docrev 0.9.11 → 0.9.14

package/lib/crossref.ts

@@ -1,781 +1,781 @@
-/**
- * Cross-reference handling - dynamic figure/table references
- *
- * Enables:
- * - @fig:label syntax in source (auto-numbered)
- * - Conversion to "Figure 1" in Word output
- * - Auto-conversion back during import
- */
-
-import * as fs from 'fs';
-import * as path from 'path';
-import YAML from 'yaml';
-import type {
-  RefNumber,
-  HardcodedRef,
-  DynamicRef,
-  FigureInfo,
-  Registry,
-  RefStatus,
-  ConversionResult,
-} from './types.js';
-
-// =============================================================================
-// Constants
-// =============================================================================
-
-/** Characters of context to check before a reference for deduplication */
-const REF_CONTEXT_WINDOW = 100;
-
-/** Minimum word length for similarity calculations */
-const MIN_WORD_LENGTH = 2;
-
-// =============================================================================
-// Type Definitions (Internal)
-// =============================================================================
-
-/**
- * Reference info (internal use in registry building)
- */
-interface RefInfo {
-  label: string;
-  num: number;
-  isSupp: boolean;
-  file: string;
-}
-
-/**
- * Parsed reference number components
- */
-interface ParsedRefNumber {
-  isSupp: boolean;
-  num: number;
-  suffix: string | null;
-}
-
-/**
- * Detected reference with parsed numbers
- */
-interface DetectedRef {
-  type: 'fig' | 'tbl' | 'eq';
-  match: string;
-  numbers: ParsedRefNumber[];
-  position: number;
-}
-
-// =============================================================================
-// Internal Helpers
-// =============================================================================
-
-/**
- * Discover section files from a directory by reading config files
- * Only returns files explicitly defined in rev.yaml or sections.yaml
- * Returns empty array if no config found (caller should handle this)
- */
-function discoverSectionFiles(directory: string): string[] {
-  // Try rev.yaml first
-  const revYamlPath = path.join(directory, 'rev.yaml');
-  if (fs.existsSync(revYamlPath)) {
-    try {
-      const config = YAML.parse(fs.readFileSync(revYamlPath, 'utf-8'));
-      if (config.sections && Array.isArray(config.sections) && config.sections.length > 0) {
-        return config.sections.filter((f: string) => fs.existsSync(path.join(directory, f)));
-      }
-    } catch (e) {
-      if (process.env.DEBUG) {
-        console.warn('crossref: YAML parse error in rev.yaml:', (e as Error).message);
-      }
-    }
-  }
-
-  // Try sections.yaml
-  const sectionsPath = path.join(directory, 'sections.yaml');
-  if (fs.existsSync(sectionsPath)) {
-    try {
-      const config = YAML.parse(fs.readFileSync(sectionsPath, 'utf-8'));
-      if (config.sections) {
-        const sectionOrder = Object.entries(config.sections)
-          .sort((a, b) => ((a[1] as any).order ?? 999) - ((b[1] as any).order ?? 999))
-          .map(([file]) => file);
-        return sectionOrder.filter((f) => fs.existsSync(path.join(directory, f)));
-      }
-    } catch (e) {
-      if (process.env.DEBUG) {
-        console.warn('crossref: YAML parse error in sections.yaml:', (e as Error).message);
-      }
-    }
-  }
-
-  // No config found - return empty array
-  // Caller must handle this (either error or use explicit sections)
-  return [];
-}
-
-// =============================================================================
-// Detection Patterns
-// =============================================================================
-
-/**
- * Patterns for detecting hardcoded references
- * Matches complex patterns including:
- * - Simple: "Figure 1", "Fig. 2a", "Table S1"
- * - Ranges: "Figures 1-3", "Fig. 1a-c", "Figs. 1a-3b"
- * - Lists: "Figures 1, 2, and 3", "Fig. 1a, b, c", "Tables 1 & 2"
- * - Mixed: "Figs. 1, 3-5, and 7"
- *
- * Uses a simpler base pattern and parses the full match for lists
- */
-const DETECTION_PATTERNS: Record<string, RegExp> = {
-  // Captures the full reference including lists with "and"
-  // Group 1: type prefix (Figure, Fig., etc.)
-  // Group 2: reference list (parsed by parseReferenceList())
-  // Matches: "1", "1a", "1-3", "1a-c", "1, 2, 3", "1 and 2", "1, 2 and 3", "1, 2, and 3"
-  // Separator: comma/dash/ampersand, optionally followed by "and"
-  // Standalone letters must be followed by separator, punctuation, or word boundary
-  // Also handles: "see Figure 1", "(Fig. 1)", "in Figures 1–3"
-  // Note: 'gi' flag makes these case-insensitive, so "figure 1" is also matched
-  figure: /\b(Figures?|Figs?\.?)\s+((?:\d+|S\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+|S\d+)[a-z]?|[a-z]\b))*)/gi,
-
-  table: /\b(Tables?|Tabs?\.?)\s+((?:\d+|S\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+|S\d+)[a-z]?|[a-z]\b))*)/gi,
-
-  equation: /\b(Equations?|Eqs?\.?)\s+((?:\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+)[a-z]?|[a-z]\b))*)/gi,
-};
-
-/**
- * Patterns to EXCLUDE from detection (false positives)
- * These look like references but aren't (e.g., "Table of Contents", "Figure skating")
- */
-const EXCLUSION_PATTERNS = [
-  /\bTable\s+of\s+Contents?\b/gi,
-  /\bFigure\s+skating\b/gi,
-  /\bFigure\s+out\b/gi,
-  /\bFigure\s+it\b/gi,
-  /\bTable\s+setting/gi,
-  /\bEquation\s+editor\b/gi,
-];
-
-/**
- * Pattern for extracting anchors from markdown: {#fig:label}, {#tbl:label}
- */
-const ANCHOR_PATTERN = /\{#(fig|tbl|eq):([a-zA-Z0-9_-]+)/gi;
-
-/**
- * Pattern for @-style references: @fig:label, @tbl:label
- */
-const REF_PATTERN = /@(fig|tbl|eq):([a-zA-Z0-9_-]+)/gi;
-
-// =============================================================================
-// Public API
-// =============================================================================
-
-/**
- * Normalize a reference type to standard form
- */
-export function normalizeType(typeStr: string): 'fig' | 'tbl' | 'eq' | string {
-  if (typeof typeStr !== 'string') {
-    throw new TypeError(`typeStr must be a string, got ${typeof typeStr}`);
-  }
-  const lower = typeStr.toLowerCase().replace(/\.$/, '');
-  if (lower.startsWith('fig')) return 'fig';
-  if (lower.startsWith('tab')) return 'tbl';
-  if (lower.startsWith('eq')) return 'eq';
-  return lower;
-}
-
-/**
- * Parse a reference number, handling supplementary (S1, S2) and letter suffixes (1a, 1b)
- */
-export function parseRefNumber(numStr: string, suffix: string | null = null): ParsedRefNumber {
-  if (!numStr || typeof numStr !== 'string') {
-    return { isSupp: false, num: 0, suffix: suffix || null };
-  }
-  const isSupp = numStr.toUpperCase().startsWith('S');
-  const numPart = isSupp ? numStr.slice(1) : numStr;
-  // Extract suffix if embedded in numStr (e.g., "1a")
-  const match = numPart.match(/^(\d+)([a-z])?$/i);
-  const num = match && match[1] ? parseInt(match[1], 10) : parseInt(numPart, 10);
-  const extractedSuffix = suffix || (match && match[2]) || null;
-  return { isSupp, num, suffix: extractedSuffix ? extractedSuffix.toLowerCase() : null };
-}
-
-/**
- * Parse a reference list string like "1, 2, and 3" or "1a-c" or "1a-3b"
- * Returns an array of {num, isSupp, suffix} objects
- */
-export function parseReferenceList(listStr: string): ParsedRefNumber[] {
-  const results: ParsedRefNumber[] = [];
-  if (!listStr || typeof listStr !== 'string') return results;
-
-  // Normalize: replace "and" with comma, normalize dashes
-  let normalized = listStr
-    .replace(/\s+and\s+/gi, ', ')
-    .replace(/[–—]/g, '-') // en-dash, em-dash → hyphen
-    .replace(/&/g, ', '); // & → comma
-
-  // Split by comma (but not by dash, which indicates ranges)
-  const parts = normalized.split(/\s*,\s*/).filter((p) => p.trim());
-
-  let lastFullRef: { num: number; isSupp: boolean } | null = null; // Track the last full reference for implicit prefixes
-
-  for (const part of parts) {
-    const trimmed = part.trim();
-    if (!trimmed) continue;
-
-    // Check if this is a range (contains -)
-    if (trimmed.includes('-')) {
-      const parts = trimmed.split('-').map((s) => s.trim());
-      const start = parts[0] || '';
-      const end = parts[1] || '';
-
-      // Check if end is just a letter (e.g., "1a-c" where end is "c")
-      const endIsLetterOnly = /^[a-z]$/i.test(end);
-
-      const startRef = parseRefNumber(start);
-      // For letter-only end, don't parse as number
-      const endRef = endIsLetterOnly
-        ? { num: startRef.num, isSupp: startRef.isSupp, suffix: end.toLowerCase() }
-        : parseRefNumber(end);
-
-      // Handle different range types:
-      // 1. Suffix-only range on same number: "1a-c" → 1a, 1b, 1c
-      // 2. Number range: "1-3" → 1, 2, 3
-      // 3. Cross-number suffix range: "1a-3b" → 1a...1z, 2a...2z, 3a, 3b (limited)
-
-      if (startRef.suffix && endRef.suffix && startRef.num !== endRef.num) {
-        // Cross-number suffix range: "1a-3b"
-        // For academic papers, limit intermediate figures to same suffix range
-        // e.g., "1a-3b" typically means 1a, 1b, 2a, 2b, 3a, 3b
-        const maxSuffix = Math.max(
-          startRef.suffix.charCodeAt(0),
-          endRef.suffix.charCodeAt(0)
-        );
-
-        for (let n = startRef.num; n <= endRef.num; n++) {
-          const suffixStart =
-            n === startRef.num ? startRef.suffix.charCodeAt(0) : 'a'.charCodeAt(0);
-          const suffixEnd = n === endRef.num ? endRef.suffix.charCodeAt(0) : maxSuffix;
-
-          for (let s = suffixStart; s <= suffixEnd; s++) {
-            results.push({
-              num: n,
-              isSupp: startRef.isSupp,
-              suffix: String.fromCharCode(s),
-            });
-          }
-        }
-        lastFullRef = { num: endRef.num, isSupp: startRef.isSupp };
-      } else if (startRef.suffix || endRef.suffix) {
-        // Suffix range on same number: "1a-c"
-        const num: number = startRef.num !== 0 ? startRef.num : (lastFullRef ? lastFullRef.num : 1);
-        const isSupp: boolean = startRef.isSupp ? startRef.isSupp : (lastFullRef ? lastFullRef.isSupp : false);
-        const startCode = (startRef.suffix || 'a').charCodeAt(0);
-        const endCode = (endRef.suffix || 'a').charCodeAt(0);
-
-        for (let code = startCode; code <= endCode; code++) {
-          results.push({
-            num,
-            isSupp,
-            suffix: String.fromCharCode(code),
-          });
-        }
-        lastFullRef = { num, isSupp };
-      } else {
-        // Pure number range: "1-3"
-        for (let n = startRef.num; n <= endRef.num; n++) {
-          results.push({
-            num: n,
-            isSupp: startRef.isSupp,
-            suffix: null,
-          });
-        }
-        lastFullRef = { num: endRef.num, isSupp: startRef.isSupp };
-      }
-    } else {
-      // Single reference or implicit suffix
-      // Check if it's just a letter (implicit prefix from previous number)
-      if (/^[a-z]$/i.test(trimmed) && lastFullRef) {
-        // Implicit prefix: "b" after "1a" means "1b"
-        results.push({
-          num: lastFullRef.num,
-          isSupp: lastFullRef.isSupp,
-          suffix: trimmed.toLowerCase(),
-        });
-      } else {
-        // Full reference: "1", "1a", "S1", "S1a"
-        const ref = parseRefNumber(trimmed);
-        results.push(ref);
-        lastFullRef = { num: ref.num, isSupp: ref.isSupp };
-      }
-    }
-  }
-
-  return results;
-}
-
-/**
- * Build a registry of figure/table labels from .md files
- * Scans for {#fig:label} and {#tbl:label} anchors
- *
- * IMPORTANT: This function requires either explicit sections or a rev.yaml/sections.yaml config.
- * It will NOT guess by scanning all .md files, as this leads to incorrect numbering
- * when temporary files (paper_clean.md, etc.) exist in the directory.
- */
-export function buildRegistry(directory: string, sections?: string[]): Registry {
-  if (typeof directory !== 'string') {
-    throw new TypeError(`directory must be a string, got ${typeof directory}`);
-  }
-
-  const figures = new Map<string, FigureInfo>();
-  const tables = new Map<string, FigureInfo>();
-  const equations = new Map<string, FigureInfo>();
-
-  // Counters for numbering (separate for main and supplementary)
-  let figNum = 0;
-  let figSuppNum = 0;
-  let tblNum = 0;
-  let tblSuppNum = 0;
-  let eqNum = 0;
-
-  let orderedFiles: string[];
-
-  if (Array.isArray(sections) && sections.length > 0) {
-    // Use explicitly provided section files - most reliable
-    orderedFiles = sections.filter((f) => fs.existsSync(path.join(directory, f)));
-  } else {
-    // Try to determine sections from config files (rev.yaml or sections.yaml)
-    orderedFiles = discoverSectionFiles(directory);
-    // If no config found, return empty registry rather than guessing
-    // This prevents bugs from scanning wrong files
-  }
-
-  // Determine if a file is supplementary
-  const isSupplementary = (filename: string): boolean =>
-    filename.toLowerCase().includes('supp') || filename.toLowerCase().includes('appendix');
-
-  // Process each file in order
-  for (const file of orderedFiles) {
-    const filePath = path.join(directory, file);
-    const content = fs.readFileSync(filePath, 'utf-8');
-    const isSupp = isSupplementary(file);
-
-    // Find all anchors
-    let match: RegExpExecArray | null;
-    ANCHOR_PATTERN.lastIndex = 0;
-    while ((match = ANCHOR_PATTERN.exec(content)) !== null) {
-      const typeRaw = match[1];
-      const labelRaw = match[2];
-      if (!typeRaw || !labelRaw) continue;
-
-      const type = typeRaw.toLowerCase();
-      const label = labelRaw;
-
-      if (type === 'fig') {
-        if (isSupp) {
-          figSuppNum++;
-          figures.set(label, { label, num: figSuppNum, isSupp: true, file });
-        } else {
-          figNum++;
-          figures.set(label, { label, num: figNum, isSupp: false, file });
-        }
-      } else if (type === 'tbl') {
-        if (isSupp) {
-          tblSuppNum++;
-          tables.set(label, { label, num: tblSuppNum, isSupp: true, file });
-        } else {
-          tblNum++;
-          tables.set(label, { label, num: tblNum, isSupp: false, file });
-        }
-      } else if (type === 'eq') {
-        eqNum++;
-        equations.set(label, { label, num: eqNum, isSupp: false, file });
-      }
-    }
-  }
-
-  // Build reverse lookup: number → label
-  const byNumber: Registry['byNumber'] = {
-    fig: new Map(),
-    figS: new Map(),
-    tbl: new Map(),
-    tblS: new Map(),
-    eq: new Map(),
-  };
-
-  for (const [label, info] of figures) {
-    const key = info.isSupp ? 'figS' : 'fig';
-    byNumber[key].set(info.num, label);
-  }
-  for (const [label, info] of tables) {
-    const key = info.isSupp ? 'tblS' : 'tbl';
-    byNumber[key].set(info.num, label);
-  }
-  for (const [label, info] of equations) {
-    byNumber.eq.set(info.num, label);
-  }
-
-  return { figures, tables, equations, byNumber };
-}
-
-/**
- * Get the display string for a label (e.g., "Figure 1", "Table S2")
- */
-export function labelToDisplay(
-  type: 'fig' | 'tbl' | 'eq',
-  label: string,
-  registry: Registry
-): string | null {
-  if (!registry || !registry.figures) return null;
-
-  const collection =
-    type === 'fig' ? registry.figures : type === 'tbl' ? registry.tables : registry.equations;
-
-  const info = collection.get(label);
-  if (!info) return null;
-
-  const prefix = type === 'fig' ? 'Figure' : type === 'tbl' ? 'Table' : 'Equation';
-  const numStr = info.isSupp ? `S${info.num}` : `${info.num}`;
-
-  return `${prefix} ${numStr}`;
-}
-
-/**
- * Get the label for a display number (e.g., "fig:heatmap" from Figure 1)
- */
-export function numberToLabel(
-  type: 'fig' | 'tbl' | 'eq',
-  num: number,
-  isSupp: boolean,
-  registry: Registry
-): string | null {
-  if (!registry || !registry.byNumber) return null;
-
-  const key = isSupp ? (`${type}S` as keyof Registry['byNumber']) : type;
-  return registry.byNumber[key]?.get(num) || null;
-}
-
-/**
- * Detect all hardcoded references in text
- */
-export function detectHardcodedRefs(text: string): DetectedRef[] {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  const refs: DetectedRef[] = [];
-
-  for (const [type, pattern] of Object.entries(DETECTION_PATTERNS)) {
-    pattern.lastIndex = 0;
-    let match: RegExpExecArray | null;
-
-    while ((match = pattern.exec(text)) !== null) {
-      // Pattern groups:
-      // [1] = type prefix (Figure, Fig., etc.)
-      // [2] = reference list string (e.g., "1, 2, and 3" or "1a-3b")
-
-      const listStr = match[2];
-      if (!listStr) continue;
-      const numbers = parseReferenceList(listStr);
-
-      // Skip if no valid numbers were parsed
-      if (numbers.length === 0) continue;
-
-      refs.push({
-        type: normalizeType(type) as 'fig' | 'tbl' | 'eq',
-        match: match[0],
-        numbers,
-        position: match.index,
-      });
-    }
-  }
-
-  // Sort by position
-  refs.sort((a, b) => a.position - b.position);
-  return refs;
-}
-
-/**
- * Convert hardcoded references to @-style references
- */
-export function convertHardcodedRefs(text: string, registry: Registry): ConversionResult {
-  // Input validation delegated to detectHardcodedRefs
-  const refs = detectHardcodedRefs(text);
-  const conversions: Array<{ from: string; to: string }> = [];
-  const warnings: string[] = [];
-
-  // Process in reverse order to preserve positions
-  let result = text;
-  for (let i = refs.length - 1; i >= 0; i--) {
-    const ref = refs[i];
-    if (!ref) continue;
-
-    // Build replacement
-    const labels: string[] = [];
-    for (const { num, isSupp } of ref.numbers) {
-      const label = numberToLabel(ref.type, num, isSupp, registry);
-      if (label) {
-        labels.push(`@${ref.type}:${label}`);
-      } else {
-        const displayNum = isSupp ? `S${num}` : `${num}`;
-        warnings.push(`Unknown reference: ${ref.type} ${displayNum} (no matching label)`);
-        labels.push(ref.match); // Keep original if no match
-      }
-    }
-
-    if (labels.length > 0 && !labels.includes(ref.match)) {
-      const replacement = labels.join('; ');
-
-      // Skip if the @-syntax already appears in the preceding text
-      // This prevents duplication when import restores @fig:x and then we see "Fig. 1"
-      // e.g., "@fig:map@fig:map{++@fig:map++}" or "@fig:mapFigure 1" patterns
-      const textBefore = result.slice(Math.max(0, ref.position - REF_CONTEXT_WINDOW), ref.position);
-      const alreadyHasRef = labels.some((label) => textBefore.includes(label));
-      if (alreadyHasRef) {
-        continue; // Skip - ref already present nearby
-      }
-
-      result =
-        result.slice(0, ref.position) + replacement + result.slice(ref.position + ref.match.length);
-
-      conversions.push({
-        from: ref.match,
-        to: replacement,
-      });
-    }
-  }
-
-  return { converted: result, conversions, warnings };
-}
-
-/**
- * Detect @-style references in text
- */
-export function detectDynamicRefs(text: string): DynamicRef[] {
-  if (typeof text !== 'string') {
-    throw new TypeError(`text must be a string, got ${typeof text}`);
-  }
-
-  const refs: DynamicRef[] = [];
-  REF_PATTERN.lastIndex = 0;
-  let match: RegExpExecArray | null;
-
-  while ((match = REF_PATTERN.exec(text)) !== null) {
-    const type = match[1];
-    const label = match[2];
-    if (!type || !label) continue;
-    refs.push({
-      type: type as 'fig' | 'tbl' | 'eq',
-      label: label,
-      match: match[0],
-      position: match.index,
-    });
-  }
-
-  return refs;
-}
-
-/**
- * Get reference status for a file/text
- */
-export function getRefStatus(text: string, registry: Registry): RefStatus {
-  const dynamic = detectDynamicRefs(text);
-  const hardcoded = detectHardcodedRefs(text) as HardcodedRef[];
-
-  // Count anchors in this text
-  ANCHOR_PATTERN.lastIndex = 0;
-  let figCount = 0,
-    tblCount = 0,
-    eqCount = 0;
-  let match: RegExpExecArray | null;
-  while ((match = ANCHOR_PATTERN.exec(text)) !== null) {
-    const type = match[1];
-    if (!type) continue;
-    if (type === 'fig') figCount++;
-    else if (type === 'tbl') tblCount++;
-    else if (type === 'eq') eqCount++;
-  }
-
-  return {
-    dynamic,
-    hardcoded,
-    anchors: { figures: figCount, tables: tblCount, equations: eqCount },
-  };
-}
-
-/**
- * Detect forward references in combined text
- * A forward reference is a @ref that appears before its {#anchor} definition
- */
-export function detectForwardRefs(text: string): {
-  forwardRefs: Array<{ type: string; label: string; match: string; position: number }>;
-  anchorPositions: Map<string, number>;
-} {
-  // Build map of anchor positions: "fig:label" -> position
-  const anchorPositions = new Map<string, number>();
-  ANCHOR_PATTERN.lastIndex = 0;
-  let match: RegExpExecArray | null;
-  while ((match = ANCHOR_PATTERN.exec(text)) !== null) {
-    const type = match[1];
-    const label = match[2];
-    if (!type || !label) continue;
-    const key = `${type}:${label}`;
-    // Only store first occurrence (in case of duplicates)
-    if (!anchorPositions.has(key)) {
-      anchorPositions.set(key, match.index);
-    }
-  }
-
-  // Find all references
-  const refs = detectDynamicRefs(text);
-
-  // Filter to only forward references
-  const forwardRefs = refs.filter((ref) => {
-    const key = `${ref.type}:${ref.label}`;
-    const anchorPos = anchorPositions.get(key);
-    // Forward ref if anchor doesn't exist or appears after the reference
-    return anchorPos === undefined || ref.position < anchorPos;
-  });
-
-  return { forwardRefs, anchorPositions };
-}
-
-/**
- * Resolve forward references to display format
- * Only resolves refs that appear before their anchor definition
- * Leaves other refs for pandoc-crossref to handle (preserves clickable links)
- */
-export function resolveForwardRefs(
-  text: string,
-  registry: Registry
-): {
-  text: string;
-  resolved: Array<{ from: string; to: string; position: number }>;
-  unresolved: Array<{ ref: string; position: number }>;
-} {
-  const { forwardRefs } = detectForwardRefs(text);
-  const resolved: Array<{ from: string; to: string; position: number }> = [];
-  const unresolved: Array<{ ref: string; position: number }> = [];
-
-  // Process in reverse order to preserve positions
-  let result = text;
-  for (let i = forwardRefs.length - 1; i >= 0; i--) {
-    const ref = forwardRefs[i];
-    if (!ref) continue;
-    const display = labelToDisplay(ref.type as 'fig' | 'tbl' | 'eq', ref.label, registry);
-
-    if (display) {
-      result =
-        result.slice(0, ref.position) + display + result.slice(ref.position + ref.match.length);
-      resolved.push({
-        from: ref.match,
-        to: display,
-        position: ref.position,
-      });
-    } else {
-      unresolved.push({
-        ref: ref.match,
-        position: ref.position,
-      });
-    }
-  }
-
-  return { text: result, resolved, unresolved };
-}
-
-/**
- * Resolve ALL supplementary references and strip supplementary anchor labels.
- *
- * pandoc-crossref cannot produce "Figure S1" numbering — it numbers all figures
- * sequentially. This function resolves every @fig:label / @tbl:label that points
- * to a supplementary item to plain text ("Figure S1", "Table S1") and removes
- * the {#fig:label} / {#tbl:label} attributes so pandoc-crossref ignores them.
- */
-export function resolveSupplementaryRefs(
-  text: string,
-  registry: Registry
-): {
-  text: string;
-  resolved: Array<{ from: string; to: string }>;
-} {
-  const resolved: Array<{ from: string; to: string }> = [];
-  let result = text;
-
-  // Collect supplementary labels
-  const suppLabels = new Set<string>();
-  for (const [label, info] of registry.figures) {
-    if (info.isSupp) suppLabels.add(`fig:${label}`);
-  }
-  for (const [label, info] of registry.tables) {
-    if (info.isSupp) suppLabels.add(`tbl:${label}`);
-  }
-
-  if (suppLabels.size === 0) return { text: result, resolved };
-
-  // 1. Replace all @fig:label / @tbl:label references to supplementary items
-  const refs = detectDynamicRefs(result);
-  // Process in reverse to preserve positions
-  for (let i = refs.length - 1; i >= 0; i--) {
-    const ref = refs[i];
-    if (!ref) continue;
-    const key = `${ref.type}:${ref.label}`;
-    if (!suppLabels.has(key)) continue;
-
-    const display = labelToDisplay(ref.type as 'fig' | 'tbl' | 'eq', ref.label, registry);
-    if (display) {
-      result =
-        result.slice(0, ref.position) + display + result.slice(ref.position + ref.match.length);
-      resolved.push({ from: ref.match, to: display });
-    }
-  }
-
-  // 2. Strip {#fig:label} and {#tbl:label} attributes from supplementary anchors
-  //    so pandoc-crossref does not re-number them
-  for (const key of suppLabels) {
-    // Match {#fig:label ...} or just {#fig:label}
-    const escaped = key.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-    const pattern = new RegExp(`\\{#${escaped}(?:\\s[^}]*)?\\}`, 'g');
-    result = result.replace(pattern, (match) => {
-      resolved.push({ from: match, to: '(stripped)' });
-      return '';
-    });
-  }
-
-  return { text: result, resolved };
-}
-
-/**
- * Format registry for display
- */
-export function formatRegistry(registry: Registry): string {
-  const lines: string[] = [];
-
-  if (registry.figures.size > 0) {
-    lines.push('Figures:');
-    for (const [label, info] of registry.figures) {
-      const num = info.isSupp ? `S${info.num}` : info.num;
-      lines.push(`  Figure ${num}: @fig:${label} (${info.file})`);
-    }
-  }
-
-  if (registry.tables.size > 0) {
-    if (lines.length > 0) lines.push('');
-    lines.push('Tables:');
-    for (const [label, info] of registry.tables) {
-      const num = info.isSupp ? `S${info.num}` : info.num;
-      lines.push(`  Table ${num}: @tbl:${label} (${info.file})`);
-    }
-  }
-
-  if (registry.equations.size > 0) {
-    if (lines.length > 0) lines.push('');
-    lines.push('Equations:');
-    for (const [label, info] of registry.equations) {
-      lines.push(`  Equation ${info.num}: @eq:${label} (${info.file})`);
-    }
-  }
-
-  if (lines.length === 0) {
-    lines.push('No figure/table anchors found.');
-  }
-
-  return lines.join('\n');
-}
+/**
+ * Cross-reference handling - dynamic figure/table references
+ *
+ * Enables:
+ * - @fig:label syntax in source (auto-numbered)
+ * - Conversion to "Figure 1" in Word output
+ * - Auto-conversion back during import
+ */
+
+import * as fs from 'fs';
+import * as path from 'path';
+import YAML from 'yaml';
+import type {
+  RefNumber,
+  HardcodedRef,
+  DynamicRef,
+  FigureInfo,
+  Registry,
+  RefStatus,
+  ConversionResult,
+} from './types.js';
+
+// =============================================================================
+// Constants
+// =============================================================================
+
+/** Characters of context to check before a reference for deduplication */
+const REF_CONTEXT_WINDOW = 100;
+
+/** Minimum word length for similarity calculations */
+const MIN_WORD_LENGTH = 2;
+
+// =============================================================================
+// Type Definitions (Internal)
+// =============================================================================
+
+/**
+ * Reference info (internal use in registry building)
+ */
+interface RefInfo {
+  label: string;
+  num: number;
+  isSupp: boolean;
+  file: string;
+}
+
+/**
+ * Parsed reference number components
+ */
+interface ParsedRefNumber {
+  isSupp: boolean;
+  num: number;
+  suffix: string | null;
+}
+
+/**
+ * Detected reference with parsed numbers
+ */
+interface DetectedRef {
+  type: 'fig' | 'tbl' | 'eq';
+  match: string;
+  numbers: ParsedRefNumber[];
+  position: number;
+}
+
+// =============================================================================
+// Internal Helpers
+// =============================================================================
+
+/**
+ * Discover section files from a directory by reading config files
+ * Only returns files explicitly defined in rev.yaml or sections.yaml
+ * Returns empty array if no config found (caller should handle this)
+ */
+function discoverSectionFiles(directory: string): string[] {
+  // Try rev.yaml first
+  const revYamlPath = path.join(directory, 'rev.yaml');
+  if (fs.existsSync(revYamlPath)) {
+    try {
+      const config = YAML.parse(fs.readFileSync(revYamlPath, 'utf-8'));
+      if (config.sections && Array.isArray(config.sections) && config.sections.length > 0) {
+        return config.sections.filter((f: string) => fs.existsSync(path.join(directory, f)));
+      }
+    } catch (e) {
+      if (process.env.DEBUG) {
+        console.warn('crossref: YAML parse error in rev.yaml:', (e as Error).message);
+      }
+    }
+  }
+
+  // Try sections.yaml
+  const sectionsPath = path.join(directory, 'sections.yaml');
+  if (fs.existsSync(sectionsPath)) {
+    try {
+      const config = YAML.parse(fs.readFileSync(sectionsPath, 'utf-8'));
+      if (config.sections) {
+        const sectionOrder = Object.entries(config.sections)
+          .sort((a, b) => ((a[1] as any).order ?? 999) - ((b[1] as any).order ?? 999))
+          .map(([file]) => file);
+        return sectionOrder.filter((f) => fs.existsSync(path.join(directory, f)));
+      }
+    } catch (e) {
+      if (process.env.DEBUG) {
+        console.warn('crossref: YAML parse error in sections.yaml:', (e as Error).message);
+      }
+    }
+  }
+
+  // No config found - return empty array
+  // Caller must handle this (either error or use explicit sections)
+  return [];
+}
+
+// =============================================================================
+// Detection Patterns
+// =============================================================================
+
+/**
+ * Patterns for detecting hardcoded references
+ * Matches complex patterns including:
+ * - Simple: "Figure 1", "Fig. 2a", "Table S1"
+ * - Ranges: "Figures 1-3", "Fig. 1a-c", "Figs. 1a-3b"
+ * - Lists: "Figures 1, 2, and 3", "Fig. 1a, b, c", "Tables 1 & 2"
+ * - Mixed: "Figs. 1, 3-5, and 7"
+ *
+ * Uses a simpler base pattern and parses the full match for lists
+ */
+const DETECTION_PATTERNS: Record<string, RegExp> = {
+  // Captures the full reference including lists with "and"
+  // Group 1: type prefix (Figure, Fig., etc.)
+  // Group 2: reference list (parsed by parseReferenceList())
+  // Matches: "1", "1a", "1-3", "1a-c", "1, 2, 3", "1 and 2", "1, 2 and 3", "1, 2, and 3"
+  // Separator: comma/dash/ampersand, optionally followed by "and"
+  // Standalone letters must be followed by separator, punctuation, or word boundary
+  // Also handles: "see Figure 1", "(Fig. 1)", "in Figures 1–3"
+  // Note: 'gi' flag makes these case-insensitive, so "figure 1" is also matched
+  figure: /\b(Figures?|Figs?\.?)\s+((?:\d+|S\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+|S\d+)[a-z]?|[a-z]\b))*)/gi,
+
+  table: /\b(Tables?|Tabs?\.?)\s+((?:\d+|S\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+|S\d+)[a-z]?|[a-z]\b))*)/gi,
+
+  equation: /\b(Equations?|Eqs?\.?)\s+((?:\d+)[a-z]?(?:(?:\s*[-–—,&]\s*(?:and\s+)?|\s+and\s+)(?:(?:\d+)[a-z]?|[a-z]\b))*)/gi,
+};
+
+/**
+ * Patterns to EXCLUDE from detection (false positives)
+ * These look like references but aren't (e.g., "Table of Contents", "Figure skating")
+ */
+const EXCLUSION_PATTERNS = [
+  /\bTable\s+of\s+Contents?\b/gi,
+  /\bFigure\s+skating\b/gi,
+  /\bFigure\s+out\b/gi,
+  /\bFigure\s+it\b/gi,
+  /\bTable\s+setting/gi,
+  /\bEquation\s+editor\b/gi,
+];
+
+/**
+ * Pattern for extracting anchors from markdown: {#fig:label}, {#tbl:label}
+ */
+const ANCHOR_PATTERN = /\{#(fig|tbl|eq):([a-zA-Z0-9_-]+)/gi;
+
+/**
+ * Pattern for @-style references: @fig:label, @tbl:label
+ */
+const REF_PATTERN = /@(fig|tbl|eq):([a-zA-Z0-9_-]+)/gi;
+
+// =============================================================================
+// Public API
+// =============================================================================
+
+/**
+ * Normalize a reference type to standard form
+ */
+export function normalizeType(typeStr: string): 'fig' | 'tbl' | 'eq' | string {
+  if (typeof typeStr !== 'string') {
+    throw new TypeError(`typeStr must be a string, got ${typeof typeStr}`);
+  }
+  const lower = typeStr.toLowerCase().replace(/\.$/, '');
+  if (lower.startsWith('fig')) return 'fig';
+  if (lower.startsWith('tab')) return 'tbl';
+  if (lower.startsWith('eq')) return 'eq';
+  return lower;
+}
+
+/**
+ * Parse a reference number, handling supplementary (S1, S2) and letter suffixes (1a, 1b)
+ */
+export function parseRefNumber(numStr: string, suffix: string | null = null): ParsedRefNumber {
+  if (!numStr || typeof numStr !== 'string') {
+    return { isSupp: false, num: 0, suffix: suffix || null };
+  }
+  const isSupp = numStr.toUpperCase().startsWith('S');
+  const numPart = isSupp ? numStr.slice(1) : numStr;
+  // Extract suffix if embedded in numStr (e.g., "1a")
+  const match = numPart.match(/^(\d+)([a-z])?$/i);
+  const num = match && match[1] ? parseInt(match[1], 10) : parseInt(numPart, 10);
+  const extractedSuffix = suffix || (match && match[2]) || null;
+  return { isSupp, num, suffix: extractedSuffix ? extractedSuffix.toLowerCase() : null };
+}
+
+/**
+ * Parse a reference list string like "1, 2, and 3" or "1a-c" or "1a-3b"
+ * Returns an array of {num, isSupp, suffix} objects
+ */
+export function parseReferenceList(listStr: string): ParsedRefNumber[] {
+  const results: ParsedRefNumber[] = [];
+  if (!listStr || typeof listStr !== 'string') return results;
+
+  // Normalize: replace "and" with comma, normalize dashes
+  let normalized = listStr
+    .replace(/\s+and\s+/gi, ', ')
+    .replace(/[–—]/g, '-') // en-dash, em-dash → hyphen
+    .replace(/&/g, ', '); // & → comma
+
+  // Split by comma (but not by dash, which indicates ranges)
+  const parts = normalized.split(/\s*,\s*/).filter((p) => p.trim());
+
+  let lastFullRef: { num: number; isSupp: boolean } | null = null; // Track the last full reference for implicit prefixes
+
+  for (const part of parts) {
+    const trimmed = part.trim();
+    if (!trimmed) continue;
+
+    // Check if this is a range (contains -)
+    if (trimmed.includes('-')) {
+      const parts = trimmed.split('-').map((s) => s.trim());
+      const start = parts[0] || '';
+      const end = parts[1] || '';
+
+      // Check if end is just a letter (e.g., "1a-c" where end is "c")
+      const endIsLetterOnly = /^[a-z]$/i.test(end);
+
+      const startRef = parseRefNumber(start);
+      // For letter-only end, don't parse as number
+      const endRef = endIsLetterOnly
+        ? { num: startRef.num, isSupp: startRef.isSupp, suffix: end.toLowerCase() }
+        : parseRefNumber(end);
+
+      // Handle different range types:
+      // 1. Suffix-only range on same number: "1a-c" → 1a, 1b, 1c
+      // 2. Number range: "1-3" → 1, 2, 3
+      // 3. Cross-number suffix range: "1a-3b" → 1a...1z, 2a...2z, 3a, 3b (limited)
+
+      if (startRef.suffix && endRef.suffix && startRef.num !== endRef.num) {
+        // Cross-number suffix range: "1a-3b"
+        // For academic papers, limit intermediate figures to same suffix range
+        // e.g., "1a-3b" typically means 1a, 1b, 2a, 2b, 3a, 3b
+        const maxSuffix = Math.max(
+          startRef.suffix.charCodeAt(0),
+          endRef.suffix.charCodeAt(0)
+        );
+
+        for (let n = startRef.num; n <= endRef.num; n++) {
+          const suffixStart =
+            n === startRef.num ? startRef.suffix.charCodeAt(0) : 'a'.charCodeAt(0);
+          const suffixEnd = n === endRef.num ? endRef.suffix.charCodeAt(0) : maxSuffix;
+
+          for (let s = suffixStart; s <= suffixEnd; s++) {
+            results.push({
+              num: n,
+              isSupp: startRef.isSupp,
+              suffix: String.fromCharCode(s),
+            });
+          }
+        }
+        lastFullRef = { num: endRef.num, isSupp: startRef.isSupp };
+      } else if (startRef.suffix || endRef.suffix) {
+        // Suffix range on same number: "1a-c"
+        const num: number = startRef.num !== 0 ? startRef.num : (lastFullRef ? lastFullRef.num : 1);
+        const isSupp: boolean = startRef.isSupp ? startRef.isSupp : (lastFullRef ? lastFullRef.isSupp : false);
+        const startCode = (startRef.suffix || 'a').charCodeAt(0);
+        const endCode = (endRef.suffix || 'a').charCodeAt(0);
+
+        for (let code = startCode; code <= endCode; code++) {
+          results.push({
+            num,
+            isSupp,
+            suffix: String.fromCharCode(code),
+          });
+        }
+        lastFullRef = { num, isSupp };
+      } else {
+        // Pure number range: "1-3"
+        for (let n = startRef.num; n <= endRef.num; n++) {
+          results.push({
+            num: n,
+            isSupp: startRef.isSupp,
+            suffix: null,
+          });
+        }
+        lastFullRef = { num: endRef.num, isSupp: startRef.isSupp };
+      }
+    } else {
+      // Single reference or implicit suffix
+      // Check if it's just a letter (implicit prefix from previous number)
+      if (/^[a-z]$/i.test(trimmed) && lastFullRef) {
+        // Implicit prefix: "b" after "1a" means "1b"
+        results.push({
+          num: lastFullRef.num,
+          isSupp: lastFullRef.isSupp,
+          suffix: trimmed.toLowerCase(),
+        });
+      } else {
+        // Full reference: "1", "1a", "S1", "S1a"
+        const ref = parseRefNumber(trimmed);
+        results.push(ref);
+        lastFullRef = { num: ref.num, isSupp: ref.isSupp };
+      }
+    }
+  }
+
+  return results;
+}
+
+/**
+ * Build a registry of figure/table labels from .md files
+ * Scans for {#fig:label} and {#tbl:label} anchors
+ *
+ * IMPORTANT: This function requires either explicit sections or a rev.yaml/sections.yaml config.
+ * It will NOT guess by scanning all .md files, as this leads to incorrect numbering
+ * when temporary files (paper_clean.md, etc.) exist in the directory.
+ */
+export function buildRegistry(directory: string, sections?: string[]): Registry {
+  if (typeof directory !== 'string') {
+    throw new TypeError(`directory must be a string, got ${typeof directory}`);
+  }
+
+  const figures = new Map<string, FigureInfo>();
+  const tables = new Map<string, FigureInfo>();
+  const equations = new Map<string, FigureInfo>();
+
+  // Counters for numbering (separate for main and supplementary)
+  let figNum = 0;
+  let figSuppNum = 0;
+  let tblNum = 0;
+  let tblSuppNum = 0;
+  let eqNum = 0;
+
+  let orderedFiles: string[];
+
+  if (Array.isArray(sections) && sections.length > 0) {
+    // Use explicitly provided section files - most reliable
+    orderedFiles = sections.filter((f) => fs.existsSync(path.join(directory, f)));
+  } else {
+    // Try to determine sections from config files (rev.yaml or sections.yaml)
+    orderedFiles = discoverSectionFiles(directory);
+    // If no config found, return empty registry rather than guessing
+    // This prevents bugs from scanning wrong files
+  }
+
+  // Determine if a file is supplementary
+  const isSupplementary = (filename: string): boolean =>
+    filename.toLowerCase().includes('supp') || filename.toLowerCase().includes('appendix');
+
+  // Process each file in order
+  for (const file of orderedFiles) {
+    const filePath = path.join(directory, file);
+    const content = fs.readFileSync(filePath, 'utf-8');
+    const isSupp = isSupplementary(file);
+
+    // Find all anchors
+    let match: RegExpExecArray | null;
+    ANCHOR_PATTERN.lastIndex = 0;
+    while ((match = ANCHOR_PATTERN.exec(content)) !== null) {
+      const typeRaw = match[1];
+      const labelRaw = match[2];
+      if (!typeRaw || !labelRaw) continue;
+
+      const type = typeRaw.toLowerCase();
+      const label = labelRaw;
+
+      if (type === 'fig') {
+        if (isSupp) {
+          figSuppNum++;
+          figures.set(label, { label, num: figSuppNum, isSupp: true, file });
+        } else {
+          figNum++;
+          figures.set(label, { label, num: figNum, isSupp: false, file });
+        }
+      } else if (type === 'tbl') {
+        if (isSupp) {
+          tblSuppNum++;
+          tables.set(label, { label, num: tblSuppNum, isSupp: true, file });
+        } else {
+          tblNum++;
+          tables.set(label, { label, num: tblNum, isSupp: false, file });
+        }
+      } else if (type === 'eq') {
+        eqNum++;
+        equations.set(label, { label, num: eqNum, isSupp: false, file });
+      }
+    }
+  }
+
+  // Build reverse lookup: number → label
+  const byNumber: Registry['byNumber'] = {
+    fig: new Map(),
+    figS: new Map(),
+    tbl: new Map(),
+    tblS: new Map(),
+    eq: new Map(),
+  };
+
+  for (const [label, info] of figures) {
+    const key = info.isSupp ? 'figS' : 'fig';
+    byNumber[key].set(info.num, label);
+  }
+  for (const [label, info] of tables) {
+    const key = info.isSupp ? 'tblS' : 'tbl';
+    byNumber[key].set(info.num, label);
+  }
+  for (const [label, info] of equations) {
+    byNumber.eq.set(info.num, label);
+  }
+
+  return { figures, tables, equations, byNumber };
+}
+
+/**
+ * Get the display string for a label (e.g., "Figure 1", "Table S2")
+ */
+export function labelToDisplay(
+  type: 'fig' | 'tbl' | 'eq',
+  label: string,
+  registry: Registry
+): string | null {
+  if (!registry || !registry.figures) return null;
+
+  const collection =
+    type === 'fig' ? registry.figures : type === 'tbl' ? registry.tables : registry.equations;
+
+  const info = collection.get(label);
+  if (!info) return null;
+
+  const prefix = type === 'fig' ? 'Figure' : type === 'tbl' ? 'Table' : 'Equation';
+  const numStr = info.isSupp ? `S${info.num}` : `${info.num}`;
+
+  return `${prefix} ${numStr}`;
+}
+
+/**
+ * Get the label for a display number (e.g., "fig:heatmap" from Figure 1)
+ */
+export function numberToLabel(
+  type: 'fig' | 'tbl' | 'eq',
+  num: number,
+  isSupp: boolean,
+  registry: Registry
+): string | null {
+  if (!registry || !registry.byNumber) return null;
+
+  const key = isSupp ? (`${type}S` as keyof Registry['byNumber']) : type;
+  return registry.byNumber[key]?.get(num) || null;
+}
+
+/**
+ * Detect all hardcoded references in text
+ */
+export function detectHardcodedRefs(text: string): DetectedRef[] {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+
+  const refs: DetectedRef[] = [];
+
+  for (const [type, pattern] of Object.entries(DETECTION_PATTERNS)) {
+    pattern.lastIndex = 0;
+    let match: RegExpExecArray | null;
+
+    while ((match = pattern.exec(text)) !== null) {
+      // Pattern groups:
+      // [1] = type prefix (Figure, Fig., etc.)
+      // [2] = reference list string (e.g., "1, 2, and 3" or "1a-3b")
+
+      const listStr = match[2];
+      if (!listStr) continue;
+      const numbers = parseReferenceList(listStr);
+
+      // Skip if no valid numbers were parsed
+      if (numbers.length === 0) continue;
+
+      refs.push({
+        type: normalizeType(type) as 'fig' | 'tbl' | 'eq',
+        match: match[0],
+        numbers,
+        position: match.index,
+      });
+    }
+  }
+
+  // Sort by position
+  refs.sort((a, b) => a.position - b.position);
+  return refs;
+}
+
+/**
+ * Convert hardcoded references to @-style references
+ */
+export function convertHardcodedRefs(text: string, registry: Registry): ConversionResult {
+  // Input validation delegated to detectHardcodedRefs
+  const refs = detectHardcodedRefs(text);
+  const conversions: Array<{ from: string; to: string }> = [];
+  const warnings: string[] = [];
+
+  // Process in reverse order to preserve positions
+  let result = text;
+  for (let i = refs.length - 1; i >= 0; i--) {
+    const ref = refs[i];
+    if (!ref) continue;
+
+    // Build replacement
+    const labels: string[] = [];
+    for (const { num, isSupp } of ref.numbers) {
+      const label = numberToLabel(ref.type, num, isSupp, registry);
+      if (label) {
+        labels.push(`@${ref.type}:${label}`);
+      } else {
+        const displayNum = isSupp ? `S${num}` : `${num}`;
+        warnings.push(`Unknown reference: ${ref.type} ${displayNum} (no matching label)`);
+        labels.push(ref.match); // Keep original if no match
+      }
+    }
+
+    if (labels.length > 0 && !labels.includes(ref.match)) {
+      const replacement = labels.join('; ');
+
+      // Skip if the @-syntax already appears in the preceding text
+      // This prevents duplication when import restores @fig:x and then we see "Fig. 1"
+      // e.g., "@fig:map@fig:map{++@fig:map++}" or "@fig:mapFigure 1" patterns
+      const textBefore = result.slice(Math.max(0, ref.position - REF_CONTEXT_WINDOW), ref.position);
+      const alreadyHasRef = labels.some((label) => textBefore.includes(label));
+      if (alreadyHasRef) {
+        continue; // Skip - ref already present nearby
+      }
+
+      result =
+        result.slice(0, ref.position) + replacement + result.slice(ref.position + ref.match.length);
+
+      conversions.push({
+        from: ref.match,
+        to: replacement,
+      });
+    }
+  }
+
+  return { converted: result, conversions, warnings };
+}
+
+/**
+ * Detect @-style references in text
+ */
+export function detectDynamicRefs(text: string): DynamicRef[] {
+  if (typeof text !== 'string') {
+    throw new TypeError(`text must be a string, got ${typeof text}`);
+  }
+
+  const refs: DynamicRef[] = [];
+  REF_PATTERN.lastIndex = 0;
+  let match: RegExpExecArray | null;
+
+  while ((match = REF_PATTERN.exec(text)) !== null) {
+    const type = match[1];
+    const label = match[2];
+    if (!type || !label) continue;
+    refs.push({
+      type: type as 'fig' | 'tbl' | 'eq',
+      label: label,
+      match: match[0],
+      position: match.index,
+    });
+  }
+
+  return refs;
+}
+
+/**
+ * Get reference status for a file/text
+ */
+export function getRefStatus(text: string, registry: Registry): RefStatus {
+  const dynamic = detectDynamicRefs(text);
+  const hardcoded = detectHardcodedRefs(text) as HardcodedRef[];
+
+  // Count anchors in this text
+  ANCHOR_PATTERN.lastIndex = 0;
+  let figCount = 0,
+    tblCount = 0,
+    eqCount = 0;
+  let match: RegExpExecArray | null;
+  while ((match = ANCHOR_PATTERN.exec(text)) !== null) {
+    const type = match[1];
+    if (!type) continue;
+    if (type === 'fig') figCount++;
+    else if (type === 'tbl') tblCount++;
+    else if (type === 'eq') eqCount++;
+  }
+
+  return {
+    dynamic,
+    hardcoded,
+    anchors: { figures: figCount, tables: tblCount, equations: eqCount },
+  };
+}
+
+/**
+ * Detect forward references in combined text
+ * A forward reference is a @ref that appears before its {#anchor} definition
+ */
+export function detectForwardRefs(text: string): {
+  forwardRefs: Array<{ type: string; label: string; match: string; position: number }>;
+  anchorPositions: Map<string, number>;
+} {
+  // Build map of anchor positions: "fig:label" -> position
+  const anchorPositions = new Map<string, number>();
+  ANCHOR_PATTERN.lastIndex = 0;
+  let match: RegExpExecArray | null;
+  while ((match = ANCHOR_PATTERN.exec(text)) !== null) {
+    const type = match[1];
+    const label = match[2];
+    if (!type || !label) continue;
+    const key = `${type}:${label}`;
+    // Only store first occurrence (in case of duplicates)
+    if (!anchorPositions.has(key)) {
+      anchorPositions.set(key, match.index);
+    }
+  }
+
+  // Find all references
+  const refs = detectDynamicRefs(text);
+
+  // Filter to only forward references
+  const forwardRefs = refs.filter((ref) => {
+    const key = `${ref.type}:${ref.label}`;
+    const anchorPos = anchorPositions.get(key);
+    // Forward ref if anchor doesn't exist or appears after the reference
+    return anchorPos === undefined || ref.position < anchorPos;
+  });
+
+  return { forwardRefs, anchorPositions };
+}
+
+/**
+ * Resolve forward references to display format
+ * Only resolves refs that appear before their anchor definition
+ * Leaves other refs for pandoc-crossref to handle (preserves clickable links)
+ */
+export function resolveForwardRefs(
+  text: string,
+  registry: Registry
+): {
+  text: string;
+  resolved: Array<{ from: string; to: string; position: number }>;
+  unresolved: Array<{ ref: string; position: number }>;
+} {
+  const { forwardRefs } = detectForwardRefs(text);
+  const resolved: Array<{ from: string; to: string; position: number }> = [];
+  const unresolved: Array<{ ref: string; position: number }> = [];
+
+  // Process in reverse order to preserve positions
+  let result = text;
+  for (let i = forwardRefs.length - 1; i >= 0; i--) {
+    const ref = forwardRefs[i];
+    if (!ref) continue;
+    const display = labelToDisplay(ref.type as 'fig' | 'tbl' | 'eq', ref.label, registry);
+
+    if (display) {
+      result =
+        result.slice(0, ref.position) + display + result.slice(ref.position + ref.match.length);
+      resolved.push({
+        from: ref.match,
+        to: display,
+        position: ref.position,
+      });
+    } else {
+      unresolved.push({
+        ref: ref.match,
+        position: ref.position,
+      });
+    }
+  }
+
+  return { text: result, resolved, unresolved };
+}
+
+/**
+ * Resolve ALL supplementary references and strip supplementary anchor labels.
+ *
+ * pandoc-crossref cannot produce "Figure S1" numbering — it numbers all figures
+ * sequentially. This function resolves every @fig:label / @tbl:label that points
+ * to a supplementary item to plain text ("Figure S1", "Table S1") and removes
+ * the {#fig:label} / {#tbl:label} attributes so pandoc-crossref ignores them.
+ */
+export function resolveSupplementaryRefs(
+  text: string,
+  registry: Registry
+): {
+  text: string;
+  resolved: Array<{ from: string; to: string }>;
+} {
+  const resolved: Array<{ from: string; to: string }> = [];
+  let result = text;
+
+  // Collect supplementary labels
+  const suppLabels = new Set<string>();
+  for (const [label, info] of registry.figures) {
+    if (info.isSupp) suppLabels.add(`fig:${label}`);
+  }
+  for (const [label, info] of registry.tables) {
+    if (info.isSupp) suppLabels.add(`tbl:${label}`);
+  }
+
+  if (suppLabels.size === 0) return { text: result, resolved };
+
+  // 1. Replace all @fig:label / @tbl:label references to supplementary items
+  const refs = detectDynamicRefs(result);
+  // Process in reverse to preserve positions
+  for (let i = refs.length - 1; i >= 0; i--) {
+    const ref = refs[i];
+    if (!ref) continue;
+    const key = `${ref.type}:${ref.label}`;
+    if (!suppLabels.has(key)) continue;
+
+    const display = labelToDisplay(ref.type as 'fig' | 'tbl' | 'eq', ref.label, registry);
+    if (display) {
+      result =
+        result.slice(0, ref.position) + display + result.slice(ref.position + ref.match.length);
+      resolved.push({ from: ref.match, to: display });
+    }
+  }
+
+  // 2. Strip {#fig:label} and {#tbl:label} attributes from supplementary anchors
+  //    so pandoc-crossref does not re-number them
+  for (const key of suppLabels) {
+    // Match {#fig:label ...} or just {#fig:label}
+    const escaped = key.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+    const pattern = new RegExp(`\\{#${escaped}(?:\\s[^}]*)?\\}`, 'g');
+    result = result.replace(pattern, (match) => {
+      resolved.push({ from: match, to: '(stripped)' });
+      return '';
+    });
+  }
+
+  return { text: result, resolved };
+}
+
+/**
+ * Format registry for display
+ */
+export function formatRegistry(registry: Registry): string {
+  const lines: string[] = [];
+
+  if (registry.figures.size > 0) {
+    lines.push('Figures:');
+    for (const [label, info] of registry.figures) {
+      const num = info.isSupp ? `S${info.num}` : info.num;
+      lines.push(`  Figure ${num}: @fig:${label} (${info.file})`);
+    }
+  }
+
+  if (registry.tables.size > 0) {
+    if (lines.length > 0) lines.push('');
+    lines.push('Tables:');
+    for (const [label, info] of registry.tables) {
+      const num = info.isSupp ? `S${info.num}` : info.num;
+      lines.push(`  Table ${num}: @tbl:${label} (${info.file})`);
+    }
+  }
+
+  if (registry.equations.size > 0) {
+    if (lines.length > 0) lines.push('');
+    lines.push('Equations:');
+    for (const [label, info] of registry.equations) {
+      lines.push(`  Equation ${info.num}: @eq:${label} (${info.file})`);
+    }
+  }
+
+  if (lines.length === 0) {
+    lines.push('No figure/table anchors found.');
+  }
+
+  return lines.join('\n');
+}