@clazic/kordoc 2.4.17 → 2.4.18

package/dist/index.js

@@ -11346,17 +11346,39 @@ 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 +11389,6 @@ 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 +11408,6 @@ 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 +11522,11 @@ 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 +11985,6 @@ 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);