docrev 0.11.1 → 0.11.3

package/lib/annotations.ts

@@ -1,664 +1,695 @@
-/**
- * CriticMarkup annotation parsing and manipulation
- *
- * Syntax:
- *   {++inserted text++}     - Insertions
- *   {--deleted text--}      - Deletions
- *   {~~old~>new~~}          - Substitutions
- *   {>>Author: comment<<}   - Comments
- *   {==text==}              - Highlights
- */
-
-import type { Annotation, AnnotationCounts, StripOptions, CommentFilterOptions } from './types.js';
-
-// =============================================================================
-// Constants
-// =============================================================================
-
-/** Window size for context lookup (characters before/after position) */
-const CONTEXT_WINDOW_SIZE = 2000;
-
-/** Characters of context to include in annotation results */
-const CONTEXT_SNIPPET_SIZE = 50;
-
-/** Maximum iterations for nested annotation stripping (safety limit) */
-const MAX_STRIP_ITERATIONS = 20;
-
-/** Maximum author name length in comments */
-const MAX_AUTHOR_LENGTH = 30;
-
-/** Maximum content length before heuristic assumes it's not a comment */
-const MAX_COMMENT_CONTENT_LENGTH = 200;
-
-// =============================================================================
-// Patterns
-// =============================================================================
-
-// Patterns for each annotation type
-const PATTERNS = {
-  insert: /\{\+\+(.+?)\+\+\}/gs,
-  delete: /\{--(.+?)--\}/gs,
-  substitute: /\{~~(.+?)~>(.+?)~~\}/gs,
-  comment: /\{>>(.+?)<<\}/gs,
-  highlight: /\{==(.+?)==\}/gs,
-};
-
-/**
- * Check if a potential comment is actually a false positive
- * (e.g., figure caption, nested inside other annotation, code block, etc.)
- * @param commentContent - The content inside {>>...<<}
- * @param fullText - The full document text
- * @param position - Position of the comment in the text
- * @returns true if this is a false positive (not a real comment)
- */
-function isCommentFalsePositive(commentContent: string, fullText: string, position: number): boolean {
-  // Check if inside a code block (fenced or indented)
-  const textBefore = fullText.slice(Math.max(0, position - CONTEXT_WINDOW_SIZE), position);
-  const textAfter = fullText.slice(position, Math.min(fullText.length, position + CONTEXT_WINDOW_SIZE));
-
-  // Count unclosed fenced code blocks (``` or ~~~)
-  const fenceOpens = (textBefore.match(/^```|^~~~/gm) || []).length;
-  const fenceCloses = (textBefore.match(/```$|~~~$/gm) || []).length;
-  if (fenceOpens > fenceCloses) return true; // Inside code block
-
-  // Check if on an indented line (4+ spaces or tab at line start = code)
-  const lineStart = textBefore.lastIndexOf('\n') + 1;
-  const linePrefix = fullText.slice(lineStart, position);
-  if (/^(\t|    )/.test(linePrefix)) return true; // Indented code
-
-  // Check if inside inline code backticks
-  const backticksBefore = (linePrefix.match(/`/g) || []).length;
-  if (backticksBefore % 2 === 1) return true; // Inside inline code
-
-  // Check if nested inside a deletion or insertion block
-  const nearTextBefore = fullText.slice(Math.max(0, position - 500), position);
-
-  // Count unclosed deletion markers
-  const delOpens = (nearTextBefore.match(/\{--/g) || []).length;
-  const delCloses = (nearTextBefore.match(/--\}/g) || []).length;
-  if (delOpens > delCloses) return true; // Nested inside deletion
-
-  // Count unclosed insertion markers
-  const insOpens = (nearTextBefore.match(/\{\+\+/g) || []).length;
-  const insCloses = (nearTextBefore.match(/\+\+\}/g) || []).length;
-  if (insOpens > insCloses) return true; // Nested inside insertion
-
-  // Heuristics for figure captions and other false positives:
-
-  // Contains image/figure path patterns
-  if (/\(figures?\/|\(images?\/|\.png|\.jpg|\.jpeg|\.gif|\.svg|\.pdf/i.test(commentContent)) return true;
-
-  // Contains markdown figure reference syntax
-  if (/\{#fig:|!\[/.test(commentContent)) return true;
-
-  // Real comments typically have "Author:" at start. Accept hyphens, apostrophes,
-  // periods, and Unicode letters so names like "Jens-Christian Svenning" or
-  // "Camilla T Colding-Jørgensen" don't get rejected. See gcol33/docrev#1.
-  const hasAuthorPrefix = /^[\p{L}][\p{L}\s\-'.]{0,30}:\s/u.test(commentContent.trim());
-  const hasResolvedMark = /^[✓✔]\s/.test(commentContent.trim());
-
-  // Contains URL patterns (likely a link, not a comment) — only filter when
-  // there is no real author prefix, since reviewers legitimately cite URLs/DOIs.
-  if (!hasAuthorPrefix && /https?:\/\/|www\./i.test(commentContent) && commentContent.length < 150) return true;
-
-  // Looks like code (contains programming patterns)
-  if (/function\s*\(|=>|import\s+|export\s+|const\s+|let\s+|var\s+/.test(commentContent)) return true;
-
-  // Very long without clear author pattern (likely caption, not comment)
-  if (!hasAuthorPrefix && !hasResolvedMark && commentContent.length > MAX_COMMENT_CONTENT_LENGTH) return true;
-
-  // Looks like a figure caption (starts with "Fig" or contains typical caption words)
-  if (/^(Fig\.?|Figure|Table|Sankey|Diagram|Proportion|Distribution|Map|Chart|Graph|Plot|Panel)/i.test(commentContent.trim())) {
-    return true;
-  }
-
-  // Contains LaTeX-like patterns (likely equation, not comment)
-  if (/\\[a-z]+\{|\\frac|\\sum|\\int|\\begin\{/.test(commentContent)) return true;
-
-  // Looks like BibTeX entry (not a comment)
-  if (/@article\{|@book\{|@inproceedings\{/i.test(commentContent)) return true;
-
-  return false;
-}
-
-// Combined pattern for any track change (not comments)
-const TRACK_CHANGE_PATTERN = /(\{\+\+.+?\+\+\}|\{--.+?--\}|\{~~.+?~>.+?~~\})/gs;
-
-// =============================================================================
-// Public API
-// =============================================================================
-
-/**
- * Parse all annotations from text
- * @param text - Markdown text containing CriticMarkup annotations
- * @returns Array of parsed annotations sorted by position
- * @throws TypeError If text is not a string
- */
-export function parseAnnotations(text: string): Annotation[] {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  const annotations: Annotation[] = [];
-
-  // Build line number lookup
-  const lines = text.split('\n');
-  let pos = 0;
-  const lineStarts = lines.map((line) => {
-    const start = pos;
-    pos += line.length + 1;
-    return start;
-  });
-
-  function getLine(position: number): number {
-    for (let i = 0; i < lineStarts.length; i++) {
-      const start = lineStarts[i];
-      if (start !== undefined && start > position) return i;
-    }
-    return lineStarts.length;
-  }
-
-  function getContext(position: number, length: number): { before: string; after: string } {
-    const start = Math.max(0, position - CONTEXT_SNIPPET_SIZE);
-    const end = Math.min(text.length, position + length + CONTEXT_SNIPPET_SIZE);
-    const before = text.slice(start, position).split('\n').pop() || '';
-    const after = text.slice(position + length, end).split('\n')[0] || '';
-    return { before, after };
-  }
-
-  // Parse insertions
-  for (const match of text.matchAll(PATTERNS.insert)) {
-    if (match.index === undefined) continue;
-    const ctx = getContext(match.index, match[0].length);
-    annotations.push({
-      type: 'insert',
-      match: match[0],
-      content: match[1] || '',
-      position: match.index,
-      line: getLine(match.index),
-      ...ctx,
-    });
-  }
-
-  // Parse deletions
-  for (const match of text.matchAll(PATTERNS.delete)) {
-    if (match.index === undefined) continue;
-    const ctx = getContext(match.index, match[0].length);
-    annotations.push({
-      type: 'delete',
-      match: match[0],
-      content: match[1] || '',
-      position: match.index,
-      line: getLine(match.index),
-      ...ctx,
-    });
-  }
-
-  // Parse substitutions
-  for (const match of text.matchAll(PATTERNS.substitute)) {
-    if (match.index === undefined) continue;
-    const ctx = getContext(match.index, match[0].length);
-    annotations.push({
-      type: 'substitute',
-      match: match[0],
-      content: match[1] || '',
-      replacement: match[2] || '',
-      position: match.index,
-      line: getLine(match.index),
-      ...ctx,
-    });
-  }
-
-  // Parse comments (with false positive filtering)
-  for (const match of text.matchAll(PATTERNS.comment)) {
-    if (match.index === undefined) continue;
-    // Skip false positives (figure captions, nested annotations, etc.)
-    const commentContent = match[1] || '';
-    if (isCommentFalsePositive(commentContent, text, match.index)) {
-      continue;
-    }
-
-    const ctx = getContext(match.index, match[0].length);
-    let commentText = commentContent;
-    let author = '';
-
-    // Extract author if present (format: "Author: comment")
-    const colonIdx = commentText.indexOf(':');
-    if (colonIdx > 0 && colonIdx < MAX_AUTHOR_LENGTH) {
-      author = commentText.slice(0, colonIdx).trim();
-      commentText = commentText.slice(colonIdx + 1).trim();
-    }
-
-    annotations.push({
-      type: 'comment',
-      match: match[0],
-      content: commentText,
-      author,
-      position: match.index,
-      line: getLine(match.index),
-      ...ctx,
-    });
-  }
-
-  // Sort by position
-  annotations.sort((a, b) => a.position - b.position);
-  return annotations;
-}
-
-/**
- * Strip annotations from text, applying changes
- * Handles nested annotations by iterating until stable
- * @param text - Markdown text with CriticMarkup annotations
- * @param options - Strip options
- * @returns Clean text with annotations applied/removed
- * @throws TypeError If text is not a string
- */
-export function stripAnnotations(text: string, options: StripOptions = {}): string {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  const { keepComments = false } = options;
-
-  // Iterate until no more changes (handles nested annotations)
-  let prev: string;
-  let iterations = 0;
-
-  do {
-    prev = text;
-
-    // Apply substitutions: {~~old~>new~~} → new
-    text = text.replace(PATTERNS.substitute, '$2');
-
-    // Apply insertions: {++text++} → text
-    text = text.replace(PATTERNS.insert, '$1');
-
-    // Apply deletions: {--text--} → nothing
-    // Don't touch surrounding whitespace - just remove the annotation
-    text = text.replace(PATTERNS.delete, '');
-
-    // Remove highlights: {==text==} → text
-    text = text.replace(PATTERNS.highlight, '$1');
-
-    // Remove comments unless keeping
-    if (!keepComments) {
-      text = text.replace(PATTERNS.comment, '');
-    }
-
-    // Strip pandoc highlight spans: [text]{.mark} → text.
-    // When `keepComments=true`, preserve `[anchor]{.mark}` that is the
-    // anchor of a kept `{>>...<<}` comment. The dual-build flow runs
-    // stripAnnotations() before prepareMarkdownWithMarkers(), and stripping
-    // the anchor span here would leave the marker generator with no anchor
-    // text — collapsing every multi-word anchor to a single fallback word
-    // in the rebuilt docx.
-    text = keepComments
-      ? text.replace(/(?<!<<\}\s{0,3})\[([^\]]*)\]\{\.mark\}/g, '$1')
-      : text.replace(/\[([^\]]*)\]\{\.mark\}/g, '$1');
-
-    // Clean up partial/orphaned markers within the loop
-    // This handles cases where nested annotations leave behind fragments
-
-    // Empty annotations (from nested stripping)
-    text = text.replace(/\{----\}/g, '');
-    text = text.replace(/\{\+\+\+\+\}/g, '');
-    text = text.replace(/\{--\s*--\}/g, '');
-    text = text.replace(/\{\+\+\s*\+\+\}/g, '');
-
-    // Orphaned substitution fragments: ~>text~~} or {~~text (no proper pairs)
-    text = text.replace(/~>[^{]*?~~\}/g, '');
-    text = text.replace(/\{~~[^~}]*$/gm, '');
-
-    // Handle malformed substitution from nested: {~~{~~old → just strip the {~~
-    text = text.replace(/\{~~\{~~/g, '{~~');
-    text = text.replace(/~~\}~~\}/g, '~~}');
-
-    iterations++;
-  } while (text !== prev && iterations < MAX_STRIP_ITERATIONS);
-
-  // Final cleanup of any remaining orphaned markers
-  // Orphaned closing markers
-  text = text.replace(/--\}(?:--\})+/g, '');
-  text = text.replace(/\+\+\}(?:\+\+\})+/g, '');
-  text = text.replace(/~~\}(?:~~\})+/g, '');
-  text = text.replace(/--\}/g, '');
-  text = text.replace(/\+\+\}/g, '');
-  text = text.replace(/~~\}/g, '');
-
-  // Orphaned opening markers
-  text = text.replace(/\{--(?:\{--)+/g, '');
-  text = text.replace(/\{\+\+(?:\{\+\+)+/g, '');
-  text = text.replace(/\{~~(?:\{~~)+/g, '');
-  text = text.replace(/\{--/g, '');
-  text = text.replace(/\{\+\+/g, '');
-  text = text.replace(/\{~~/g, '');
-  text = text.replace(/~>/g, '');
-
-  // Remove orphan [ from stripped {.mark} spans where the closing ]{.mark}
-  // was inside a comment. A [ is orphan if no `]` follows before end of line.
-  // We deliberately allow other `[` between the candidate and the matching `]`
-  // — otherwise nested forms like `[[0..9]]{.mark}` would have their outer
-  // `[` stripped because the lookahead saw the inner `[` as a barrier.
-  text = text.replace(/\[(?![^\]\n]*\])/g, '');
-
-  return text;
-}
-
-/**
- * Collapse multiple spaces to single space, preserving table formatting
- * Useful for cleaning up messy Word imports
- * @param text - Text to normalize
- * @returns Text with multiple spaces collapsed to single spaces
- * @throws TypeError If text is not a string
- */
-export function stripToSingleSpace(text: string): string {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  const lines = text.split('\n');
-  let inTable = false;
-
-  // Helper to check if a line looks like table content
-  const looksLikeTableRow = (ln: string): boolean => {
-    const trimmed = ln.trim();
-    if (!trimmed) return false;
-    // Has multiple consecutive spaces (column spacing)
-    // OR italicized category header with trailing spaces
-    return /\S\s{2,}\S/.test(trimmed) || (/^\*[^*]+\*\s*$/.test(trimmed) && /\s{2,}$/.test(ln));
-  };
-
-  for (let i = 0; i < lines.length; i++) {
-    const line = lines[i];
-    if (!line) continue;
-
-    // Detect table separator line
-    const isTableSeparator = /^\|?[\s-]*[-]{3,}[\s|:-]+[-]{3,}/.test(line) ||
-                              /^[-]{3,}\s{2,}[-]{3,}/.test(line);
-
-    if (isTableSeparator) {
-      inTable = true;
-      continue;
-    }
-
-    // Check if we're exiting the table
-    if (inTable && line.trim() === '') {
-      let nextContentLine = '';
-      for (let j = i + 1; j < lines.length && j < i + 5; j++) {
-        const nextLine = lines[j];
-        if (nextLine && nextLine.trim() !== '') {
-          nextContentLine = nextLine;
-          break;
-        }
-      }
-      if (!looksLikeTableRow(nextContentLine) && !/^[-]{3,}/.test(nextContentLine.trim())) {
-        inTable = false;
-      }
-      continue;
-    }
-
-    // Only collapse spaces outside tables
-    if (!inTable) {
-      lines[i] = line.replace(/  +/g, ' ');
-    }
-  }
-
-  return lines.join('\n');
-}
-
-/**
- * Check if text contains any CriticMarkup annotations
- * @param text - Text to check
- * @returns True if text contains any annotations
- * @throws TypeError If text is not a string
- */
-// Non-global copies for membership tests. `.test()` on a global regex advances
-// its lastIndex, so reusing PATTERNS here would make the result depend on prior
-// calls and intermittently miss real annotations.
-const ANNOTATION_TESTERS = [
-  PATTERNS.insert,
-  PATTERNS.delete,
-  PATTERNS.substitute,
-  PATTERNS.comment,
-  PATTERNS.highlight,
-].map((re) => new RegExp(re.source, re.flags.replace('g', '')));
-
-export function hasAnnotations(text: string): boolean {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  return ANNOTATION_TESTERS.some((re) => re.test(text));
-}
-
-/**
- * Replace the specific annotation occurrence at `position` rather than the
- * first textual match. Two identical annotations (e.g. the same `{++the++}`
- * twice) would otherwise have the wrong one edited. Falls back to a
- * first-occurrence replace only if the recorded position no longer lines up.
- */
-function replaceAnnotationAt(text: string, match: string, position: number, replacement: string): string {
-  if (position >= 0 && text.startsWith(match, position)) {
-    return text.slice(0, position) + replacement + text.slice(position + match.length);
-  }
-  return text.replace(match, replacement);
-}
-
-/**
- * Apply a decision to a single annotation (accept or reject)
- * @param text - Document text containing the annotation
- * @param annotation - Annotation object from parseAnnotations()
- * @param accept - True to accept the change, false to reject
- * @returns Updated text with the decision applied
- * @throws TypeError If text is not a string or annotation is invalid
- */
-export function applyDecision(text: string, annotation: Annotation, accept: boolean): string {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-  if (!annotation || typeof annotation.type !== 'string' || typeof annotation.match !== 'string') {
-    throw new TypeError('annotation must have type and match properties');
-  }
-  let replacement: string;
-
-  // Extract any comments embedded in the annotation content
-  // These should be preserved when accepting deletions or rejecting insertions
-  const commentPattern = /\{>>[\s\S]*?<<\}/g;
-  const embeddedComments = (annotation.match || '').match(commentPattern) || [];
-
-  switch (annotation.type) {
-    case 'insert':
-      if (accept) {
-        replacement = annotation.content;
-      } else {
-        // Rejecting insertion - preserve any comments that were inside
-        replacement = embeddedComments.length > 0 ? embeddedComments.join('') : '';
-      }
-      break;
-    case 'delete':
-      if (accept) {
-        // Accepting deletion - preserve any comments by placing them before
-        replacement = embeddedComments.length > 0 ? embeddedComments.join('') : '';
-      } else {
-        replacement = annotation.content;
-      }
-      break;
-    case 'substitute':
-      if (accept) {
-        // For substitutions, check if comments are in the old text being replaced
-        const oldTextComments = (annotation.content || '').match(commentPattern) || [];
-        replacement = annotation.replacement || '';
-        if (oldTextComments.length > 0) {
-          // Prepend comments that were in the old text
-          replacement = oldTextComments.join('') + replacement;
-        }
-      } else {
-        replacement = annotation.content;
-      }
-      break;
-    default:
-      return text;
-  }
-
-  return replaceAnnotationAt(text, annotation.match, annotation.position, replacement);
-}
-
-/**
- * Get track changes only (no comments)
- * @param text - Markdown text with CriticMarkup annotations
- * @returns Array of insert/delete/substitute annotations
- * @throws TypeError If text is not a string
- */
-export function getTrackChanges(text: string): Annotation[] {
-  // Input validation delegated to parseAnnotations
-  return parseAnnotations(text).filter((a) => a.type !== 'comment');
-}
-
-/**
- * Get comments only
- * @param text - Markdown text with CriticMarkup annotations
- * @param options - Filter options
- * @returns Array of comment annotations
- * @throws TypeError If text is not a string
- */
-export function getComments(text: string, options: CommentFilterOptions = {}): Annotation[] {
-  // Input validation delegated to parseAnnotations
-  const { pendingOnly = false, resolvedOnly = false } = options;
-  let comments = parseAnnotations(text).filter((a) => a.type === 'comment');
-
-  // Check for resolved status marker at end of comment
-  comments = comments.map((c) => {
-    const resolved = c.content.endsWith('[RESOLVED]') || c.content.endsWith('[✓]');
-    return {
-      ...c,
-      resolved,
-      content: resolved
-        ? c.content.replace(/\s*\[(RESOLVED|✓)\]$/, '').trim()
-        : c.content,
-    };
-  });
-
-  if (pendingOnly) {
-    comments = comments.filter((c) => !c.resolved);
-  }
-  if (resolvedOnly) {
-    comments = comments.filter((c) => c.resolved);
-  }
-
-  return comments;
-}
-
-/**
- * Mark a comment as resolved or pending
- * @param text - Document text containing the comment
- * @param comment - Comment annotation object from getComments()
- * @param resolved - True to mark resolved, false to mark pending
- * @returns Updated text with status marker applied
- * @throws TypeError If text is not a string or comment is invalid
- */
-export function setCommentStatus(text: string, comment: Annotation, resolved: boolean): string {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-  if (!comment || typeof comment.match !== 'string') {
-    throw new TypeError('comment must have a match property');
-  }
-  // Find the comment in the text
-  const originalMatch = comment.match;
-
-  if (resolved) {
-    // Add [RESOLVED] marker before the closing <<
-    const newMatch = originalMatch.replace(/<<\}$/, ' [RESOLVED]<<}');
-    return replaceAnnotationAt(text, originalMatch, comment.position, newMatch);
-  } else {
-    // Remove resolved markers
-    const newMatch = originalMatch.replace(/\s*\[(RESOLVED|✓)\]<<\}$/, '<<}');
-    return replaceAnnotationAt(text, originalMatch, comment.position, newMatch);
-  }
-}
-
-/**
- * Count annotations by type
- * @param text - Markdown text with CriticMarkup annotations
- * @returns Counts by annotation type
- * @throws TypeError If text is not a string
- */
-export function countAnnotations(text: string): AnnotationCounts {
-  // Input validation delegated to parseAnnotations
-  const annotations = parseAnnotations(text);
-  const counts: AnnotationCounts = { inserts: 0, deletes: 0, substitutes: 0, comments: 0, total: 0 };
-
-  for (const a of annotations) {
-    counts.total++;
-    switch (a.type) {
-      case 'insert':
-        counts.inserts++;
-        break;
-      case 'delete':
-        counts.deletes++;
-        break;
-      case 'substitute':
-        counts.substitutes++;
-        break;
-      case 'comment':
-        counts.comments++;
-        break;
-    }
-  }
-
-  return counts;
-}
-
-/**
- * Clean up orphaned/malformed CriticMarkup markers
- * This can happen when track changes span across comment boundaries
- * @param text - Document text with potentially malformed markers
- * @returns Cleaned text with orphaned markers removed
- * @throws TypeError If text is not a string
- */
-export function cleanupOrphanedMarkers(text: string): string {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-  let result = text;
-
-  // Remove orphaned insertion end markers (++} not preceded by {++)
-  // These occur when an insertion's start was inside something that got deleted/replaced
-  result = result.replace(/(?<!\{\+\+[^}]*)\+\+\}/g, '');
-
-  // Remove orphaned deletion end markers (--} not preceded by {--)
-  result = result.replace(/(?<!\{--[^}]*)--\}/g, '');
-
-  // Remove orphaned substitution end markers (~~} not preceded by {~~)
-  result = result.replace(/(?<!\{~~[^}]*)~~\}/g, '');
-
-  // Fix unclosed insertions: {++ without matching ++}
-  // Find {++ and check if there's a matching ++} before the next { marker
-  result = result.replace(/\{\+\+([^+]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, content) => {
-    // If content has no ++}, it's unclosed - just keep the content
-    if (!content.includes('++}')) {
-      return content;
-    }
-    return match;
-  });
-
-  // Fix unclosed deletions: {-- without matching --}
-  result = result.replace(/\{--([^-]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, content) => {
-    if (!content.includes('--}')) {
-      return content;
-    }
-    return match;
-  });
-
-  // Fix unclosed substitutions: {~~ without matching ~~}
-  // This is trickier because we need both ~> and ~~}
-  result = result.replace(/\{~~([^~]*?)~>([^~]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, old, newText) => {
-    if (!match.includes('~~}')) {
-      // Unclosed substitution - keep the new text
-      return newText;
-    }
-    return match;
-  });
-
-  return result;
-}
+/**
+ * CriticMarkup annotation parsing and manipulation
+ *
+ * Syntax:
+ *   {++inserted text++}     - Insertions
+ *   {--deleted text--}      - Deletions
+ *   {~~old~>new~~}          - Substitutions
+ *   {>>Author: comment<<}   - Comments
+ *   {==text==}              - Highlights
+ */
+
+import type { Annotation, AnnotationCounts, StripOptions, CommentFilterOptions } from './types.js';
+
+// =============================================================================
+// Constants
+// =============================================================================
+
+/** Window size for context lookup (characters before/after position) */
+const CONTEXT_WINDOW_SIZE = 2000;
+
+/** Characters of context to include in annotation results */
+const CONTEXT_SNIPPET_SIZE = 50;
+
+/** Maximum iterations for nested annotation stripping (safety limit) */
+const MAX_STRIP_ITERATIONS = 20;
+
+/** Maximum author name length in comments */
+const MAX_AUTHOR_LENGTH = 30;
+
+/** Maximum content length before heuristic assumes it's not a comment */
+const MAX_COMMENT_CONTENT_LENGTH = 200;
+
+// =============================================================================
+// Patterns
+// =============================================================================
+
+// Patterns for each annotation type
+const PATTERNS = {
+  insert: /\{\+\+(.+?)\+\+\}/gs,
+  delete: /\{--(.+?)--\}/gs,
+  substitute: /\{~~(.+?)~>(.+?)~~\}/gs,
+  comment: /\{>>(.+?)<<\}/gs,
+  highlight: /\{==(.+?)==\}/gs,
+};
+
+/**
+ * Check if a potential comment is actually a false positive
+ * (e.g., figure caption, nested inside other annotation, code block, etc.)
+ * @param commentContent - The content inside {>>...<<}
+ * @param fullText - The full document text
+ * @param position - Position of the comment in the text
+ * @returns true if this is a false positive (not a real comment)
+ */
+function isCommentFalsePositive(commentContent: string, fullText: string, position: number): boolean {
+  // Check if inside a code block (fenced or indented)
+  const textBefore = fullText.slice(Math.max(0, position - CONTEXT_WINDOW_SIZE), position);
+  const textAfter = fullText.slice(position, Math.min(fullText.length, position + CONTEXT_WINDOW_SIZE));
+
+  // Count unclosed fenced code blocks (``` or ~~~)
+  const fenceOpens = (textBefore.match(/^```|^~~~/gm) || []).length;
+  const fenceCloses = (textBefore.match(/```$|~~~$/gm) || []).length;
+  if (fenceOpens > fenceCloses) return true; // Inside code block
+
+  // Check if on an indented line (4+ spaces or tab at line start = code)
+  const lineStart = textBefore.lastIndexOf('\n') + 1;
+  const linePrefix = fullText.slice(lineStart, position);
+  if (/^(\t|    )/.test(linePrefix)) return true; // Indented code
+
+  // Check if inside inline code backticks
+  const backticksBefore = (linePrefix.match(/`/g) || []).length;
+  if (backticksBefore % 2 === 1) return true; // Inside inline code
+
+  // Check if nested inside a deletion or insertion block
+  const nearTextBefore = fullText.slice(Math.max(0, position - 500), position);
+
+  // Count unclosed deletion markers
+  const delOpens = (nearTextBefore.match(/\{--/g) || []).length;
+  const delCloses = (nearTextBefore.match(/--\}/g) || []).length;
+  if (delOpens > delCloses) return true; // Nested inside deletion
+
+  // Count unclosed insertion markers
+  const insOpens = (nearTextBefore.match(/\{\+\+/g) || []).length;
+  const insCloses = (nearTextBefore.match(/\+\+\}/g) || []).length;
+  if (insOpens > insCloses) return true; // Nested inside insertion
+
+  // Heuristics for figure captions and other false positives:
+
+  // Contains image/figure path patterns
+  if (/\(figures?\/|\(images?\/|\.png|\.jpg|\.jpeg|\.gif|\.svg|\.pdf/i.test(commentContent)) return true;
+
+  // Contains markdown figure reference syntax
+  if (/\{#fig:|!\[/.test(commentContent)) return true;
+
+  // Real comments typically have "Author:" at start. Accept hyphens, apostrophes,
+  // periods, and Unicode letters so names like "Jens-Christian Svenning" or
+  // "Camilla T Colding-Jørgensen" don't get rejected. See gcol33/docrev#1.
+  const hasAuthorPrefix = /^[\p{L}][\p{L}\s\-'.]{0,30}:\s/u.test(commentContent.trim());
+  const hasResolvedMark = /^[✓✔]\s/.test(commentContent.trim());
+
+  // Contains URL patterns (likely a link, not a comment) — only filter when
+  // there is no real author prefix, since reviewers legitimately cite URLs/DOIs.
+  if (!hasAuthorPrefix && /https?:\/\/|www\./i.test(commentContent) && commentContent.length < 150) return true;
+
+  // Looks like code (contains programming patterns)
+  if (/function\s*\(|=>|import\s+|export\s+|const\s+|let\s+|var\s+/.test(commentContent)) return true;
+
+  // Very long without clear author pattern (likely caption, not comment)
+  if (!hasAuthorPrefix && !hasResolvedMark && commentContent.length > MAX_COMMENT_CONTENT_LENGTH) return true;
+
+  // Looks like a figure caption (starts with "Fig" or contains typical caption words)
+  if (/^(Fig\.?|Figure|Table|Sankey|Diagram|Proportion|Distribution|Map|Chart|Graph|Plot|Panel)/i.test(commentContent.trim())) {
+    return true;
+  }
+
+  // Contains LaTeX-like patterns (likely equation, not comment)
+  if (/\\[a-z]+\{|\\frac|\\sum|\\int|\\begin\{/.test(commentContent)) return true;
+
+  // Looks like BibTeX entry (not a comment)
+  if (/@article\{|@book\{|@inproceedings\{/i.test(commentContent)) return true;
+
+  return false;
+}
+
+// Combined pattern for any track change (not comments)
+const TRACK_CHANGE_PATTERN = /(\{\+\+.+?\+\+\}|\{--.+?--\}|\{~~.+?~>.+?~~\})/gs;
+
+/**
+ * Reject binary content before it reaches the CriticMarkup regexes. A `.docx`
+ * is a ZIP; read as UTF-8 it starts with `PK\x03\x04` and is riddled with NUL
+ * bytes from the DEFLATE stream. Running the annotation patterns over that
+ * yields accidental byte-matches that look like a small, plausible count —
+ * silent garbage. This guard turns that whole failure class into a loud error.
+ * Callers that legitimately handle Word documents route them through
+ * `readAnnotatedInput` (lib/input.ts), which converts the docx to CriticMarkup
+ * first, so this only fires on a raw binary that reached a text-only path.
+ *
+ * UTF-8 Markdown never contains a NUL byte, so the check cannot false-positive
+ * on real input; it scans only the head to stay O(1) on large documents.
+ * @throws Error if the text is binary rather than text/Markdown.
+ */
+function assertNotBinary(text: string): void {
+  // ZIP local-file-header magic "PK\x03\x04" (docx/xlsx/pptx read as text),
+  // or any NUL byte in the head: a reliable binary marker absent from UTF-8
+  // Markdown, so this never false-positives on real input. Scans only the
+  // head to stay cheap on large documents.
+  if (text.startsWith('PK\u0003\u0004') || text.slice(0, 8192).includes('\u0000')) {
+    throw new Error(
+      'Expected text/Markdown but received binary content. If this is a Word ' +
+        'document, convert it first with "rev import <docx>" (rev status/comments ' +
+        'read .docx directly; other commands need the imported Markdown).',
+    );
+  }
+}
+
+// =============================================================================
+// Public API
+// =============================================================================
+
+/**
+ * Parse all annotations from text
+ * @param text - Markdown text containing CriticMarkup annotations
+ * @returns Array of parsed annotations sorted by position
+ * @throws TypeError If text is not a string
+ */
+export function parseAnnotations(text: string): Annotation[] {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  assertNotBinary(text);
+
+  const annotations: Annotation[] = [];
+
+  // Build line number lookup
+  const lines = text.split('\n');
+  let pos = 0;
+  const lineStarts = lines.map((line) => {
+    const start = pos;
+    pos += line.length + 1;
+    return start;
+  });
+
+  function getLine(position: number): number {
+    for (let i = 0; i < lineStarts.length; i++) {
+      const start = lineStarts[i];
+      if (start !== undefined && start > position) return i;
+    }
+    return lineStarts.length;
+  }
+
+  function getContext(position: number, length: number): { before: string; after: string } {
+    const start = Math.max(0, position - CONTEXT_SNIPPET_SIZE);
+    const end = Math.min(text.length, position + length + CONTEXT_SNIPPET_SIZE);
+    const before = text.slice(start, position).split('\n').pop() || '';
+    const after = text.slice(position + length, end).split('\n')[0] || '';
+    return { before, after };
+  }
+
+  // Parse insertions
+  for (const match of text.matchAll(PATTERNS.insert)) {
+    if (match.index === undefined) continue;
+    const ctx = getContext(match.index, match[0].length);
+    annotations.push({
+      type: 'insert',
+      match: match[0],
+      content: match[1] || '',
+      position: match.index,
+      line: getLine(match.index),
+      ...ctx,
+    });
+  }
+
+  // Parse deletions
+  for (const match of text.matchAll(PATTERNS.delete)) {
+    if (match.index === undefined) continue;
+    const ctx = getContext(match.index, match[0].length);
+    annotations.push({
+      type: 'delete',
+      match: match[0],
+      content: match[1] || '',
+      position: match.index,
+      line: getLine(match.index),
+      ...ctx,
+    });
+  }
+
+  // Parse substitutions
+  for (const match of text.matchAll(PATTERNS.substitute)) {
+    if (match.index === undefined) continue;
+    const ctx = getContext(match.index, match[0].length);
+    annotations.push({
+      type: 'substitute',
+      match: match[0],
+      content: match[1] || '',
+      replacement: match[2] || '',
+      position: match.index,
+      line: getLine(match.index),
+      ...ctx,
+    });
+  }
+
+  // Parse comments (with false positive filtering)
+  for (const match of text.matchAll(PATTERNS.comment)) {
+    if (match.index === undefined) continue;
+    // Skip false positives (figure captions, nested annotations, etc.)
+    const commentContent = match[1] || '';
+    if (isCommentFalsePositive(commentContent, text, match.index)) {
+      continue;
+    }
+
+    const ctx = getContext(match.index, match[0].length);
+    let commentText = commentContent;
+    let author = '';
+
+    // Extract author if present (format: "Author: comment")
+    const colonIdx = commentText.indexOf(':');
+    if (colonIdx > 0 && colonIdx < MAX_AUTHOR_LENGTH) {
+      author = commentText.slice(0, colonIdx).trim();
+      commentText = commentText.slice(colonIdx + 1).trim();
+    }
+
+    annotations.push({
+      type: 'comment',
+      match: match[0],
+      content: commentText,
+      author,
+      position: match.index,
+      line: getLine(match.index),
+      ...ctx,
+    });
+  }
+
+  // Sort by position
+  annotations.sort((a, b) => a.position - b.position);
+  return annotations;
+}
+
+/**
+ * Strip annotations from text, applying changes
+ * Handles nested annotations by iterating until stable
+ * @param text - Markdown text with CriticMarkup annotations
+ * @param options - Strip options
+ * @returns Clean text with annotations applied/removed
+ * @throws TypeError If text is not a string
+ */
+export function stripAnnotations(text: string, options: StripOptions = {}): string {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  assertNotBinary(text);
+
+  const { keepComments = false } = options;
+
+  // Iterate until no more changes (handles nested annotations)
+  let prev: string;
+  let iterations = 0;
+
+  do {
+    prev = text;
+
+    // Apply substitutions: {~~old~>new~~} → new
+    text = text.replace(PATTERNS.substitute, '$2');
+
+    // Apply insertions: {++text++} → text
+    text = text.replace(PATTERNS.insert, '$1');
+
+    // Apply deletions: {--text--} → nothing
+    // Don't touch surrounding whitespace - just remove the annotation
+    text = text.replace(PATTERNS.delete, '');
+
+    // Remove highlights: {==text==} → text
+    text = text.replace(PATTERNS.highlight, '$1');
+
+    // Remove comments unless keeping
+    if (!keepComments) {
+      text = text.replace(PATTERNS.comment, '');
+    }
+
+    // Strip pandoc highlight spans: [text]{.mark} → text.
+    // When `keepComments=true`, preserve `[anchor]{.mark}` that is the
+    // anchor of a kept `{>>...<<}` comment. The dual-build flow runs
+    // stripAnnotations() before prepareMarkdownWithMarkers(), and stripping
+    // the anchor span here would leave the marker generator with no anchor
+    // text — collapsing every multi-word anchor to a single fallback word
+    // in the rebuilt docx.
+    text = keepComments
+      ? text.replace(/(?<!<<\}\s{0,3})\[([^\]]*)\]\{\.mark\}/g, '$1')
+      : text.replace(/\[([^\]]*)\]\{\.mark\}/g, '$1');
+
+    // Clean up partial/orphaned markers within the loop
+    // This handles cases where nested annotations leave behind fragments
+
+    // Empty annotations (from nested stripping)
+    text = text.replace(/\{----\}/g, '');
+    text = text.replace(/\{\+\+\+\+\}/g, '');
+    text = text.replace(/\{--\s*--\}/g, '');
+    text = text.replace(/\{\+\+\s*\+\+\}/g, '');
+
+    // Orphaned substitution fragments: ~>text~~} or {~~text (no proper pairs)
+    text = text.replace(/~>[^{]*?~~\}/g, '');
+    text = text.replace(/\{~~[^~}]*$/gm, '');
+
+    // Handle malformed substitution from nested: {~~{~~old → just strip the {~~
+    text = text.replace(/\{~~\{~~/g, '{~~');
+    text = text.replace(/~~\}~~\}/g, '~~}');
+
+    iterations++;
+  } while (text !== prev && iterations < MAX_STRIP_ITERATIONS);
+
+  // Final cleanup of any remaining orphaned markers
+  // Orphaned closing markers
+  text = text.replace(/--\}(?:--\})+/g, '');
+  text = text.replace(/\+\+\}(?:\+\+\})+/g, '');
+  text = text.replace(/~~\}(?:~~\})+/g, '');
+  text = text.replace(/--\}/g, '');
+  text = text.replace(/\+\+\}/g, '');
+  text = text.replace(/~~\}/g, '');
+
+  // Orphaned opening markers
+  text = text.replace(/\{--(?:\{--)+/g, '');
+  text = text.replace(/\{\+\+(?:\{\+\+)+/g, '');
+  text = text.replace(/\{~~(?:\{~~)+/g, '');
+  text = text.replace(/\{--/g, '');
+  text = text.replace(/\{\+\+/g, '');
+  text = text.replace(/\{~~/g, '');
+  text = text.replace(/~>/g, '');
+
+  // Remove orphan [ from stripped {.mark} spans where the closing ]{.mark}
+  // was inside a comment. A [ is orphan if no `]` follows before end of line.
+  // We deliberately allow other `[` between the candidate and the matching `]`
+  // — otherwise nested forms like `[[0..9]]{.mark}` would have their outer
+  // `[` stripped because the lookahead saw the inner `[` as a barrier.
+  text = text.replace(/\[(?![^\]\n]*\])/g, '');
+
+  return text;
+}
+
+/**
+ * Collapse multiple spaces to single space, preserving table formatting
+ * Useful for cleaning up messy Word imports
+ * @param text - Text to normalize
+ * @returns Text with multiple spaces collapsed to single spaces
+ * @throws TypeError If text is not a string
+ */
+export function stripToSingleSpace(text: string): string {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+
+  const lines = text.split('\n');
+  let inTable = false;
+
+  // Helper to check if a line looks like table content
+  const looksLikeTableRow = (ln: string): boolean => {
+    const trimmed = ln.trim();
+    if (!trimmed) return false;
+    // Has multiple consecutive spaces (column spacing)
+    // OR italicized category header with trailing spaces
+    return /\S\s{2,}\S/.test(trimmed) || (/^\*[^*]+\*\s*$/.test(trimmed) && /\s{2,}$/.test(ln));
+  };
+
+  for (let i = 0; i < lines.length; i++) {
+    const line = lines[i];
+    if (!line) continue;
+
+    // Detect table separator line
+    const isTableSeparator = /^\|?[\s-]*[-]{3,}[\s|:-]+[-]{3,}/.test(line) ||
+                              /^[-]{3,}\s{2,}[-]{3,}/.test(line);
+
+    if (isTableSeparator) {
+      inTable = true;
+      continue;
+    }
+
+    // Check if we're exiting the table
+    if (inTable && line.trim() === '') {
+      let nextContentLine = '';
+      for (let j = i + 1; j < lines.length && j < i + 5; j++) {
+        const nextLine = lines[j];
+        if (nextLine && nextLine.trim() !== '') {
+          nextContentLine = nextLine;
+          break;
+        }
+      }
+      if (!looksLikeTableRow(nextContentLine) && !/^[-]{3,}/.test(nextContentLine.trim())) {
+        inTable = false;
+      }
+      continue;
+    }
+
+    // Only collapse spaces outside tables
+    if (!inTable) {
+      lines[i] = line.replace(/  +/g, ' ');
+    }
+  }
+
+  return lines.join('\n');
+}
+
+/**
+ * Check if text contains any CriticMarkup annotations
+ * @param text - Text to check
+ * @returns True if text contains any annotations
+ * @throws TypeError If text is not a string
+ */
+// Non-global copies for membership tests. `.test()` on a global regex advances
+// its lastIndex, so reusing PATTERNS here would make the result depend on prior
+// calls and intermittently miss real annotations.
+const ANNOTATION_TESTERS = [
+  PATTERNS.insert,
+  PATTERNS.delete,
+  PATTERNS.substitute,
+  PATTERNS.comment,
+  PATTERNS.highlight,
+].map((re) => new RegExp(re.source, re.flags.replace('g', '')));
+
+export function hasAnnotations(text: string): boolean {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  assertNotBinary(text);
+
+  return ANNOTATION_TESTERS.some((re) => re.test(text));
+}
+
+/**
+ * Replace the specific annotation occurrence at `position` rather than the
+ * first textual match. Two identical annotations (e.g. the same `{++the++}`
+ * twice) would otherwise have the wrong one edited. Falls back to a
+ * first-occurrence replace only if the recorded position no longer lines up.
+ */
+function replaceAnnotationAt(text: string, match: string, position: number, replacement: string): string {
+  if (position >= 0 && text.startsWith(match, position)) {
+    return text.slice(0, position) + replacement + text.slice(position + match.length);
+  }
+  return text.replace(match, replacement);
+}
+
+/**
+ * Apply a decision to a single annotation (accept or reject)
+ * @param text - Document text containing the annotation
+ * @param annotation - Annotation object from parseAnnotations()
+ * @param accept - True to accept the change, false to reject
+ * @returns Updated text with the decision applied
+ * @throws TypeError If text is not a string or annotation is invalid
+ */
+export function applyDecision(text: string, annotation: Annotation, accept: boolean): string {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  if (!annotation || typeof annotation.type !== 'string' || typeof annotation.match !== 'string') {
+    throw new TypeError('annotation must have type and match properties');
+  }
+  let replacement: string;
+
+  // Extract any comments embedded in the annotation content
+  // These should be preserved when accepting deletions or rejecting insertions
+  const commentPattern = /\{>>[\s\S]*?<<\}/g;
+  const embeddedComments = (annotation.match || '').match(commentPattern) || [];
+
+  switch (annotation.type) {
+    case 'insert':
+      if (accept) {
+        replacement = annotation.content;
+      } else {
+        // Rejecting insertion - preserve any comments that were inside
+        replacement = embeddedComments.length > 0 ? embeddedComments.join('') : '';
+      }
+      break;
+    case 'delete':
+      if (accept) {
+        // Accepting deletion - preserve any comments by placing them before
+        replacement = embeddedComments.length > 0 ? embeddedComments.join('') : '';
+      } else {
+        replacement = annotation.content;
+      }
+      break;
+    case 'substitute':
+      if (accept) {
+        // For substitutions, check if comments are in the old text being replaced
+        const oldTextComments = (annotation.content || '').match(commentPattern) || [];
+        replacement = annotation.replacement || '';
+        if (oldTextComments.length > 0) {
+          // Prepend comments that were in the old text
+          replacement = oldTextComments.join('') + replacement;
+        }
+      } else {
+        replacement = annotation.content;
+      }
+      break;
+    default:
+      return text;
+  }
+
+  return replaceAnnotationAt(text, annotation.match, annotation.position, replacement);
+}
+
+/**
+ * Get track changes only (no comments)
+ * @param text - Markdown text with CriticMarkup annotations
+ * @returns Array of insert/delete/substitute annotations
+ * @throws TypeError If text is not a string
+ */
+export function getTrackChanges(text: string): Annotation[] {
+  // Input validation delegated to parseAnnotations
+  return parseAnnotations(text).filter((a) => a.type !== 'comment');
+}
+
+/**
+ * Get comments only
+ * @param text - Markdown text with CriticMarkup annotations
+ * @param options - Filter options
+ * @returns Array of comment annotations
+ * @throws TypeError If text is not a string
+ */
+export function getComments(text: string, options: CommentFilterOptions = {}): Annotation[] {
+  // Input validation delegated to parseAnnotations
+  const { pendingOnly = false, resolvedOnly = false } = options;
+  let comments = parseAnnotations(text).filter((a) => a.type === 'comment');
+
+  // Check for resolved status marker at end of comment
+  comments = comments.map((c) => {
+    const resolved = c.content.endsWith('[RESOLVED]') || c.content.endsWith('[✓]');
+    return {
+      ...c,
+      resolved,
+      content: resolved
+        ? c.content.replace(/\s*\[(RESOLVED|✓)\]$/, '').trim()
+        : c.content,
+    };
+  });
+
+  if (pendingOnly) {
+    comments = comments.filter((c) => !c.resolved);
+  }
+  if (resolvedOnly) {
+    comments = comments.filter((c) => c.resolved);
+  }
+
+  return comments;
+}
+
+/**
+ * Mark a comment as resolved or pending
+ * @param text - Document text containing the comment
+ * @param comment - Comment annotation object from getComments()
+ * @param resolved - True to mark resolved, false to mark pending
+ * @returns Updated text with status marker applied
+ * @throws TypeError If text is not a string or comment is invalid
+ */
+export function setCommentStatus(text: string, comment: Annotation, resolved: boolean): string {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  if (!comment || typeof comment.match !== 'string') {
+    throw new TypeError('comment must have a match property');
+  }
+  // Find the comment in the text
+  const originalMatch = comment.match;
+
+  if (resolved) {
+    // Add [RESOLVED] marker before the closing <<
+    const newMatch = originalMatch.replace(/<<\}$/, ' [RESOLVED]<<}');
+    return replaceAnnotationAt(text, originalMatch, comment.position, newMatch);
+  } else {
+    // Remove resolved markers
+    const newMatch = originalMatch.replace(/\s*\[(RESOLVED|✓)\]<<\}$/, '<<}');
+    return replaceAnnotationAt(text, originalMatch, comment.position, newMatch);
+  }
+}
+
+/**
+ * Count annotations by type
+ * @param text - Markdown text with CriticMarkup annotations
+ * @returns Counts by annotation type
+ * @throws TypeError If text is not a string
+ */
+export function countAnnotations(text: string): AnnotationCounts {
+  // Input validation delegated to parseAnnotations
+  const annotations = parseAnnotations(text);
+  const counts: AnnotationCounts = { inserts: 0, deletes: 0, substitutes: 0, comments: 0, total: 0 };
+
+  for (const a of annotations) {
+    counts.total++;
+    switch (a.type) {
+      case 'insert':
+        counts.inserts++;
+        break;
+      case 'delete':
+        counts.deletes++;
+        break;
+      case 'substitute':
+        counts.substitutes++;
+        break;
+      case 'comment':
+        counts.comments++;
+        break;
+    }
+  }
+
+  return counts;
+}
+
+/**
+ * Clean up orphaned/malformed CriticMarkup markers
+ * This can happen when track changes span across comment boundaries
+ * @param text - Document text with potentially malformed markers
+ * @returns Cleaned text with orphaned markers removed
+ * @throws TypeError If text is not a string
+ */
+export function cleanupOrphanedMarkers(text: string): string {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+  let result = text;
+
+  // Remove orphaned insertion end markers (++} not preceded by {++)
+  // These occur when an insertion's start was inside something that got deleted/replaced
+  result = result.replace(/(?<!\{\+\+[^}]*)\+\+\}/g, '');
+
+  // Remove orphaned deletion end markers (--} not preceded by {--)
+  result = result.replace(/(?<!\{--[^}]*)--\}/g, '');
+
+  // Remove orphaned substitution end markers (~~} not preceded by {~~)
+  result = result.replace(/(?<!\{~~[^}]*)~~\}/g, '');
+
+  // Fix unclosed insertions: {++ without matching ++}
+  // Find {++ and check if there's a matching ++} before the next { marker
+  result = result.replace(/\{\+\+([^+]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, content) => {
+    // If content has no ++}, it's unclosed - just keep the content
+    if (!content.includes('++}')) {
+      return content;
+    }
+    return match;
+  });
+
+  // Fix unclosed deletions: {-- without matching --}
+  result = result.replace(/\{--([^-]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, content) => {
+    if (!content.includes('--}')) {
+      return content;
+    }
+    return match;
+  });
+
+  // Fix unclosed substitutions: {~~ without matching ~~}
+  // This is trickier because we need both ~> and ~~}
+  result = result.replace(/\{~~([^~]*?)~>([^~]*?)(?=\{[+\-~>]|\{>>|$)/g, (match, old, newText) => {
+    if (!match.includes('~~}')) {
+      // Unclosed substitution - keep the new text
+      return newText;
+    }
+    return match;
+  });
+
+  return result;
+}