docrev 0.9.5 → 0.9.6

package/dist/lib/import.js

@@ -1,522 +1,26 @@
 /**
  * Import functionality - convert Word docs to annotated Markdown
+ *
+ * Orchestration workflows + re-exports from extraction/diff/restore modules
  */
 import * as fs from 'fs';
 import * as path from 'path';
-import { diffWords } from 'diff';
 import { stripAnnotations } from './annotations.js';
 import { readImageRegistry } from './image-registry.js';
 import { exec } from 'child_process';
 import { promisify } from 'util';
-import { extractMarkdownPrefix, protectAnchors, restoreAnchors, protectCrossrefs, restoreCrossrefs, protectMath, restoreMath, replaceRenderedMath, protectCitations, restoreCitations, replaceRenderedCitations, protectImages, restoreImages, matchWordImagesToOriginal, protectTables, restoreTables, } from './protect-restore.js';
-import { normalizeWhitespace } from './utils.js';
+// Import from split modules
+import { extractFromWord, } from './word-extraction.js';
+import { generateSmartDiff, cleanupAnnotations, fixCitationAnnotations, } from './diff-engine.js';
+import { restoreCrossrefFromWord, restoreImagesFromRegistry, convertVisibleComments, } from './restore-references.js';
+// Re-export everything so existing imports from './import.js' still work
+export { extractFromWord, extractWordComments, extractCommentAnchors, extractWordTables, } from './word-extraction.js';
+export { generateSmartDiff, generateAnnotatedDiff, cleanupAnnotations, fixCitationAnnotations, } from './diff-engine.js';
+export { restoreCrossrefFromWord, restoreImagesFromRegistry, parseVisibleComments, convertVisibleComments, } from './restore-references.js';
 const execAsync = promisify(exec);
 // ============================================
 // Functions
 // ============================================
-/**
- * Extract comments directly from Word docx comments.xml
- */
-export async function extractWordComments(docxPath) {
-    const AdmZip = (await import('adm-zip')).default;
-    const { parseStringPromise } = await import('xml2js');
-    const comments = [];
-    // Validate file exists
-    if (!fs.existsSync(docxPath)) {
-        throw new Error(`File not found: ${docxPath}`);
-    }
-    try {
-        let zip;
-        try {
-            zip = new AdmZip(docxPath);
-        }
-        catch (err) {
-            throw new Error(`Invalid Word document (not a valid .docx file): ${err.message}`);
-        }
-        const commentsEntry = zip.getEntry('word/comments.xml');
-        if (!commentsEntry) {
-            return comments;
-        }
-        let commentsXml;
-        try {
-            commentsXml = commentsEntry.getData().toString('utf8');
-        }
-        catch (err) {
-            throw new Error(`Failed to read comments from document: ${err.message}`);
-        }
-        const parsed = await parseStringPromise(commentsXml, { explicitArray: false });
-        const ns = 'w:';
-        const commentsRoot = parsed['w:comments'];
-        if (!commentsRoot || !commentsRoot['w:comment']) {
-            return comments;
-        }
-        // Ensure it's an array
-        const commentNodes = Array.isArray(commentsRoot['w:comment'])
-            ? commentsRoot['w:comment']
-            : [commentsRoot['w:comment']];
-        for (const comment of commentNodes) {
-            const id = comment.$?.['w:id'] || '';
-            const author = comment.$?.['w:author'] || 'Unknown';
-            const date = comment.$?.['w:date'] || '';
-            // Extract text from nested w:p/w:r/w:t elements
-            let text = '';
-            const extractText = (node) => {
-                if (!node)
-                    return;
-                if (typeof node === 'string') {
-                    text += node;
-                    return;
-                }
-                if (node['w:t']) {
-                    const t = node['w:t'];
-                    text += typeof t === 'string' ? t : (t._ || t);
-                }
-                if (node['w:r']) {
-                    const runs = Array.isArray(node['w:r']) ? node['w:r'] : [node['w:r']];
-                    runs.forEach(extractText);
-                }
-                if (node['w:p']) {
-                    const paras = Array.isArray(node['w:p']) ? node['w:p'] : [node['w:p']];
-                    paras.forEach(extractText);
-                }
-            };
-            extractText(comment);
-            comments.push({ id, author, date: date.slice(0, 10), text: text.trim() });
-        }
-    }
-    catch (err) {
-        // Re-throw with more context if it's already an Error we created
-        if (err.message.includes('Invalid Word document') || err.message.includes('File not found')) {
-            throw err;
-        }
-        throw new Error(`Error extracting comments from ${path.basename(docxPath)}: ${err.message}`);
-    }
-    return comments;
-}
-/**
- * Extract comment anchor texts from document.xml with surrounding context
- * Returns map of comment ID -> {anchor, before, after, docPosition, isEmpty} for better matching
- * Also returns fullDocText for section boundary matching
- */
-export async function extractCommentAnchors(docxPath) {
-    const AdmZip = (await import('adm-zip')).default;
-    const anchors = new Map();
-    let fullDocText = '';
-    try {
-        const zip = new AdmZip(docxPath);
-        const docEntry = zip.getEntry('word/document.xml');
-        if (!docEntry) {
-            return { anchors, fullDocText };
-        }
-        const docXml = docEntry.getData().toString('utf8');
-        // ========================================
-        // STEP 1: Build text position mapping
-        // ========================================
-        const textNodePattern = /<w:t[^>]*>([^<]*)<\/w:t>/g;
-        const textNodes = [];
-        let textPosition = 0;
-        let nodeMatch;
-        while ((nodeMatch = textNodePattern.exec(docXml)) !== null) {
-            const rawText = nodeMatch[1] ?? '';
-            const decodedText = decodeXmlEntities(rawText);
-            textNodes.push({
-                xmlStart: nodeMatch.index,
-                xmlEnd: nodeMatch.index + nodeMatch[0].length,
-                textStart: textPosition,
-                textEnd: textPosition + decodedText.length,
-                text: decodedText
-            });
-            textPosition += decodedText.length;
-        }
-        fullDocText = textNodes.map(n => n.text).join('');
-        // Helper: convert XML position to text position
-        function xmlPosToTextPos(xmlPos) {
-            for (let i = 0; i < textNodes.length; i++) {
-                const node = textNodes[i];
-                if (!node)
-                    continue;
-                if (xmlPos >= node.xmlStart && xmlPos < node.xmlEnd) {
-                    return node.textStart;
-                }
-                if (xmlPos < node.xmlStart) {
-                    return node.textStart;
-                }
-            }
-            const lastNode = textNodes[textNodes.length - 1];
-            return lastNode ? lastNode.textEnd : 0;
-        }
-        // Helper: extract context before a position
-        function getContextBefore(position, maxLength = 150) {
-            const beforeText = fullDocText.slice(Math.max(0, position - maxLength), position);
-            const sentenceStart = beforeText.search(/[.!?]\s+[A-Z][^.!?]*$/);
-            return sentenceStart >= 0
-                ? beforeText.slice(sentenceStart + 2).trim()
-                : beforeText.slice(-80).trim();
-        }
-        // Helper: extract context after a position
-        function getContextAfter(position, maxLength = 150) {
-            const afterText = fullDocText.slice(position, position + maxLength);
-            const sentenceEnd = afterText.search(/[.!?]\s/);
-            return sentenceEnd >= 0
-                ? afterText.slice(0, sentenceEnd + 1).trim()
-                : afterText.slice(0, 80).trim();
-        }
-        // ========================================
-        // STEP 2: Collect all start/end markers separately
-        // ========================================
-        const startPattern = /<w:commentRangeStart[^>]*w:id="(\d+)"[^>]*\/?>/g;
-        const endPattern = /<w:commentRangeEnd[^>]*w:id="(\d+)"[^>]*\/?>/g;
-        const starts = new Map(); // id -> position after start tag
-        const ends = new Map(); // id -> position before end tag
-        let match;
-        while ((match = startPattern.exec(docXml)) !== null) {
-            const id = match[1];
-            if (!starts.has(id)) {
-                starts.set(id, match.index + match[0].length);
-            }
-        }
-        while ((match = endPattern.exec(docXml)) !== null) {
-            const id = match[1];
-            if (!ends.has(id)) {
-                ends.set(id, match.index);
-            }
-        }
-        // ========================================
-        // STEP 3: Process each comment range by ID
-        // ========================================
-        for (const [id, startXmlPos] of starts) {
-            const endXmlPos = ends.get(id);
-            // Missing end marker - skip with warning
-            if (endXmlPos === undefined) {
-                console.warn(`Comment ${id}: missing end marker`);
-                continue;
-            }
-            // Calculate text position
-            const docPosition = xmlPosToTextPos(startXmlPos);
-            // Handle empty or inverted ranges
-            if (endXmlPos <= startXmlPos) {
-                anchors.set(id, {
-                    anchor: '',
-                    before: getContextBefore(docPosition),
-                    after: getContextAfter(docPosition),
-                    docPosition,
-                    docLength: fullDocText.length,
-                    isEmpty: true
-                });
-                continue;
-            }
-            // Extract XML segment between markers
-            const segment = docXml.slice(startXmlPos, endXmlPos);
-            // Extract text from w:t (regular) AND w:delText (deleted text in track changes)
-            const textInRangePattern = /<w:t[^>]*>([^<]*)<\/w:t>|<w:delText[^>]*>([^<]*)<\/w:delText>/g;
-            let anchorText = '';
-            let tm;
-            while ((tm = textInRangePattern.exec(segment)) !== null) {
-                anchorText += tm[1] || tm[2] || '';
-            }
-            anchorText = decodeXmlEntities(anchorText);
-            // Get context
-            const anchorLength = anchorText.length;
-            const before = getContextBefore(docPosition);
-            const after = getContextAfter(docPosition + anchorLength);
-            // ALWAYS add entry (even if anchor is empty)
-            anchors.set(id, {
-                anchor: anchorText.trim(),
-                before,
-                after,
-                docPosition,
-                docLength: fullDocText.length,
-                isEmpty: !anchorText.trim()
-            });
-        }
-    }
-    catch (err) {
-        console.error('Error extracting comment anchors:', err.message);
-        return { anchors, fullDocText: '' };
-    }
-    return { anchors, fullDocText };
-}
-/**
- * Decode XML entities in text
- */
-function decodeXmlEntities(text) {
-    return text
-        .replace(/&amp;/g, '&')
-        .replace(/&lt;/g, '<')
-        .replace(/&gt;/g, '>')
-        .replace(/&quot;/g, '"')
-        .replace(/&apos;/g, "'")
-        .replace(/&#(\d+);/g, (_, code) => String.fromCharCode(parseInt(code, 10)))
-        .replace(/&#x([0-9a-fA-F]+);/g, (_, code) => String.fromCharCode(parseInt(code, 16)));
-}
-/**
- * Extract text content from a Word XML cell
- */
-function extractCellText(cellXml) {
-    const parts = [];
-    // Check for OMML math - replace with [math] placeholder
-    if (cellXml.includes('<m:oMath')) {
-        // Try to extract the text representation of math
-        const mathTextMatches = cellXml.match(/<m:t>([^<]*)<\/m:t>/g) || [];
-        if (mathTextMatches.length > 0) {
-            const mathText = mathTextMatches.map((t) => t.replace(/<[^>]+>/g, '')).join('');
-            parts.push(mathText);
-        }
-        else {
-            parts.push('[math]');
-        }
-    }
-    // Extract regular text from w:t elements
-    const textMatches = cellXml.match(/<w:t[^>]*>([^<]*)<\/w:t>/g) || [];
-    for (const match of textMatches) {
-        const text = match.replace(/<[^>]+>/g, '');
-        if (text) {
-            parts.push(text);
-        }
-    }
-    let result = parts.join('').trim();
-    result = decodeXmlEntities(result);
-    // Escape pipe characters in cell content (would break table)
-    result = result.replace(/\|/g, '\\|');
-    return result;
-}
-/**
- * Parse a table row, handling merged cells (gridSpan)
- */
-function parseTableRow(rowXml, expectedCols) {
-    // Match cells - handle both <w:tc> and <w:tc ...>
-    const cellMatches = rowXml.match(/<w:tc(?:\s[^>]*)?>[\s\S]*?<\/w:tc>/g) || [];
-    const cells = [];
-    const colSpans = [];
-    for (const cellXml of cellMatches) {
-        // Check for horizontal merge (gridSpan)
-        const gridSpanMatch = cellXml.match(/<w:gridSpan\s+w:val="(\d+)"/);
-        const span = gridSpanMatch ? parseInt(gridSpanMatch[1], 10) : 1;
-        // Check for vertical merge continuation (vMerge without restart)
-        // If vMerge is present without w:val="restart", it's a continuation - use empty
-        const vMergeMatch = cellXml.match(/<w:vMerge(?:\s+w:val="([^"]+)")?/);
-        const isVMergeContinuation = vMergeMatch && vMergeMatch[1] !== 'restart';
-        const cellText = isVMergeContinuation ? '' : extractCellText(cellXml);
-        // Add the cell content
-        cells.push(cellText);
-        colSpans.push(span);
-        // For gridSpan > 1, add empty cells to maintain column alignment
-        for (let i = 1; i < span; i++) {
-            cells.push('');
-            colSpans.push(0); // 0 indicates this is a spanned cell
-        }
-    }
-    return { cells, colSpans };
-}
-/**
- * Determine table grid column count from table XML
- */
-function getTableGridCols(tableXml) {
-    // Try to get from tblGrid
-    const gridColMatches = tableXml.match(/<w:gridCol/g) || [];
-    if (gridColMatches.length > 0) {
-        return gridColMatches.length;
-    }
-    // Fallback: count max cells in any row
-    const rowMatches = tableXml.match(/<w:tr[\s\S]*?<\/w:tr>/g) || [];
-    let maxCols = 0;
-    for (const rowXml of rowMatches) {
-        const { cells } = parseTableRow(rowXml, 0);
-        maxCols = Math.max(maxCols, cells.length);
-    }
-    return maxCols;
-}
-/**
- * Extract tables directly from Word document XML and convert to markdown pipe tables
- */
-export async function extractWordTables(docxPath) {
-    const AdmZip = (await import('adm-zip')).default;
-    const tables = [];
-    try {
-        const zip = new AdmZip(docxPath);
-        const docEntry = zip.getEntry('word/document.xml');
-        if (!docEntry) {
-            return tables;
-        }
-        const xml = docEntry.getData().toString('utf8');
-        // Find all table elements
-        const tableMatches = xml.match(/<w:tbl>[\s\S]*?<\/w:tbl>/g) || [];
-        for (const tableXml of tableMatches) {
-            // Determine expected column count from grid
-            const expectedCols = getTableGridCols(tableXml);
-            // Extract rows
-            const rowMatches = tableXml.match(/<w:tr[\s\S]*?<\/w:tr>/g) || [];
-            const rows = [];
-            for (const rowXml of rowMatches) {
-                const { cells } = parseTableRow(rowXml, expectedCols);
-                if (cells.length > 0) {
-                    rows.push(cells);
-                }
-            }
-            if (rows.length > 0) {
-                // Convert to markdown pipe table
-                const markdown = convertRowsToMarkdownTable(rows);
-                tables.push({ markdown, rowCount: rows.length, colCount: expectedCols || rows[0]?.length || 0 });
-            }
-        }
-    }
-    catch (err) {
-        console.error('Error extracting tables from Word:', err.message);
-    }
-    return tables;
-}
-/**
- * Convert array of rows (each row is array of cell strings) to markdown pipe table
- */
-function convertRowsToMarkdownTable(rows) {
-    if (rows.length === 0)
-        return '';
-    // Normalize column count (use max across all rows)
-    const colCount = Math.max(...rows.map((r) => r.length));
-    // Pad rows to have consistent column count
-    const normalizedRows = rows.map((row) => {
-        while (row.length < colCount) {
-            row.push('');
-        }
-        return row;
-    });
-    // Build markdown table
-    const lines = [];
-    // Header row
-    const header = normalizedRows[0];
-    lines.push('| ' + header.join(' | ') + ' |');
-    // Separator row
-    lines.push('|' + header.map(() => '---').join('|') + '|');
-    // Data rows
-    for (let i = 1; i < normalizedRows.length; i++) {
-        lines.push('| ' + normalizedRows[i].join(' | ') + ' |');
-    }
-    return lines.join('\n');
-}
-/**
- * Extract text from Word document using pandoc with track changes preserved
- */
-export async function extractFromWord(docxPath, options = {}) {
-    let text;
-    let messages = [];
-    let extractedMedia = [];
-    let hasTrackChanges = false;
-    let trackChangeStats = { insertions: 0, deletions: 0 };
-    // Determine media extraction directory
-    const docxDir = path.dirname(docxPath);
-    const mediaDir = options.mediaDir || path.join(docxDir, 'media');
-    // Skip media extraction if figures already exist (e.g., when re-importing with existing source)
-    const skipMediaExtraction = options.skipMediaExtraction || false;
-    // Extract tables directly from Word XML (reliable, no heuristics)
-    const wordTables = await extractWordTables(docxPath);
-    // Try pandoc first with --track-changes=all to preserve reviewer edits
-    try {
-        // Build pandoc command
-        let pandocCmd = `pandoc "${docxPath}" -t markdown --wrap=none --track-changes=all`;
-        if (!skipMediaExtraction) {
-            pandocCmd += ` --extract-media="${mediaDir}"`;
-        }
-        const { stdout } = await execAsync(pandocCmd, { maxBuffer: 50 * 1024 * 1024 });
-        text = stdout;
-        // Convert pandoc's track change format to CriticMarkup
-        const origLength = text.length;
-        // Use a more robust pattern that handles nested content
-        text = text.replace(/\[([^\[\]]*(?:\[[^\[\]]*\][^\[\]]*)*)\]\{\.insertion[^}]*\}/g, (match, content) => {
-            if (content.trim()) {
-                trackChangeStats.insertions++;
-                return `{++${content}++}`;
-            }
-            return ''; // Empty insertions are removed
-        });
-        text = text.replace(/\[([^\[\]]*(?:\[[^\[\]]*\][^\[\]]*)*)\]\{\.deletion[^}]*\}/g, (match, content) => {
-            if (content.trim()) {
-                trackChangeStats.deletions++;
-                return `{--${content}--}`;
-            }
-            return ''; // Empty deletions are removed
-        });
-        // Handle any remaining pandoc track change patterns
-        let prevText;
-        do {
-            prevText = text;
-            text = text.replace(/\[([^\]]*)\]\{\.insertion[^}]*\}/g, (match, content) => {
-                if (content.trim()) {
-                    trackChangeStats.insertions++;
-                    return `{++${content}++}`;
-                }
-                return '';
-            });
-            text = text.replace(/\[([^\]]*)\]\{\.deletion[^}]*\}/g, (match, content) => {
-                if (content.trim()) {
-                    trackChangeStats.deletions++;
-                    return `{--${content}--}`;
-                }
-                return '';
-            });
-        } while (text !== prevText);
-        // Handle pandoc comment patterns - remove comment text from body
-        text = text.replace(/\[[^\]]*\]\{\.comment-start[^}]*\}/g, '');
-        text = text.replace(/\[\]\{\.comment-end[^}]*\}/g, '');
-        // Also handle {.mark} spans
-        text = text.replace(/\[([^\]]*)\]\{\.mark\}/g, '$1');
-        hasTrackChanges = trackChangeStats.insertions > 0 || trackChangeStats.deletions > 0;
-        if (hasTrackChanges) {
-            messages.push({
-                type: 'info',
-                message: `Found ${trackChangeStats.insertions} insertion(s) and ${trackChangeStats.deletions} deletion(s) from track changes`
-            });
-        }
-        // Find extracted media files
-        const mediaSubdir = path.join(mediaDir, 'media');
-        if (fs.existsSync(mediaSubdir)) {
-            extractedMedia = fs.readdirSync(mediaSubdir)
-                .filter(f => /\.(png|jpg|jpeg|gif|svg|emf|wmf|tiff?)$/i.test(f))
-                .map(f => path.join(mediaSubdir, f));
-            if (extractedMedia.length > 0) {
-                messages.push({
-                    type: 'info',
-                    message: `Extracted ${extractedMedia.length} image(s) to ${mediaSubdir}`
-                });
-            }
-        }
-    }
-    catch (pandocErr) {
-        // Pandoc not available — use XML-based extraction with track change support
-        const { extractPlainTextWithTrackChanges } = await import('./word.js');
-        const { getInstallInstructions } = await import('./dependencies.js');
-        const installCmd = getInstallInstructions('pandoc');
-        const xmlResult = await extractPlainTextWithTrackChanges(docxPath);
-        text = xmlResult.text;
-        hasTrackChanges = xmlResult.hasTrackChanges;
-        trackChangeStats = xmlResult.stats;
-        if (hasTrackChanges) {
-            messages.push({
-                type: 'warning',
-                message: `Pandoc not installed. Using built-in XML extractor (${trackChangeStats.insertions} insertions, ${trackChangeStats.deletions} deletions preserved). Formatting may differ. Install pandoc for best results: ${installCmd}`
-            });
-        }
-        else {
-            messages.push({
-                type: 'warning',
-                message: `Pandoc not installed. Using built-in XML extractor (no track changes found). Install pandoc for better formatting: ${installCmd}`
-            });
-        }
-    }
-    // Extract comments directly from docx XML
-    const comments = await extractWordComments(docxPath);
-    // Extract comment anchor texts
-    const { anchors } = await extractCommentAnchors(docxPath);
-    return {
-        text,
-        comments,
-        anchors,
-        messages,
-        extractedMedia,
-        tables: wordTables,
-        hasTrackChanges,
-        trackChangeStats,
-    };
-}
 /**
  * Insert comments into markdown text based on anchor texts with context
  */
@@ -829,530 +333,6 @@ export function insertCommentsIntoMarkdown(markdown, comments, anchors, options
     }
     return result;
 }
-/**
- * Fix citation and math annotations by preserving original markdown syntax
- */
-function fixCitationAnnotations(text, originalMd) {
-    // Fix math annotations - preserve inline and display math
-    text = text.replace(/\{--(\$[^$]+\$)--\}/g, '$1');
-    text = text.replace(/\{--(\$\$[^$]+\$\$)--\}/g, '$1');
-    text = text.replace(/\{~~(\$[^$]+\$)~>[^~]+~~\}/g, '$1');
-    text = text.replace(/\{~~(\$\$[^$]+\$\$)~>[^~]+~~\}/g, '$1');
-    // Extract all citations from original markdown
-    const citationPattern = /\[@[^\]]+\]/g;
-    const originalCitations = [...originalMd.matchAll(citationPattern)].map(m => m[0]);
-    // Fix substitutions where left side has markdown citation
-    text = text.replace(/\{~~(\[@[^\]]+\])~>[^~]+~~\}/g, '$1');
-    // Fix substitutions where left side STARTS with markdown citation
-    text = text.replace(/\{~~(\[@[^\]]+\])\s*([^~]*)~>([^~]*)~~\}/g, (match, cite, oldText, newText) => {
-        if (oldText.trim() === '' && newText.trim() === '') {
-            return cite;
-        }
-        if (oldText.trim() || newText.trim()) {
-            return cite + (oldText.trim() !== newText.trim() ? ` {~~${oldText.trim()}~>${newText.trim()}~~}` : ` ${newText}`);
-        }
-        return cite;
-    });
-    // Fix deletions of markdown citations
-    text = text.replace(/\{--(\[@[^\]]+\])--\}/g, '$1');
-    // Fix insertions of rendered citations
-    text = text.replace(/\{\+\+\([A-Z][^)]*\d{4}[^)]*\)\+\+\}/g, '');
-    // Clean up broken multi-part substitutions
-    text = text.replace(/\{~~(@[A-Za-z]+\d{4})~>[^~]+~~\}/g, '[$1]');
-    // Fix citations split across substitution boundaries
-    text = text.replace(/\{~~\[@~>[^~]*~~\}([A-Za-z]+\d{4})\]/g, '[@$1]');
-    // Clean up any remaining partial citations
-    text = text.replace(/\{~~;\s*@([A-Za-z]+\d{4})\]~>[^~]*~~\}/g, '; [@$1]');
-    // Remove rendered citation insertions (with Unicode support)
-    text = text.replace(/\{\+\+\(\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?(?:[;,]\s*\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?)*\)\+\+\}/gu, '');
-    text = text.replace(/\{\+\+\(\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?(?:[;,]\s*\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?)*\)\.\s*\+\+\}/gu, '');
-    // Trailing citation fragments
-    text = text.replace(/\{\+\+\d{4}[a-z]?(?:[;,]\s*(?:\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+)?\d{4}[a-z]?)*\)\.\s*\+\+\}/gu, '');
-    text = text.replace(/\{\+\+\d{4}[a-z]?(?:[;,]\s*(?:\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+)?\d{4}[a-z]?)*\)\s*\+\+\}/gu, '');
-    // Just year with closing paren
-    text = text.replace(/\{\+\+\d{4}[a-z]?\)\.\s*\+\+\}/g, '');
-    text = text.replace(/\{\+\+\d{4}[a-z]?\)\s*\+\+\}/g, '');
-    // Leading citation fragments
-    text = text.replace(/\{\+\+\(?\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s*\+\+\}/gu, '');
-    // Semicolon-separated fragments
-    text = text.replace(/\{\+\+[;,]\s*\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?\+\+\}/gu, '');
-    // Year ranges with authors
-    text = text.replace(/\{\+\+\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?(?:[;,]\s*\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?)*\)\s*\+\+\}/gu, '');
-    text = text.replace(/\{\+\+\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?(?:[;,]\s*\p{Lu}\p{L}*(?:\s+et\s+al\.?)?\s+\d{4}[a-z]?)*\)\.\s*\+\+\}/gu, '');
-    // Clean up double spaces and orphaned punctuation
-    text = text.replace(/  +/g, ' ');
-    text = text.replace(/\s+\./g, '.');
-    text = text.replace(/\s+,/g, ',');
-    // Final cleanup - remove empty annotations
-    text = text.replace(/\{~~\s*~>\s*~~\}/g, '');
-    text = text.replace(/\{\+\+\s*\+\+\}/g, '');
-    text = text.replace(/\{--\s*--\}/g, '');
-    return text;
-}
-/**
- * Strip markdown syntax to get plain text
- */
-function stripMarkdownSyntax(md) {
-    return md
-        .replace(/^---[\s\S]*?---\n*/m, '')
-        .replace(/^#{1,6}\s+/gm, '')
-        .replace(/(\*\*|__)(.*?)\1/g, '$2')
-        .replace(/(\*|_)(.*?)\1/g, '$2')
-        .replace(/\[([^\]]+)\]\([^)]+\)/g, '$1')
-        .replace(/!\[([^\]]*)\]\([^)]+\)/g, '')
-        .replace(/`([^`]+)`/g, '$1')
-        .replace(/```[\s\S]*?```/g, '')
-        .replace(/^>\s*/gm, '')
-        .replace(/^[-*_]{3,}\s*$/gm, '')
-        .replace(/^[\s]*[-*+]\s+/gm, '')
-        .replace(/^[\s]*\d+\.\s+/gm, '')
-        .replace(/\|/g, ' ')
-        .replace(/^[-:]+$/gm, '')
-        .replace(/\n{3,}/g, '\n\n')
-        .trim();
-}
-/**
- * Generate annotated markdown by diffing original MD against Word text
- */
-export function generateAnnotatedDiff(originalMd, wordText, author = 'Reviewer') {
-    const normalizedOriginal = normalizeWhitespace(originalMd);
-    const normalizedWord = normalizeWhitespace(wordText);
-    const changes = diffWords(normalizedOriginal, normalizedWord);
-    let result = '';
-    for (const part of changes) {
-        if (part.added) {
-            result += `{++${part.value}++}`;
-        }
-        else if (part.removed) {
-            result += `{--${part.value}--}`;
-        }
-        else {
-            result += part.value;
-        }
-    }
-    return result;
-}
-/**
- * Inject Word tables (extracted from XML) into pandoc text output
- */
-function injectWordTables(pandocText, wordTables) {
-    if (!wordTables || wordTables.length === 0) {
-        return pandocText;
-    }
-    let result = pandocText;
-    for (const table of wordTables) {
-        const firstLine = table.markdown.split('\n')[0];
-        const headerCells = firstLine
-            .split('|')
-            .map((c) => c.trim())
-            .filter((c) => c.length > 0);
-        if (headerCells.length === 0)
-            continue;
-        const firstCell = headerCells[0];
-        const startIdx = result.indexOf(firstCell);
-        if (startIdx === -1)
-            continue;
-        const lastLine = table.markdown.split('\n').pop();
-        const lastCells = lastLine
-            .split('|')
-            .map((c) => c.trim())
-            .filter((c) => c.length > 0);
-        const lastCell = lastCells[lastCells.length - 1] || lastCells[0];
-        const endIdx = result.indexOf(lastCell, startIdx);
-        if (endIdx === -1)
-            continue;
-        let regionStart = result.lastIndexOf('\n\n', startIdx);
-        if (regionStart === -1)
-            regionStart = 0;
-        else
-            regionStart += 2;
-        let regionEnd = result.indexOf('\n\n', endIdx + lastCell.length);
-        if (regionEnd === -1)
-            regionEnd = result.length;
-        result = result.slice(0, regionStart) + table.markdown + '\n\n' + result.slice(regionEnd);
-    }
-    return result;
-}
-/**
- * Smart paragraph-level diff that preserves markdown structure
- */
-export function generateSmartDiff(originalMd, wordText, author = 'Reviewer', options = {}) {
-    const { wordTables = [], imageRegistry = null } = options;
-    // Inject Word tables into pandoc output
-    let wordTextWithTables = injectWordTables(wordText, wordTables);
-    // Protect markdown tables
-    const { text: mdWithTablesProtected, tables } = protectTables(originalMd);
-    // Also protect tables in Word text
-    const { text: wordWithTablesProtected, tables: wordTableBlocks } = protectTables(wordTextWithTables);
-    // Protect images
-    const { text: mdWithImagesProtected, images: origImages } = protectImages(mdWithTablesProtected, imageRegistry);
-    const { text: wordWithImagesProtected, images: wordImages } = protectImages(wordWithTablesProtected, imageRegistry);
-    // Match Word images to original images
-    const imageMapping = matchWordImagesToOriginal(origImages, wordImages, imageRegistry);
-    // Replace Word image placeholders with matching original placeholders
-    let wordWithMappedImages = wordWithImagesProtected;
-    for (const [wordPlaceholder, origPlaceholder] of imageMapping) {
-        wordWithMappedImages = wordWithMappedImages.split(wordPlaceholder).join(origPlaceholder);
-    }
-    // Protect figure/table anchors
-    const { text: mdWithAnchorsProtected, anchors: figAnchors } = protectAnchors(mdWithImagesProtected);
-    // Protect cross-references
-    const { text: mdWithXrefsProtected, crossrefs } = protectCrossrefs(mdWithAnchorsProtected);
-    // Protect math
-    const { text: mdWithMathProtected, mathBlocks } = protectMath(mdWithXrefsProtected);
-    // Protect citations
-    const { text: mdProtected, citations } = protectCitations(mdWithMathProtected);
-    // Replace rendered elements in Word text
-    let wordProtected = wordWithMappedImages;
-    wordProtected = replaceRenderedMath(wordProtected, mathBlocks);
-    wordProtected = replaceRenderedCitations(wordProtected, citations.length);
-    // Split into paragraphs
-    const originalParas = mdProtected.split(/\n\n+/);
-    const wordParas = wordProtected.split(/\n\n+/);
-    const result = [];
-    // Try to match paragraphs intelligently
-    let wordIdx = 0;
-    for (let i = 0; i < originalParas.length; i++) {
-        const orig = originalParas[i] || '';
-        const { prefix: mdPrefix, content: origContent } = extractMarkdownPrefix(orig.split('\n')[0]);
-        // Find best matching word paragraph
-        let bestMatch = -1;
-        let bestScore = 0;
-        for (let j = wordIdx; j < Math.min(wordIdx + 3, wordParas.length); j++) {
-            const wordPara = wordParas[j] || '';
-            const origWords = new Set(origContent.toLowerCase().split(/\s+/));
-            const wordWords = wordPara.toLowerCase().split(/\s+/);
-            const common = wordWords.filter((w) => origWords.has(w)).length;
-            const score = common / Math.max(origWords.size, wordWords.length);
-            if (score > bestScore && score > 0.3) {
-                bestScore = score;
-                bestMatch = j;
-            }
-        }
-        if (bestMatch === -1) {
-            if (mdPrefix && wordIdx < wordParas.length) {
-                const wordPara = wordParas[wordIdx];
-                if (wordPara.toLowerCase().includes(origContent.toLowerCase().slice(0, 20))) {
-                    bestMatch = wordIdx;
-                }
-            }
-        }
-        if (bestMatch >= 0) {
-            const word = wordParas[bestMatch];
-            const origStripped = stripMarkdownSyntax(orig);
-            const wordNormalized = normalizeWhitespace(word);
-            if (origStripped === wordNormalized) {
-                result.push(orig);
-            }
-            else {
-                const changes = diffWords(origStripped, wordNormalized);
-                let annotated = mdPrefix;
-                for (const part of changes) {
-                    if (part.added) {
-                        annotated += `{++${part.value}++}`;
-                    }
-                    else if (part.removed) {
-                        annotated += `{--${part.value}--}`;
-                    }
-                    else {
-                        annotated += part.value;
-                    }
-                }
-                result.push(annotated);
-            }
-            wordIdx = bestMatch + 1;
-        }
-        else {
-            // Paragraph deleted entirely
-            if (mdPrefix && mdPrefix.match(/^#{1,6}\s+/)) {
-                result.push(orig);
-            }
-            else {
-                result.push(`{--${orig}--}`);
-            }
-        }
-    }
-    // Any remaining word paragraphs are additions
-    for (let j = wordIdx; j < wordParas.length; j++) {
-        const word = wordParas[j];
-        if (word.trim()) {
-            result.push(`{++${word}++}`);
-        }
-    }
-    // Restore protected content
-    let finalResult = result.join('\n\n');
-    finalResult = restoreCitations(finalResult, citations);
-    finalResult = restoreMath(finalResult, mathBlocks);
-    finalResult = restoreCrossrefs(finalResult, crossrefs);
-    finalResult = restoreAnchors(finalResult, figAnchors);
-    finalResult = restoreImages(finalResult, origImages);
-    finalResult = restoreImages(finalResult, wordImages);
-    finalResult = restoreTables(finalResult, tables);
-    finalResult = restoreTables(finalResult, wordTableBlocks);
-    return finalResult;
-}
-/**
- * Clean up redundant adjacent annotations
- */
-export function cleanupAnnotations(text) {
-    // Convert adjacent delete+insert to substitution
-    text = text.replace(/\{--(.+?)--\}\s*\{\+\+(.+?)\+\+\}/g, '{~~$1~>$2~~}');
-    // Also handle insert+delete
-    text = text.replace(/\{\+\+(.+?)\+\+\}\s*\{--(.+?)--\}/g, '{~~$2~>$1~~}');
-    // Fix malformed patterns
-    text = text.replace(/\{--([^}]+?)~>([^}]+?)~~\}/g, '{~~$1~>$2~~}');
-    // Fix malformed substitutions that got split
-    text = text.replace(/\{~~([^~]+)\s*--\}/g, '{--$1--}');
-    text = text.replace(/\{\+\+([^+]+)~~\}/g, '{++$1++}');
-    // Clean up empty annotations
-    text = text.replace(/\{--\s*--\}/g, '');
-    text = text.replace(/\{\+\+\s*\+\+\}/g, '');
-    // Clean up double spaces in prose, but preserve table formatting
-    const lines = text.split('\n');
-    let inTable = false;
-    const processedLines = lines.map((line, idx) => {
-        const isSeparator = /^[-]+(\s+[-]+)+\s*$/.test(line.trim());
-        const looksLikeTableRow = /\S+\s{2,}\S+/.test(line);
-        if (isSeparator) {
-            if (!inTable) {
-                inTable = true;
-            }
-            return line;
-        }
-        if (inTable) {
-            if (line.trim() === '') {
-                let lookAhead = idx + 1;
-                let foundTableContent = false;
-                let foundEndSeparator = false;
-                while (lookAhead < lines.length && lookAhead < idx + 20) {
-                    const nextLine = lines[lookAhead].trim();
-                    if (nextLine === '') {
-                        lookAhead++;
-                        continue;
-                    }
-                    if (/^[-]+(\s+[-]+)+\s*$/.test(nextLine)) {
-                        foundEndSeparator = true;
-                        break;
-                    }
-                    if (/\S+\s{2,}\S+/.test(nextLine)) {
-                        foundTableContent = true;
-                        break;
-                    }
-                    if (/^\*[^*]+\*\s*$/.test(nextLine)) {
-                        foundTableContent = true;
-                        break;
-                    }
-                    if (lines[lookAhead].startsWith('  ')) {
-                        lookAhead++;
-                        continue;
-                    }
-                    break;
-                }
-                if (foundTableContent || foundEndSeparator) {
-                    return line;
-                }
-                inTable = false;
-                return line;
-            }
-            return line;
-        }
-        if (looksLikeTableRow) {
-            let nextIdx = idx + 1;
-            while (nextIdx < lines.length && lines[nextIdx].trim() === '') {
-                nextIdx++;
-            }
-            if (nextIdx < lines.length && /^[-]+(\s+[-]+)+\s*$/.test(lines[nextIdx].trim())) {
-                return line;
-            }
-        }
-        if (line.trim().startsWith('|')) {
-            return line;
-        }
-        return line.replace(/  +/g, ' ');
-    });
-    text = processedLines.join('\n');
-    return text;
-}
-/**
- * Parse visible comment markers from Word text
- */
-export function parseVisibleComments(text) {
-    const comments = [];
-    const pattern = /\[([^\]:]+):\s*([^\]]+)\]/g;
-    let match;
-    while ((match = pattern.exec(text)) !== null) {
-        comments.push({
-            author: match[1].trim(),
-            text: match[2].trim(),
-            position: match.index,
-        });
-    }
-    return comments;
-}
-/**
- * Convert visible comments to CriticMarkup format
- */
-export function convertVisibleComments(text) {
-    return text.replace(/\[([^\]:]+):\s*([^\]]+)\]/g, '{>>$1: $2<<}');
-}
-/**
- * Restore pandoc-crossref figure/table references from Word-rendered format
- */
-export function restoreCrossrefFromWord(text, projectDir, restoredLabels = null) {
-    const messages = [];
-    let restored = 0;
-    let result = text;
-    const registry = readImageRegistry(projectDir);
-    if (!restoredLabels) {
-        restoredLabels = new Set();
-    }
-    // Pattern 1: [Figure]{.mark} [N]{.mark}
-    result = result.replace(/\[(Figure|Table|Fig\.?)\]\{\.mark\}\s*\[(\d+|S\d+)\]\{\.mark\}/gi, (match, type, num) => {
-        const prefix = type.toLowerCase().startsWith('tab') ? 'tbl' : 'fig';
-        if (registry) {
-            const entry = registry.byNumber?.get(`${prefix}:${num}`);
-            if (entry && entry.label) {
-                restored++;
-                return `@${prefix}:${entry.label}`;
-            }
-        }
-        restored++;
-        messages.push(`Restored ${type} ${num} (no label found, using placeholder)`);
-        return `@${prefix}:fig${num}`;
-    });
-    // Pattern 2: Plain "Figure N" or "Fig. N"
-    result = result.replace(/(?<!!)\b(Figure|Fig\.?|Table|Tbl\.?)\s+(\d+|S\d+)\b(?!\s*:)/gi, (match, type, num) => {
-        const prefix = type.toLowerCase().startsWith('tab') ? 'tbl' : 'fig';
-        if (registry) {
-            const entry = registry.byNumber?.get(`${prefix}:${num}`);
-            if (entry && entry.label) {
-                restored++;
-                return `@${prefix}:${entry.label}`;
-            }
-        }
-        return match;
-    });
-    // Pattern 3: Remove duplicate plain-text captions
-    result = result.replace(/(\!\[[^\]]+\]\([^)]+\)(?:\{[^}]*\})?)\s*\n+\s*(?:Figure|Fig\.?|Table|Tbl\.?)\s+\d+[:\.]?\s*[^\n]+/gi, '$1');
-    // Pattern 4: Clean up image captions that start with "Figure N: "
-    result = result.replace(/!\[(Figure|Fig\.?|Table|Tbl\.?)\s+(\d+|S\d+)[:\.]?\s*([^\]]*)\]\(([^)]+)\)(?:\{[^}]*\})?/gi, (match, type, num, caption, imgPath) => {
-        const prefix = type.toLowerCase().startsWith('tab') ? 'tbl' : 'fig';
-        const labelKey = `${prefix}:${num}`;
-        if (registry) {
-            const entry = registry.byNumber?.get(labelKey);
-            if (entry) {
-                if (restoredLabels.has(labelKey)) {
-                    messages.push(`Skipped duplicate ${prefix}:${entry.label} (already restored)`);
-                    return `![${entry.caption}](${entry.path})`;
-                }
-                restoredLabels.add(labelKey);
-                restored++;
-                messages.push(`Restored image ${prefix}:${entry.label} from Figure ${num}`);
-                return `![${entry.caption}](${entry.path}){#${prefix}:${entry.label}}`;
-            }
-        }
-        const cleanCaption = caption.trim();
-        return `![${cleanCaption}](${imgPath})`;
-    });
-    return { text: result, restored, messages, restoredLabels };
-}
-/**
- * Restore proper markdown image syntax from Word-extracted text using image registry
- */
-export function restoreImagesFromRegistry(text, projectDir, restoredLabels = null) {
-    const messages = [];
-    let restored = 0;
-    const registry = readImageRegistry(projectDir);
-    if (!registry || !registry.figures || registry.figures.length === 0) {
-        return { text, restored: 0, messages: ['No image registry found'] };
-    }
-    if (!restoredLabels) {
-        restoredLabels = new Set();
-    }
-    let result = text;
-    // Pattern 1: Caption-like text
-    const captionPatterns = [
-        /@(fig|tbl):([a-zA-Z0-9_-]+):\s*([^\n]+)/gi,
-        /^(Figure|Fig\.?)\s+(\d+|S\d+)[.:]\s*([^\n]+)/gim,
-        /\|\s*@(fig|tbl):([a-zA-Z0-9_-]+):\s*([^|]+)\s*\|/gi,
-    ];
-    // Fix @fig:label: caption patterns
-    result = result.replace(captionPatterns[0], (match, type, label, caption) => {
-        const key = `${type}:${label}`;
-        const entry = registry.byLabel.get(key);
-        if (entry) {
-            if (restoredLabels.has(key)) {
-                messages.push(`Skipped duplicate ${key} (already restored)`);
-                return `![${entry.caption}](${entry.path})`;
-            }
-            restoredLabels.add(key);
-            restored++;
-            messages.push(`Restored ${type}:${label} from registry`);
-            return `![${entry.caption}](${entry.path}){#${type}:${label}}`;
-        }
-        return match;
-    });
-    // Fix table-wrapped captions
-    result = result.replace(captionPatterns[2], (match, type, label, caption) => {
-        const key = `${type}:${label}`;
-        const entry = registry.byLabel.get(key);
-        if (entry) {
-            if (restoredLabels.has(key)) {
-                messages.push(`Skipped duplicate ${key} from table wrapper`);
-                return `![${entry.caption}](${entry.path})`;
-            }
-            restoredLabels.add(key);
-            restored++;
-            messages.push(`Restored ${type}:${label} from table wrapper`);
-            return `![${entry.caption}](${entry.path}){#${type}:${label}}`;
-        }
-        return match;
-    });
-    // Clean up empty table structures
-    result = result.replace(/\|\s*\|\s*\n\|:--:\|\s*\n/g, '');
-    // Fix "Figure N:" standalone lines
-    result = result.replace(captionPatterns[1], (match, prefix, num, caption) => {
-        const numKey = `fig:${num}`;
-        const entry = registry.byNumber.get(numKey);
-        if (entry) {
-            const labelKey = `fig:${entry.label}`;
-            if (restoredLabels.has(labelKey)) {
-                messages.push(`Skipped duplicate Figure ${num} (already restored)`);
-                return `![${entry.caption}](${entry.path})`;
-            }
-            restoredLabels.add(labelKey);
-            restored++;
-            messages.push(`Restored Figure ${num} by number lookup`);
-            return `![${entry.caption}](${entry.path}){#fig:${entry.label}}`;
-        }
-        return match;
-    });
-    // Fix generic media paths by matching caption text
-    const genericImagePattern = /!\[([^\]]*)\]\(media\/[^)]+\)/g;
-    result = result.replace(genericImagePattern, (match, caption) => {
-        if (!caption || caption.trim() === '') {
-            return match;
-        }
-        const captionKey = caption.slice(0, 50).toLowerCase().trim();
-        const entry = registry.byCaption.get(captionKey);
-        if (entry) {
-            const labelKey = entry.label ? `${entry.type}:${entry.label}` : null;
-            if (labelKey && restoredLabels.has(labelKey)) {
-                messages.push(`Skipped duplicate by caption match: ${captionKey.slice(0, 30)}...`);
-                return `![${entry.caption}](${entry.path})`;
-            }
-            if (labelKey) {
-                restoredLabels.add(labelKey);
-            }
-            restored++;
-            messages.push(`Restored image by caption match: ${captionKey.slice(0, 30)}...`);
-            const anchor = (entry.label && !restoredLabels.has(labelKey)) ? `{#${entry.type}:${entry.label}}` : '';
-            return `![${entry.caption}](${entry.path})${anchor}`;
-        }
-        return match;
-    });
-    return { text: result, restored, messages };
-}
 /**
  * Import Word document with track changes directly as CriticMarkup
  */