {
+      el.style.display = display
+    }
+  }
+}
+
+/**
+ * Markdown → 同源 marked → 模板殼 → docx
+ */
+export async function exportMarkdownToDocxWord(
+  markdownOrChapters: string | ExportMarkdownChapter[],
+  options: ExportMarkdownToDocxWordOptions = {}
+): Promise<Blob> {
+  const {
+    filename = '報告.docx',
+    headerText = DOCX_HEADER_CLASSIFICATION,
+    showPageNumbers = true,
+    reportMeta,
+  } = options
+
+  const tocTitles =
+    reportMeta?.tocTitles ??
+    (Array.isArray(markdownOrChapters)
+      ? markdownOrChapters.map((c) => String(c.title ?? '').trim())
+      : undefined)
+
+  const bodyHtml = buildBodyHtmlFromMarkdown(markdownOrChapters, tocTitles, reportMeta ?? {}).trim()
+  if (!bodyHtml || readableTextLength(bodyHtml) < 1) {
+    throw new Error('暫無可導出的 Markdown 內容')
+  }
+
+  const host = document.createElement('div')
+  host.setAttribute('data-docx-export', 'markdown-transform')
+  host.className = 'docx-page-root'
+  host.style.cssText =
+    'position:fixed;left:-10000px;top:0;width:794px;opacity:0;pointer-events:none;'
+  host.innerHTML = bodyHtml
+  document.body.appendChild(host)
+
+  try {
+    stripWordPageBreaks(host)
+    try {
+      await rasterizeImagesForDocx(host, host)
+    } catch (e) {
+      // 圖片柵格化失敗不應阻斷正文導出
+      console.warn('rasterizeImagesForDocx soft-fail:', e)
+    }
+    await new Promise((resolve) => setTimeout(resolve, 200))
+
+    const prepared = prepareDomForDocx(host, true)
+    return finalizeAndDownload(prepared.innerHTML, {
+      filename,
+      headerText,
+      showPageNumbers,
+      reportMeta: {
+        ...reportMeta,
+        tocTitles: tocTitles?.filter(Boolean),
+      },
+    })
+  } finally {
+    host.remove()
+  }
+}