kordoc 4.16.0 → 4.16.2

package/dist/{chunk-3JWMDL3S.js → chunk-S2PL227Q.js}

@@ -2,7 +2,7 @@
 import {
   KordocError,
   OPTIONAL_DEP_INSTALL_HINT
-} from "./chunk-RC4C7Q33.js";
+} from "./chunk-P6ZDXTV6.js";
 
 // src/render/rasterize.ts
 var densityExponent = null;
@@ -88,4 +88,4 @@ export {
   rasterizeSvg,
   rasterizePageSvg
 };
-//# sourceMappingURL=chunk-3JWMDL3S.js.map
+//# sourceMappingURL=chunk-S2PL227Q.js.map

package/dist/{chunk-HIVK5V27.js → chunk-SFEQYKYJ.js}

@@ -1,86 +1,14 @@
+#!/usr/bin/env node
 import {
   HEADING_RATIO_H1,
   HEADING_RATIO_H2,
-  HEADING_RATIO_H3
-} from "./chunk-QJERJZJ6.js";
-import {
-  ensureModelsIn,
-  getModelsDir
-} from "./chunk-PQRP4NSF.js";
+  HEADING_RATIO_H3,
+  tidyScriptTags
+} from "./chunk-J3IF6KGW.js";
 import {
   safeMax,
   safeMin
-} from "./chunk-GOQBON77.js";
-
-// src/ocr/models.ts
-import { join } from "path";
-import { stat } from "fs/promises";
-var OCR_DET_MODEL = {
-  name: "PP-OCRv5 mobile det",
-  filename: "det.onnx",
-  url: "https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det_onnx/resolve/main/inference.onnx",
-  sha256: "a431985659dc921974177a95adcfbb90fd9e51989a5e04d70d0b75f597b6e61d",
-  sizeMb: 5
-};
-var OCR_REC_MODEL = {
-  name: "PP-OCRv5 korean rec",
-  filename: "rec_korean.onnx",
-  url: "https://huggingface.co/PaddlePaddle/korean_PP-OCRv5_mobile_rec_onnx/resolve/main/inference.onnx",
-  sha256: "92f0b7785e64fc9090106a241cf4c1eb97472824558272751b88a2a4476d3a08",
-  sizeMb: 13
-};
-var OCR_REC_DICT = {
-  name: "PP-OCRv5 korean dict",
-  filename: "rec_korean.yml",
-  url: "https://huggingface.co/PaddlePaddle/korean_PP-OCRv5_mobile_rec_onnx/resolve/main/inference.yml",
-  sha256: "f757fa1c40e99edcf27e9cce879b93eb2a51fa46f5ef39095689b8c37dd75998",
-  sizeMb: 1
-};
-var ALL_OCR_MODELS = [OCR_DET_MODEL, OCR_REC_MODEL, OCR_REC_DICT];
-function getOcrModelsDir() {
-  return getModelsDir("ppocr");
-}
-async function ensureOcrModels(onProgress) {
-  return ensureModelsIn(getOcrModelsDir(), ALL_OCR_MODELS, onProgress);
-}
-async function ocrModelsCached() {
-  const dir = getOcrModelsDir();
-  for (const spec of ALL_OCR_MODELS) {
-    try {
-      if (!(await stat(join(dir, spec.filename))).size) return false;
-    } catch {
-      return false;
-    }
-  }
-  return true;
-}
-function parseCharacterDict(yml) {
-  const lines = yml.split("\n");
-  const chars = [];
-  let inDict = false;
-  let dictIndent = -1;
-  for (const line of lines) {
-    if (!inDict) {
-      const m2 = /^(\s*)character_dict:\s*$/.exec(line);
-      if (m2) {
-        inDict = true;
-        dictIndent = m2[1].length;
-      }
-      continue;
-    }
-    const m = /^(\s*)- (.*)$/.exec(line);
-    if (m && m[1].length >= dictIndent) {
-      let v = m[2];
-      if (v.length >= 2 && (v.startsWith("'") && v.endsWith("'") || v.startsWith('"') && v.endsWith('"'))) {
-        v = v.slice(1, -1).replace(/''/g, "'");
-      }
-      chars.push(v);
-      continue;
-    }
-    if (line.trim() !== "") break;
-  }
-  return chars;
-}
+} from "./chunk-P6ZDXTV6.js";
 
 // src/pdf/line-extract.ts
 import { OPS } from "pdfjs-dist/legacy/build/pdf.mjs";
@@ -1085,6 +1013,12 @@ function dropInsetClipGrids(clipGrids, lineGrids) {
   const near = (a, b) => Math.abs(a - b) <= SHADE_CLIP_TOL;
   return clipGrids.filter((c) => c.clipParent || c.rowYs.length !== 2 || !lineGrids.some((l) => l.rowYs.length >= 3 && near(c.bbox.x1, l.bbox.x1) && near(c.bbox.x2, l.bbox.x2) && l.rowYs.some((top, r) => r + 1 < l.rowYs.length && top - c.bbox.y2 >= INSET_CLIP_MIN && c.bbox.y1 - l.rowYs[r + 1] >= INSET_CLIP_MIN)));
 }
+function dropHeadBandClipGrids(clipGrids, lineGrids) {
+  if (clipGrids.length === 0 || lineGrids.length === 0) return clipGrids;
+  const near = (a, b) => Math.abs(a - b) <= SHADE_CLIP_TOL;
+  const overlaps = (a, b) => Math.min(a.x2, b.x2) > Math.max(a.x1, b.x1) && Math.min(a.y2, b.y2) > Math.max(a.y1, b.y1);
+  return clipGrids.filter((c) => c.clipParent || !c.cells?.length || !lineGrids.some((l) => l.rowYs.length > c.rowYs.length && l.colXs.length === c.colXs.length && c.colXs.every((x, k) => near(x, l.colXs[k])) && c.rowYs.every((y, k) => near(y, l.rowYs[k])) && !clipGrids.some((o) => o !== c && overlaps(o.bbox, l.bbox))));
+}
 function enforceMinWidth(colXs, minWidth) {
   if (colXs.length <= 2) return colXs;
   const result = [colXs[0]];
@@ -1921,6 +1855,60 @@ function hasHorizontalLine(horizontals, y, leftX, rightX, vertexRadius) {
   return false;
 }
 
+// src/pdf/script-items.ts
+function scriptKindsOfLine(sorted) {
+  const kinds = sorted.map(() => null);
+  if (sorted.length < 2) return kinds;
+  const weight = /* @__PURE__ */ new Map();
+  for (const it of sorted) {
+    const k = Math.round(it.fontSize * 2) / 2;
+    weight.set(k, (weight.get(k) ?? 0) + it.text.trim().length);
+  }
+  let bodyFs = 0, best = -1;
+  for (const [fs, n] of weight) if (n > best || n === best && fs > bodyFs) {
+    best = n;
+    bodyFs = fs;
+  }
+  if (bodyFs <= 0) return kinds;
+  const ys = sorted.filter((it) => Math.abs(it.fontSize - bodyFs) <= bodyFs * 0.12).map((it) => it.y).sort((a, b) => a - b);
+  const bodyY = ys[ys.length >> 1];
+  const isBody = (it) => it.fontSize > bodyFs * 0.85;
+  for (let i = 1; i < sorted.length; i++) {
+    const it = sorted[i], prev = sorted[i - 1];
+    const t = it.text.trim();
+    if (!t || t.length > 16 || isBody(it)) continue;
+    if (!isBody(prev) && !kinds[i - 1]) continue;
+    const gap = it.x - (prev.x + prev.w);
+    if (gap > bodyFs * 0.35 || gap < -bodyFs * 0.5) continue;
+    const dy = it.y - bodyY;
+    if (dy >= bodyFs * 0.15) kinds[i] = "sup";
+    else if (dy <= -bodyFs * 0.08) kinds[i] = "sub";
+  }
+  return kinds;
+}
+function tagScripts(text, lines) {
+  const marks = [];
+  let cursor = 0;
+  for (const line of lines) {
+    const kinds = scriptKindsOfLine(line);
+    for (let k = 0; k < line.length; k++) {
+      const t = line[k].text;
+      if (!t) continue;
+      const at = text.indexOf(t, cursor);
+      if (at < 0 || /\S/.test(text.slice(cursor, at))) return text;
+      if (kinds[k]) marks.push([at, at + t.length, kinds[k]]);
+      cursor = at + t.length;
+    }
+  }
+  if (!marks.length) return text;
+  let out = text;
+  for (let m = marks.length - 1; m >= 0; m--) {
+    const [s, e, kind] = marks[m];
+    out = out.slice(0, s) + `<${kind}>` + out.slice(s, e) + `</${kind}>` + out.slice(e);
+  }
+  return tidyScriptTags(out);
+}
+
 // src/pdf/line-wrap.ts
 var MARKUP = /<\/?u>|~~/g;
 var bump = (m, k) => {
@@ -1970,6 +1958,13 @@ var WrapLexicon = class {
   joinedWord(left, right) {
     return this.isWord(left + right) && !this.pairs.has(left + " " + right);
   }
+  /** 두 글자+두 글자 증거만 — 한 글자 쌍으로 물러서지 않는다 (표 칸 조각처럼 줄 꺾임이 아닌 자리에 쓸 때) */
+  evidence2(left, right) {
+    const n = left.length;
+    if (n < 2 || right.length < 2) return null;
+    const k1 = pairKey(left.charCodeAt(n - 2), left.charCodeAt(n - 1)), k2 = pairKey(right.charCodeAt(0), right.charCodeAt(1));
+    return decideCounts(this.joined2.get(k1)?.get(k2) ?? 0, this.spaced2.get(k1)?.get(k2) ?? 0);
+  }
   /** 꺾인 자리 증거: "" 붙음 · " " 띄움 · null 모름 (두 글자 증거가 갈리거나 없으면 한 글자) */
   evidence(left, right) {
     const n = left.length;
@@ -2128,6 +2123,7 @@ function cellLineFills(box, contentLeft, cellRight, lineRight, fontSize) {
 }
 
 // src/pdf/cell-text.ts
+var REGION_LABELS = /* @__PURE__ */ new Set(["\uC11C\uC6B8", "\uBD80\uC0B0", "\uB300\uAD6C", "\uC778\uCC9C", "\uAD11\uC8FC", "\uB300\uC804", "\uC6B8\uC0B0", "\uC138\uC885", "\uACBD\uAE30", "\uAC15\uC6D0", "\uCDA9\uBD81", "\uCDA9\uB0A8", "\uC804\uBD81", "\uC804\uB0A8", "\uACBD\uBD81", "\uACBD\uB0A8", "\uC81C\uC8FC", "\uC804\uAD6D"]);
 var CELL_PADDING = 2;
 var SPACE_GAP_RATIO = 0.17;
 function spaceGapThreshold(fontSize) {
@@ -2222,8 +2218,8 @@ function cellTextToString(items, wrap) {
   const textLines = merged.map((line) => {
     const s = sortLineByX(line);
     if (s.length === 1) return s[0].text;
-    if (s.length === 2 && /^[가-힣]$/.test(s[0].text) && /^[가-힣]$/.test(s[1].text) && !!wrap?.lex && (wrap.lex.isWord(s[0].text + s[1].text) || wrap.lex.evidence(s[0].text, s[1].text) === "") && s[1].x - (s[0].x + s[0].w) >= Math.max(s[0].fontSize, s[1].fontSize) * 1.5 && s[1].x + s[1].w - s[0].x >= wrap.box.x2 - wrap.box.x1 - Math.max(s[0].fontSize, s[1].fontSize) * 2) return s[0].text + s[1].text;
-    const evenSpaced = detectEvenSpacedItems(s);
+    if (s.length === 2 && /^[가-힣]$/.test(s[0].text) && /^[가-힣]$/.test(s[1].text) && !!wrap?.lex && (wrap.lex.isWord(s[0].text + s[1].text) || wrap.lex.evidence(s[0].text, s[1].text) === "" || REGION_LABELS.has(s[0].text + s[1].text)) && s[1].x - (s[0].x + s[0].w) >= Math.max(s[0].fontSize, s[1].fontSize) * 1.5 && s[1].x + s[1].w - s[0].x >= wrap.box.x2 - wrap.box.x1 - Math.max(s[0].fontSize, s[1].fontSize) * 2) return s[0].text + s[1].text;
+    const evenSpaced = detectEvenSpacedItems(s, true);
     let result = s[0].text;
     for (let j = 1; j < s.length; j++) {
       if (evenSpaced[j]) {
@@ -2244,7 +2240,8 @@ function cellTextToString(items, wrap) {
     }
     return result;
   });
-  if (!wrap) return mergeCellTextLines(textLines);
+  const scripted = (s) => tagScripts(s, merged.map((l) => sortLineByX([...l])));
+  if (!wrap) return scripted(mergeCellTextLines(textLines));
   let contentLeft = Infinity;
   for (const it of items) if (it.x < contentLeft) contentLeft = it.x;
   const lineEnds = merged.map((line) => {
@@ -2257,7 +2254,7 @@ function cellTextToString(items, wrap) {
   let cellRight = -Infinity;
   for (const e of lineEnds) if (e.right > cellRight) cellRight = e.right;
   const wraps = lineEnds.slice(0, -1).map((a, i) => cellLineWraps(wrap.box, contentLeft, a.right, a.fontSize, lineEnds[i + 1].firstCharW) || lineEnds.length >= 3 && !textLines[i + 1].startsWith("(") && cellLineFills(wrap.box, contentLeft, cellRight, a.right, a.fontSize));
-  return mergeCellTextLines(textLines, { wraps, lex: wrap.lex });
+  return scripted(mergeCellTextLines(textLines, { wraps, lex: wrap.lex }));
 }
 function mergeSuperscriptRows(lines) {
   if (lines.length <= 1) return lines;
@@ -2293,9 +2290,13 @@ function mergeSuperscriptRows(lines) {
   }
   return result;
 }
-function detectEvenSpacedItems(items) {
+function detectEvenSpacedItems(items, cellLine = false) {
   const result = new Array(items.length).fill(false);
   if (items.length < 3) return result;
+  if (cellLine && items.every((it) => /^[가-힣]$/.test(it.text)) && items.slice(1).every((it) => !it.hasSpaceBefore || it.syntheticSpace) && items.slice(1).every((it, k) => it.x - (items[k].x + items[k].w) >= it.fontSize * 0.1)) {
+    markEvenRun(items, result, 0, items.length);
+    if (result.some(Boolean)) return result;
+  }
   let runStart = -1;
   for (let i = 0; i < items.length; i++) {
     const isShortKorean = /^[가-힣]{1}$/.test(items[i].text) || /^[\d]{1}$/.test(items[i].text);
@@ -2602,7 +2603,7 @@ function normalizeItems(rawItems) {
     const x = Math.round(i.transform[4]);
     const y = Math.round(i.transform[5]);
     if (!i.str.trim()) {
-      spacePositions.push({ x, y });
+      spacePositions.push({ x, y, synthetic: i.synthetic });
       continue;
     }
     const scaleX = Math.hypot(i.transform[0], i.transform[1]);
@@ -2658,7 +2659,10 @@ function normalizeItems(rawItems) {
           nearest = item;
         }
       }
-      if (nearest) nearest.hasSpaceBefore = true;
+      if (nearest) {
+        nearest.syntheticSpace = (nearest.hasSpaceBefore ? nearest.syntheticSpace === true : true) && sp.synthetic === true;
+        nearest.hasSpaceBefore = true;
+      }
     }
   }
   return deduped;
@@ -2801,6 +2805,13 @@ function mergeSuperscriptLines(lines) {
     for (const i of line) total += i.text.trim().length;
     return total > 0 && total <= 10;
   };
+  const isMarkers = (line, host) => line.length > 1 && line.length <= 16 && line.every((i) => {
+    if (i.text.trim().length > 3 || !i.text.trim()) return false;
+    return host.some((h) => {
+      const g = i.x - (h.x + h.w);
+      return g <= h.fontSize * 0.35 && g >= -h.fontSize * 0.1;
+    });
+  });
   const result = [lines[0]];
   for (let i = 1; i < lines.length; i++) {
     const prev = result[result.length - 1];
@@ -2808,8 +2819,8 @@ function mergeSuperscriptLines(lines) {
     const a = band(prev);
     const b = band(curr);
     const overlap2 = Math.min(a.top, b.top) - Math.max(a.bottom, b.bottom);
-    const prevIsFrag = isFrag(prev) && a.height <= b.height * 0.8 && overlap2 >= a.height * 0.5;
-    const currIsFrag = isFrag(curr) && b.height <= a.height * 0.8 && overlap2 >= b.height * 0.5;
+    const prevIsFrag = (isFrag(prev) || isMarkers(prev, curr)) && a.height <= b.height * 0.8 && overlap2 >= a.height * 0.5;
+    const currIsFrag = (isFrag(curr) || isMarkers(curr, prev)) && b.height <= a.height * 0.8 && overlap2 >= b.height * 0.5;
     if (prevIsFrag || currIsFrag) {
       result[result.length - 1] = [...prev, ...curr];
     } else {
@@ -3007,10 +3018,13 @@ function pushLineParagraphs(out, yLines, pageNum, lex) {
   for (let i = 0; i < lines.length; ) {
     let text = lines[i].text;
     const items = [...lines[i].items];
+    const srcLines = [lines[i].items];
     for (; i + 1 < lines.length && joins[i] !== "\n"; i++) {
       text += joins[i] + lines[i + 1].text;
       items.push(...lines[i + 1].items);
+      srcLines.push(lines[i + 1].items);
     }
+    text = tagScripts(text, srcLines.map((l) => sortLineByX([...l])));
     const block = { type: "paragraph", text, pageNumber: pageNum, bbox: computeBBox(items, pageNum), style: dominantStyle(items) };
     const faces = /* @__PURE__ */ new Map();
     for (const it of items) faces.set(it.fontName, (faces.get(it.fontName) ?? 0) + it.text.length);
@@ -3593,21 +3607,23 @@ function detectSpecialKoreanTables(blocks) {
   flushKvTable();
   return result;
 }
-function removeHeaderFooterBlocks(blocks, pageHeights, warnings) {
+function removeHeaderFooterBlocks(blocks, pageHeights, warnings, notes, tables = false) {
   const ZONE_RATIO = 0.12;
   const MIN_REPEAT = 3;
   const topEntries = [];
   const bottomEntries = [];
   for (let bi = 0; bi < blocks.length; bi++) {
     const b = blocks[bi];
-    if (!b.bbox || !b.pageNumber || !b.text?.trim()) continue;
+    const text = !tables ? b.text?.trim() : b.type === "table" && b.table ? b.table.cells.flat().map((c) => c.text.trim()).filter(Boolean).join(" | ") : void 0;
+    if (!b.bbox || !b.pageNumber || !text) continue;
     const ph = pageHeights.get(b.bbox.page) || pageHeights.get(b.pageNumber);
     if (!ph) continue;
     const blockTop = ph - (b.bbox.y + b.bbox.height);
     const blockBottom = ph - b.bbox.y;
-    const entry = { blockIdx: bi, page: b.pageNumber, text: b.text.trim() };
-    if (blockBottom <= ph * ZONE_RATIO) topEntries.push(entry);
-    else if (blockTop >= ph * (1 - ZONE_RATIO)) bottomEntries.push(entry);
+    const entry = { blockIdx: bi, page: b.pageNumber, text };
+    const boxMid = b.type === "table" && b.bbox.height <= ph * 0.05 ? (blockTop + blockBottom) / 2 : void 0;
+    if ((boxMid ?? blockBottom) <= ph * ZONE_RATIO) topEntries.push(entry);
+    else if ((boxMid ?? blockTop) >= ph * (1 - ZONE_RATIO)) bottomEntries.push(entry);
   }
   const removeSet = /* @__PURE__ */ new Set();
   for (const entries of [topEntries, bottomEntries]) {
@@ -3629,7 +3645,7 @@ function removeHeaderFooterBlocks(blocks, pageHeights, warnings) {
     for (const [p, count] of patternCount) {
       const pages = [...patternPages.get(p) ?? []];
       const span = pages.length ? Math.max(...pages) - Math.min(...pages) + 1 : 0;
-      const pageNumbered = (patternNumbers.get(p)?.size ?? 0) > 1;
+      const pageNumbered = !tables && (patternNumbers.get(p)?.size ?? 0) > 1;
       if (count >= MIN_REPEAT && pages.length >= MIN_REPEAT && (pages.length >= span * 0.4 || pageNumbered)) {
         repeatedPatterns.add(p);
       }
@@ -3637,9 +3653,11 @@ function removeHeaderFooterBlocks(blocks, pageHeights, warnings) {
     for (const e of entries) {
       const norm = e.text.replace(/\d+/g, "#");
       if (!repeatedPatterns.has(norm)) continue;
+      const noteMark = e.text.match(/^\d{1,3}\)/)?.[0];
+      if (noteMark && notes?.get(e.page)?.marks.some((m) => m.mark === noteMark)) continue;
       const cand = blocks[e.blockIdx];
       const cb = cand.bbox;
-      const tableNote = /^(?:주\s*(?:\d+\s*[).:]|[:：])|(?:자료|출처)\s*[:：])/.test(e.text) && blocks.some((o) => {
+      const tableNote = /^(?:주\s*(?:\d+\s*[).:]|[:：])|(?:자료|출처)\s*[:：])/.test(e.text.replace(/ \| /g, " ")) && blocks.some((o) => {
         if (o.type !== "table" || !o.bbox || o.bbox.page !== cb.page) return false;
         const gap = o.bbox.y - (cb.y + cb.height);
         return gap >= 0 && gap <= cb.height && cb.x >= o.bbox.x && cb.x + cb.width <= o.bbox.x + o.bbox.width;
@@ -5868,17 +5886,23 @@ function splitTwoColumnProse(items, cutX) {
     if (last && Math.abs(last[0].y - c.y) <= 3) last.push(c);
     else crossLines.push([c]);
   }
-  const bandItem = (arr) => arr.filter((i) => {
+  const scriptOf = (i, cl) => {
+    const fs = cl[0].fontSize, dy = i.y - cl[0].y;
+    if (i.fontSize > fs * 0.85 || dy > fs * 0.6 || dy < -fs * 0.35) return false;
+    return cl.some((c) => c !== i && c.fontSize > fs * 0.85 && i.x - (c.x + c.w) <= fs * 0.35 && i.x - (c.x + c.w) >= -fs * 0.1);
+  };
+  const bandItem = (arr, fits) => arr.filter((i) => {
     for (const cl of crossLines) {
-      if (Math.abs(cl[0].y - i.y) <= 3) {
+      if (fits(i, cl)) {
         cl.push(i);
         return false;
       }
     }
     return true;
   });
-  const leftRest = bandItem(left);
-  const rightRest = bandItem(right);
+  const sameY = (i, cl) => Math.abs(cl[0].y - i.y) <= 3;
+  const leftRest = bandItem(bandItem(left, sameY), scriptOf);
+  const rightRest = bandItem(bandItem(right, sameY), scriptOf);
   const boundYs = crossLines.map((cl) => cl[0].y);
   const bandOf = (y) => {
     let k = 0;
@@ -5890,7 +5914,7 @@ function splitTwoColumnProse(items, cutX) {
     const L = leftRest.filter((i) => bandOf(i.y) === k);
     const R = rightRest.filter((i) => bandOf(i.y) === k);
     groups.push(...columnPair(L, R));
-    if (k < crossLines.length) groups.push(crossLines[k]);
+    if (k < crossLines.length) groups.push(crossLines[k].sort((a, b) => b.y - a.y || a.x - b.x));
   }
   return groups;
 }
@@ -5965,7 +5989,8 @@ function topTableBand(items) {
       h: i.h,
       fontSize: i.fontSize,
       fontName: i.fontName,
-      hasSpaceBefore: i.hasSpaceBefore
+      hasSpaceBefore: i.hasSpaceBefore,
+      syntheticSpace: i.syntheticSpace
     })), 1);
     if (!candidate.some((t) => t.table.cols >= 3 && t.table.rows >= 3 && t.usedItems.size >= top.length * 0.75)) continue;
     return { top, rest: lines.slice(n).flat() };
@@ -6211,7 +6236,7 @@ function extractPageBlocksWithLines(items, pageNum, opList, pageWidth, pageHeigh
   if (underlines.size) horizontals = horizontals.filter((l) => !underlines.has(l));
   wrapUnderlineRuns(items);
   const lineGrids = detectTables ? buildTableGrids(horizontals, verticals) : [];
-  const tableClipGrids = dropInsetClipGrids(dropShadingClipGrids(clipGrids, lineGrids, extracted.fillRects, verticals), lineGrids);
+  const tableClipGrids = dropHeadBandClipGrids(dropInsetClipGrids(dropShadingClipGrids(clipGrids, lineGrids, extracted.fillRects, verticals), lineGrids), lineGrids);
   const grids = [...tableClipGrids, ...dropGridsInside(lineGrids, tableClipGrids, clipResult.containers)];
   if (grids.length === 1 && grids[0].rowYs.length === 2 && grids[0].colXs.length === 2 && grids[0].bbox.x2 - grids[0].bbox.x1 > pageWidth * 1.2 && grids[0].bbox.y2 - grids[0].bbox.y1 > pageHeight * 1.2) {
     return extractPageBlocksFallback(items, pageNum, true, detectTables, lex);
@@ -6415,6 +6440,7 @@ function extractBlocksWithGrids(items, pageNum, pageWidth, pageHeight, grids, ho
       fontSize: i.fontSize,
       fontName: i.fontName,
       hasSpaceBefore: i.hasSpaceBefore,
+      syntheticSpace: i.syntheticSpace,
       seq: i.seq
     }));
     const cellTextMap = mapTextToCells(textItems, cells);
@@ -6527,6 +6553,7 @@ function extractBlocksWithGrids(items, pageNum, pageWidth, pageHeight, grids, ho
           fontSize: i.fontSize,
           fontName: i.fontName,
           hasSpaceBefore: i.hasSpaceBefore,
+          syntheticSpace: i.syntheticSpace,
           seq: i.seq
         })))), colSpan: 1, rowSpan: 1 })));
         finalRows++;
@@ -6623,7 +6650,8 @@ function extractBlocksWithGrids(items, pageNum, pageWidth, pageHeight, grids, ho
       h: i.h,
       fontSize: i.fontSize,
       fontName: i.fontName,
-      hasSpaceBefore: i.hasSpaceBefore
+      hasSpaceBefore: i.hasSpaceBefore,
+      syntheticSpace: i.syntheticSpace
     }));
     const proseColumns = findTwoColumnProseCutX(clusterItems) !== null || detectPersistentColumnGutter(remaining.map((i) => ({ x: i.x, y: i.y, w: i.w, h: i.h > 0 ? i.h : i.fontSize }))) !== null;
     const clusterResults = (proseColumns ? [] : detectClusterTables(clusterItems, pageNum)).filter((cr) => {
@@ -6937,7 +6965,8 @@ function extractPageBlocksFallback(items, pageNum, fullPage = false, detectTable
     h: i.h,
     fontSize: i.fontSize,
     fontName: i.fontName,
-    hasSpaceBefore: i.hasSpaceBefore
+    hasSpaceBefore: i.hasSpaceBefore,
+    syntheticSpace: i.syntheticSpace
   }));
   const textRects = items.map((i) => ({ x: i.x, y: i.y, w: i.w, h: i.h > 0 ? i.h : i.fontSize }));
   const earlyProseCut = fullPage && detectTables ? findTwoColumnProseCutX(clusterItems) ?? persistentGutter(textRects) ?? (figures.length > 0 ? detectColumnGutter([...textRects, ...figures]) : null) : null;
@@ -7016,19 +7045,13 @@ function extractPageBlocksFallback(items, pageNum, fullPage = false, detectTable
 }
 
 export {
-  OCR_DET_MODEL,
-  OCR_REC_MODEL,
-  OCR_REC_DICT,
-  getOcrModelsDir,
-  ensureOcrModels,
-  ocrModelsCached,
-  parseCharacterDict,
   extractLines,
   extractImageRegions,
   filterHiddenText,
   collapseEvenSpacing,
   normalizeItems,
   WrapLexicon,
+  startsNewItem,
   wrapJoiner,
   joinPageBreakWraps,
   latinSoftWrap,
@@ -7056,4 +7079,4 @@ export {
   trimTrailingEmptyTableCols,
   extractPageBlocksWithLines
 };
-//# sourceMappingURL=chunk-HIVK5V27.js.map
+//# sourceMappingURL=chunk-SFEQYKYJ.js.map