@clazic/kordoc 2.4.17 → 2.4.19

package/dist/index.d.cts

@@ -106,17 +106,16 @@ interface ParseOptions {
     ocr?: OcrProvider;
     /**
      * OCR 모드 (CLI 자동 탐색용).
-     * - "auto": 설치된 CLI 자동 탐색 (gemini→claude→codex→ollama→tesseract)
-     * - "gemini"|"claude"|"codex"|"ollama"|"tesseract": 특정 도구 강제 지정
+     * - "auto": 설치된 CLI 자동 탐색 (codex→gemini→claude→ollama)
+     * - "gemini"|"claude"|"codex"|"ollama": 특정 도구 강제 지정
      * - "off": OCR 비활성화 (이미지 기반 PDF면 에러)
      * - undefined: 라이브러리 API 기존 동작 유지 (자동 탐색 안 함)
      */
     ocrMode?: OcrMode;
     /**
      * OCR 병렬 처리 수.
-     * - tesseract: 기본값은 CPU 코어 수 (병렬 처리로 속도 향상)
-     * - CLI 제공 프로바이더(gemini/claude/codex): 기본 1 (rate limit 보호)
-     * - 1: 순차 처리 (기존 동작)
+     * - CLI 제공 프로바이더(gemini/claude/codex): 기본 4 (배치 병렬 실행)
+     * - 1: 순차 처리
      */
     ocrConcurrency?: number;
     /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */
@@ -241,7 +240,7 @@ interface StructuredOcrResult {
     markdown: string;
 }
 /** OCR 모드 — CLI --ocr 옵션 허용값 */
-type OcrMode = "auto" | "gemini" | "claude" | "codex" | "ollama" | "tesseract" | "off";
+type OcrMode = "auto" | "gemini" | "claude" | "codex" | "ollama" | "off";
 /** 사용자 제공 OCR 함수 — 페이지 이미지를 받아 텍스트 또는 구조화된 결과 반환 */
 type OcrProvider = (pageImage: Uint8Array, pageNumber: number, mimeType: "image/png") => Promise<string | StructuredOcrResult>;
 interface WatchOptions {

package/dist/index.d.ts

@@ -106,17 +106,16 @@ interface ParseOptions {
     ocr?: OcrProvider;
     /**
      * OCR 모드 (CLI 자동 탐색용).
-     * - "auto": 설치된 CLI 자동 탐색 (gemini→claude→codex→ollama→tesseract)
-     * - "gemini"|"claude"|"codex"|"ollama"|"tesseract": 특정 도구 강제 지정
+     * - "auto": 설치된 CLI 자동 탐색 (codex→gemini→claude→ollama)
+     * - "gemini"|"claude"|"codex"|"ollama": 특정 도구 강제 지정
      * - "off": OCR 비활성화 (이미지 기반 PDF면 에러)
      * - undefined: 라이브러리 API 기존 동작 유지 (자동 탐색 안 함)
      */
     ocrMode?: OcrMode;
     /**
      * OCR 병렬 처리 수.
-     * - tesseract: 기본값은 CPU 코어 수 (병렬 처리로 속도 향상)
-     * - CLI 제공 프로바이더(gemini/claude/codex): 기본 1 (rate limit 보호)
-     * - 1: 순차 처리 (기존 동작)
+     * - CLI 제공 프로바이더(gemini/claude/codex): 기본 4 (배치 병렬 실행)
+     * - 1: 순차 처리
      */
     ocrConcurrency?: number;
     /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */
@@ -241,7 +240,7 @@ interface StructuredOcrResult {
     markdown: string;
 }
 /** OCR 모드 — CLI --ocr 옵션 허용값 */
-type OcrMode = "auto" | "gemini" | "claude" | "codex" | "ollama" | "tesseract" | "off";
+type OcrMode = "auto" | "gemini" | "claude" | "codex" | "ollama" | "off";
 /** 사용자 제공 OCR 함수 — 페이지 이미지를 받아 텍스트 또는 구조화된 결과 반환 */
 type OcrProvider = (pageImage: Uint8Array, pageNumber: number, mimeType: "image/png") => Promise<string | StructuredOcrResult>;
 interface WatchOptions {

package/dist/index.js

@@ -2183,7 +2183,7 @@ var auto_detect_exports = {};
 __export(auto_detect_exports, {
   detectAvailableOcr: () => detectAvailableOcr,
   getAutoFallbackChain: () => getAutoFallbackChain,
-  getTesseractFallbackMessage: () => getTesseractFallbackMessage,
+  getNoCliMessage: () => getNoCliMessage,
   validateOcrMode: () => validateOcrMode
 });
 import { execSync } from "child_process";
@@ -2191,7 +2191,7 @@ function detectAvailableOcr() {
   for (const cli of CLI_PRIORITY) {
     if (isCliInstalled(cli)) return cli;
   }
-  return "tesseract";
+  return null;
 }
 function isCliInstalled(name) {
   try {
@@ -2207,11 +2207,10 @@ function getAutoFallbackChain() {
   for (const cli of CLI_PRIORITY) {
     if (isCliInstalled(cli)) chain.push(cli);
   }
-  chain.push("tesseract");
   return chain;
 }
 function validateOcrMode(mode) {
-  if (mode === "auto" || mode === "off" || mode === "tesseract") return;
+  if (mode === "auto" || mode === "off") return;
   if (!isCliInstalled(mode)) {
     throw new Error(`'${mode}' CLI\uAC00 \uC124\uCE58\uB418\uC9C0 \uC54A\uC558\uC2B5\uB2C8\uB2E4.
 ${getInstallGuide(mode)}`);
@@ -2226,10 +2225,10 @@ function getInstallGuide(mode) {
   };
   return guides[mode] || `'${mode}'\uC744(\uB97C) \uC124\uCE58\uD574\uC8FC\uC138\uC694.`;
 }
-function getTesseractFallbackMessage() {
+function getNoCliMessage() {
   return [
-    "\uC124\uCE58\uB41C AI CLI\uAC00 \uC5C6\uC5B4 \uB0B4\uC7A5 tesseract.js\uB85C OCR\uC744 \uC218\uD589\uD569\uB2C8\uB2E4.",
-    "\uB354 \uB098\uC740 \uD488\uC9C8(\uD14C\uC774\uBE14/\uD5E4\uB529 \uAD6C\uC870 \uBCF4\uC874)\uC744 \uC704\uD574 AI CLI \uC124\uCE58\uB97C \uAD8C\uC7A5\uD569\uB2C8\uB2E4:",
+    "\uC124\uCE58\uB41C AI CLI\uAC00 \uC5C6\uC5B4 OCR\uC744 \uC218\uD589\uD560 \uC218 \uC5C6\uC2B5\uB2C8\uB2E4.",
+    "\uC774\uBBF8\uC9C0 \uAE30\uBC18 PDF \uCC98\uB9AC\uB97C \uC704\uD574 AI CLI \uC124\uCE58\uB97C \uAD8C\uC7A5\uD569\uB2C8\uB2E4:",
     "",
     "  [\uAD8C\uC7A5] Codex CLI:   npm install -g @openai/codex",
     "         Gemini CLI:  https://ai.google.dev/gemini-api/docs/cli",
@@ -2412,65 +2411,6 @@ var init_cli_provider = __esm({
   }
 });
 
-// src/ocr/tesseract-provider.ts
-var tesseract_provider_exports = {};
-__export(tesseract_provider_exports, {
-  createTesseractPoolProvider: () => createTesseractPoolProvider,
-  createTesseractProvider: () => createTesseractProvider
-});
-import { createWorker } from "tesseract.js";
-async function createTesseractProvider() {
-  const worker = await createWorker("kor+eng");
-  let terminated = false;
-  const provider = async (pageImage, _pageNumber, _mimeType) => {
-    const { data } = await worker.recognize(pageImage);
-    return data.text;
-  };
-  provider.terminate = async () => {
-    if (!terminated) {
-      await worker.terminate();
-      terminated = true;
-    }
-  };
-  return provider;
-}
-async function createTesseractPoolProvider(concurrency) {
-  const workers = await Promise.all(
-    Array.from({ length: concurrency }, () => createWorker("kor+eng"))
-  );
-  const idle = [...workers];
-  const waitQueue = [];
-  function acquire() {
-    if (idle.length > 0) return Promise.resolve(idle.pop());
-    return new Promise((resolve4) => waitQueue.push(resolve4));
-  }
-  function release(w) {
-    if (waitQueue.length > 0) {
-      waitQueue.shift()(w);
-    } else {
-      idle.push(w);
-    }
-  }
-  const provider = async (pageImage, _pageNumber, _mimeType) => {
-    const w = await acquire();
-    try {
-      const { data } = await w.recognize(pageImage);
-      return data.text;
-    } finally {
-      release(w);
-    }
-  };
-  provider.terminate = async () => {
-    await Promise.all(workers.map((w) => w.terminate()));
-  };
-  return provider;
-}
-var init_tesseract_provider = __esm({
-  "src/ocr/tesseract-provider.ts"() {
-    "use strict";
-  }
-});
-
 // src/ocr/batch-provider.ts
 var batch_provider_exports = {};
 __export(batch_provider_exports, {
@@ -2679,15 +2619,6 @@ async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
   }
   if (mode !== "auto") {
     validateOcrMode(mode);
-    if (mode === "tesseract") {
-      const { createTesseractProvider: createTesseractProvider2, createTesseractPoolProvider: createTesseractPoolProvider2 } = await Promise.resolve().then(() => (init_tesseract_provider(), tesseract_provider_exports));
-      if (concurrency && concurrency > 1) {
-        logger.log({ level: "info", event: "done", message: "Tesseract pool provider \uC120\uD0DD", meta: { concurrency } });
-        return createTesseractPoolProvider2(concurrency);
-      }
-      logger.log({ level: "info", event: "done", message: "Tesseract single provider \uC120\uD0DD" });
-      return createTesseractProvider2();
-    }
     if (mode === "gemini" || mode === "claude" || mode === "codex") {
       const { createBatchCliProvider: createBatchCliProvider2, DEFAULT_BATCH_SIZES: DEFAULT_BATCH_SIZES2 } = await Promise.resolve().then(() => (init_batch_provider(), batch_provider_exports));
       const effectiveBatch = batchSize ?? DEFAULT_BATCH_SIZES2[mode];
@@ -2703,27 +2634,16 @@ async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
   }
   const detected = detectAvailableOcr();
   logger.log({ level: "info", event: "progress", message: "OCR auto \uAC10\uC9C0 \uACB0\uACFC", meta: { detected } });
-  if (detected !== "codex") {
-    if (detected === "tesseract") {
-      warnings?.push({
-        message: getTesseractFallbackMessage(),
-        code: "OCR_CLI_FALLBACK"
-      });
-    } else {
-      warnings?.push({
-        message: `OCR: '${detected}' \uC0AC\uC6A9 \uC911 (codex CLI\uAC00 \uC5C6\uC5B4 fallback). \uB354 \uB098\uC740 \uD488\uC9C8\uC744 \uC704\uD574 codex CLI \uC124\uCE58\uB97C \uAD8C\uC7A5\uD569\uB2C8\uB2E4.`,
-        code: "OCR_CLI_FALLBACK"
-      });
-    }
+  if (!detected) {
+    throw new Error(
+      "\uC0AC\uC6A9 \uAC00\uB2A5\uD55C OCR CLI\uAC00 \uC5C6\uC2B5\uB2C8\uB2E4. \uB2E4\uC74C \uC911 \uD558\uB098\uB97C \uC124\uCE58\uD558\uC138\uC694:\n  Codex CLI:  npm install -g @openai/codex\n  Claude CLI: npm install -g @anthropic-ai/claude-code\n  Gemini CLI: https://ai.google.dev/gemini-api/docs/cli"
+    );
   }
-  if (detected === "tesseract") {
-    const { createTesseractProvider: createTesseractProvider2, createTesseractPoolProvider: createTesseractPoolProvider2 } = await Promise.resolve().then(() => (init_tesseract_provider(), tesseract_provider_exports));
-    if (concurrency && concurrency > 1) {
-      logger.log({ level: "info", event: "done", message: "AUTO: Tesseract pool provider \uC120\uD0DD", meta: { concurrency } });
-      return createTesseractPoolProvider2(concurrency);
-    }
-    logger.log({ level: "info", event: "done", message: "AUTO: Tesseract single provider \uC120\uD0DD" });
-    return createTesseractProvider2();
+  if (detected !== "codex") {
+    warnings?.push({
+      message: `OCR: '${detected}' \uC0AC\uC6A9 \uC911 (codex CLI\uAC00 \uC5C6\uC5B4 fallback). \uB354 \uB098\uC740 \uD488\uC9C8\uC744 \uC704\uD574 codex CLI \uC124\uCE58\uB97C \uAD8C\uC7A5\uD569\uB2C8\uB2E4.`,
+      code: "OCR_CLI_FALLBACK"
+    });
   }
   if (detected === "gemini" || detected === "codex" || detected === "claude") {
     const { createBatchCliProvider: createBatchCliProvider2, DEFAULT_BATCH_SIZES: DEFAULT_BATCH_SIZES2 } = await Promise.resolve().then(() => (init_batch_provider(), batch_provider_exports));
@@ -3115,7 +3035,7 @@ import JSZip2 from "jszip";
 import { DOMParser } from "@xmldom/xmldom";
 
 // src/utils.ts
-var VERSION = true ? "2.4.17" : "0.0.0-dev";
+var VERSION = true ? "2.4.19" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -8711,7 +8631,6 @@ async function parseDocxDocument(buffer, options, existingZip) {
 
 // src/index.ts
 init_cli_provider();
-init_tesseract_provider();
 init_markdown_to_blocks();
 init_logger();
 
@@ -11346,17 +11265,39 @@ var DEFAULT_STAGE_WEIGHTS = {
   render: 20,
   probe: 5,
   ocr: 45,
-  proofread: 10,
+  proofread: 0,
   merge: 5
 };
-var OCR_PROMPT2 = "Extract all text and tables from this image exactly as-is into Markdown. Do not summarize, infer, or alter the content in any way.";
-var PROOFREAD_PROMPT = [
-  "Perform non-destructive proofreading only on the Markdown below.",
-  "Rules:",
-  "- Do not add, remove, or infer any facts",
-  "- Do not change numbers, units, or proper nouns",
-  "- Correct only typos, spacing, line breaks, and Markdown structure",
-  "- Output the corrected Markdown body only"
+var OCR_PROMPT2 = [
+  "\uC774 PDF \uD398\uC774\uC9C0 \uC774\uBBF8\uC9C0\uC5D0\uC11C \uD14D\uC2A4\uD2B8\uC640 \uD45C\uB97C \uCD94\uCD9C\uD558\uC5EC Markdown\uC73C\uB85C \uBCC0\uD658\uD558\uACE0, OCR \uC624\uC778\uC2DD \uC624\uB958\uB97C \uC989\uC2DC \uAD50\uC815\uD558\uC5EC \uCD5C\uC885 \uACB0\uACFC\uBB3C\uC744 \uCD9C\uB825\uD558\uC138\uC694.",
+  "",
+  "\uCD94\uCD9C \uADDC\uCE59:",
+  "- \uD14D\uC2A4\uD2B8, \uD45C, \uC81C\uBAA9, \uB9AC\uC2A4\uD2B8\uB97C \uC6D0\uBB38 \uAD6C\uC870 \uADF8\uB300\uB85C Markdown\uC73C\uB85C \uBCC0\uD658",
+  "- \uD45C\uB294 Markdown \uD14C\uC774\uBE14 \uBB38\uBC95 \uC0AC\uC6A9 (| \uAD6C\uBD84, |---|---| \uD5E4\uB354 \uAD6C\uBD84\uC120 \uD3EC\uD568)",
+  "- \uD5E4\uB529\uC740 \uC2DC\uAC01\uC801 \uD06C\uAE30\uC5D0 \uB530\uB77C # ~ ###### \uC0AC\uC6A9",
+  "- \uB9AC\uC2A4\uD2B8\uB294 - \uB610\uB294 1. \uC0AC\uC6A9",
+  "- \uC774\uBBF8\uC9C0, \uB3C4\uD615 \uB4F1 \uBE44\uD14D\uC2A4\uD2B8 \uC694\uC18C\uB294 \uBB34\uC2DC",
+  "- \uC6D0\uBB38\uC758 \uC77D\uAE30 \uC21C\uC11C\uC640 \uAD6C\uC870\uB97C \uC720\uC9C0",
+  "",
+  "\uC808\uB300 \uAE08\uC9C0 \uC0AC\uD56D:",
+  "- \uBB38\uC7A5\xB7\uB2E8\uB77D\xB7\uD56D\uBAA9\uC744 \uCD94\uAC00\uD558\uAC70\uB098 \uC0AD\uC81C\uD558\uC9C0 \uB9D0 \uAC83",
+  "- \uC22B\uC790, \uD37C\uC13C\uD2B8, \uB0A0\uC9DC, \uB2E8\uC704, \uAE08\uC561\uC744 \uC808\uB300 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83",
+  "- \uACE0\uC720\uBA85\uC0AC, \uAE30\uAD00\uBA85, \uBC95\uB839\uBA85, \uC9C0\uBA85\uC744 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83",
+  "- \uD45C\uC758 \uC81C\uBAA9\uC744 \uBCC0\uACBD \uB610\uB294 \uC0AD\uC81C\uD558\uC9C0 \uB9D0 \uAC83",
+  "- \uD45C\uC758 \uD589\xB7\uC5F4 \uC218, \uC140 \uB0B4\uC6A9, \uD5E4\uB354\uB97C \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83",
+  "- \uC81C\uBAA9 \uC218\uC900(#, ##, ### \uB4F1)\uC744 \uC784\uC758\uB85C \uBC14\uAFB8\uC9C0 \uB9D0 \uAC83",
+  "- \uC6D0\uBB38\uC5D0 \uC5C6\uB294 \uB0B4\uC6A9\uC744 \uC694\uC57D\xB7\uBCF4\uC644\xB7\uCD94\uB860\uD558\uC9C0 \uB9D0 \uAC83",
+  "- ` ``` `\uB85C \uAC10\uC2F8\uAC70\uB098 \uC124\uBA85 \uD14D\uC2A4\uD2B8\uB97C \uCD94\uAC00\uD558\uC9C0 \uB9D0 \uAC83",
+  "",
+  "\uD5C8\uC6A9\uB418\uB294 \uAD50\uC815 \uBC94\uC704 (OCR \uC624\uC778\uC2DD \uC218\uC815):",
+  "- \uBA85\uBC31\uD55C \uAE00\uC790 \uC624\uC778\uC2DD \uC218\uC815 (\uC608: '0' \u2192 'O', 'l' \u2192 '1' \uB4F1 \uB9E5\uB77D\uC0C1 \uBA85\uD655\uD55C \uACBD\uC6B0\uB9CC)",
+  "- \uB2E8\uC5B4 \uC911\uAC04\uC5D0 \uC798\uBABB \uC0BD\uC785\uB41C \uACF5\uBC31 \uC81C\uAC70",
+  "- \uC904\uBC14\uAFC8 \uC624\uB958\uB85C \uBD84\uB9AC\uB41C \uBB38\uC7A5 \uBCD1\uD569 (\uC758\uBBF8 \uB2E8\uC704 \uAE30\uC900)",
+  "- Markdown \uBB38\uBC95 \uC624\uB958 \uC218\uC815 (\uD45C \uAD6C\uBD84\uC120 \uB204\uB77D, \uB9AC\uC2A4\uD2B8 \uB4E4\uC5EC\uC4F0\uAE30 \uB4F1)",
+  "",
+  "\uCD9C\uB825 \uADDC\uCE59:",
+  "- \uBCC0\uD658\uB41C Markdown \uBCF8\uBB38\uB9CC \uCD9C\uB825\uD560 \uAC83 (\uC124\uBA85, \uC8FC\uC11D, \uBA54\uD0C0 \uD14D\uC2A4\uD2B8 \uC5C6\uC774)",
+  "- \uD655\uC2E4\uD558\uC9C0 \uC54A\uC73C\uBA74 \uC6D0\uBB38\uC744 \uADF8\uB300\uB85C \uC720\uC9C0\uD560 \uAC83"
 ].join("\n");
 function elapsedMs(startAt) {
   return Math.round(performance.now() - startAt);
@@ -11367,7 +11308,6 @@ async function runUnifiedOcrPipeline(inputPath, options = {}) {
   const workspaceDir = resolve3(options.workspaceDir ?? join4(dirname3(absInput), `${stem}_ocr_workspace`));
   const imagesDir = join4(workspaceDir, "images");
   const rawDir = join4(workspaceDir, "ocr", "raw");
-  const proofDir = join4(workspaceDir, "ocr", "proofread");
   const diffDir = join4(workspaceDir, "ocr", "diff");
   const outputPath = resolve3(options.outputPath ?? join4(dirname3(absInput), `${stem}.md`));
   const reportPath = join4(workspaceDir, "run-report.json");
@@ -11387,7 +11327,6 @@ async function runUnifiedOcrPipeline(inputPath, options = {}) {
   const logger = (options.logger ?? createLoggerFromEnv()).withRun(runId).child({ component: "pipeline/unified-ocr.ts" });
   await mkdir(imagesDir, { recursive: true });
   await mkdir(rawDir, { recursive: true });
-  await mkdir(proofDir, { recursive: true });
   await mkdir(diffDir, { recursive: true });
   const timingsMs = {};
   const markStageStart = (stage, message) => emitProgress(options.onEvent, stage, 0, stageWeights, { message, type: "stage_start" });
@@ -11502,50 +11441,11 @@ async function runUnifiedOcrPipeline(inputPath, options = {}) {
     timingsMs.ocr = elapsedMs(ocrStart);
     markStageDone("ocr", "OCR \uC644\uB8CC");
     logStage("info", "ocr", "done", "\uD398\uC774\uC9C0 OCR \uC644\uB8CC", { elapsedMs: timingsMs.ocr });
-    const proofStart = performance.now();
-    currentStage = "proofread";
-    markStageStart("proofread", "\uBE44\uD30C\uAD34 \uAD50\uC815 \uC9C4\uD589 \uC911");
-    logStage("info", "proofread", "start", "\uBE44\uD30C\uAD34 \uAD50\uC815 \uC2DC\uC791", { pages: rawPagePaths.length });
-    const proofedPaths = [];
-    for (let i = 0; i < rawPagePaths.length; i++) {
-      const rawMd = await readFile(rawPagePaths[i], "utf-8");
-      const prompt = `${PROOFREAD_PROMPT}
-
----
-${rawMd}
----`;
-      const corrected = await ocrImageViaNim({
-        textOnlyPrompt: prompt,
-        model: selectedModel,
-        maxTokens: modelMaxTokens[selectedModel] ?? 8192,
-        baseUrl,
-        keyPool,
-        timeoutMs,
-        maxRetries: maxRetriesPerPage,
-        logger,
-        stage: "proofread"
-      });
-      const safeCorrected = preserveNumericIntegrity(rawMd, corrected);
-      const taggedCorrected = addUncertainTag(rawMd, safeCorrected);
-      const pagePath = join4(proofDir, `page_${String(i + 1).padStart(4, "0")}.md`);
-      await writeFile(pagePath, taggedCorrected, "utf-8");
-      await writeFile(
-        join4(diffDir, `page_${String(i + 1).padStart(4, "0")}.json`),
-        JSON.stringify(buildDiffSummary(rawMd, taggedCorrected), null, 2),
-        "utf-8"
-      );
-      proofedPaths.push(pagePath);
-      markStageProgress("proofread", Math.round((i + 1) / rawPagePaths.length * 100), i + 1, rawPagePaths.length, `\uAD50\uC815 ${i + 1}/${rawPagePaths.length}`);
-      logStage("debug", "proofread", "progress", "\uD398\uC774\uC9C0 \uAD50\uC815 \uC644\uB8CC", { page: i + 1, total: rawPagePaths.length });
-    }
-    timingsMs.proofread = elapsedMs(proofStart);
-    markStageDone("proofread", "\uAD50\uC815 \uC644\uB8CC");
-    logStage("info", "proofread", "done", "\uBE44\uD30C\uAD34 \uAD50\uC815 \uC644\uB8CC", { elapsedMs: timingsMs.proofread });
     const mergeStart = performance.now();
     currentStage = "merge";
     markStageStart("merge", "\uCD5C\uC885 Markdown \uBCD1\uD569 \uC911");
-    logStage("info", "merge", "start", "\uCD5C\uC885 \uBCD1\uD569 \uC2DC\uC791", { pages: proofedPaths.length });
-    const merged = await mergeMarkdownPages(proofedPaths);
+    logStage("info", "merge", "start", "\uCD5C\uC885 \uBCD1\uD569 \uC2DC\uC791", { pages: rawPagePaths.length });
+    const merged = await mergeMarkdownPages(rawPagePaths);
     await writeFile(outputPath, merged, "utf-8");
     timingsMs.merge = elapsedMs(mergeStart);
     markStageDone("merge", "\uBCD1\uD569 \uC644\uB8CC");
@@ -12004,40 +11904,6 @@ function ensureSupportedInput(path) {
     throw new UnifiedOcrError("UNSUPPORTED_INPUT", "convert", `\uC9C0\uC6D0\uD558\uC9C0 \uC54A\uB294 \uC785\uB825 \uD3EC\uB9F7: ${ext}`);
   }
 }
-function extractNumericTokens(text) {
-  return text.match(/\d[\d,./-]*/g) ?? [];
-}
-function preserveNumericIntegrity(rawText, correctedText) {
-  const rawTokens = extractNumericTokens(rawText);
-  const correctedTokens = extractNumericTokens(correctedText);
-  if (rawTokens.length !== correctedTokens.length) return rawText;
-  for (let i = 0; i < rawTokens.length; i++) {
-    if (rawTokens[i] !== correctedTokens[i]) return rawText;
-  }
-  return correctedText;
-}
-function addUncertainTag(rawText, correctedText) {
-  if (correctedText.includes("[\uD655\uC778\uD544\uC694:")) return correctedText;
-  const rawLen = rawText.trim().length;
-  const corrLen = correctedText.trim().length;
-  if (rawLen === 0 || corrLen === 0) return correctedText;
-  const rawLines = rawText.split("\n").filter(Boolean).length;
-  const corrLines = correctedText.split("\n").filter(Boolean).length;
-  const rawTableLines = rawText.split("\n").filter((l) => l.includes("|")).length;
-  const corrTableLines = correctedText.split("\n").filter((l) => l.includes("|")).length;
-  const suspicious = corrLen < rawLen * 0.75 || corrLines < Math.max(1, Math.floor(rawLines * 0.8)) || rawTableLines >= 2 && corrTableLines < Math.floor(rawTableLines * 0.7);
-  if (!suspicious) return correctedText;
-  return `${correctedText}
-
-[\uD655\uC778\uD544\uC694: \uAD50\uC815 \uACB0\uACFC\uAC00 \uCD95\uC57D\uB418\uC5C8\uC744 \uC218 \uC788\uC5B4 \uC6D0\uBB38\uACFC \uB300\uC870\uAC00 \uD544\uC694\uD569\uB2C8\uB2E4.]`;
-}
-function buildDiffSummary(before, after) {
-  return {
-    changed: before !== after,
-    beforeLength: before.length,
-    afterLength: after.length
-  };
-}
 function normalizePipelineError(err, stage) {
   if (err instanceof UnifiedOcrError) return err;
   const message = err instanceof Error ? err.message : String(err);
@@ -12123,9 +11989,6 @@ async function parseImage(buffer, options) {
     if (ocrMode === "gemini" || ocrMode === "claude" || ocrMode === "codex" || ocrMode === "ollama") {
       ocrProvider = createCliOcrProvider(ocrMode);
       actualOcrMode = ocrMode;
-    } else if (ocrMode === "tesseract") {
-      ocrProvider = await createTesseractProvider();
-      actualOcrMode = ocrMode;
     } else if (ocrMode === "auto") {
       const modesToTry = ["gemini", "claude", "codex", "ollama"];
       for (const mode of modesToTry) {
@@ -12137,10 +12000,6 @@ async function parseImage(buffer, options) {
           console.warn(`[kordoc] OCR auto-detection: ${mode} CLI not available or failed. Trying next.`, e);
         }
       }
-      if (!ocrProvider) {
-        ocrProvider = await createTesseractProvider();
-        actualOcrMode = "tesseract";
-      }
     }
     if (!ocrProvider) {
       return { success: false, fileType: "image", error: "\uC0AC\uC6A9 \uAC00\uB2A5\uD55C OCR \uD504\uB85C\uBC14\uC774\uB354\uB97C \uCC3E\uC744 \uC218 \uC5C6\uC2B5\uB2C8\uB2E4.", code: "PARSE_ERROR" };