kordoc 4.2.2 → 4.2.4

package/dist/chunk-DCZVOIEO.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/chris_gomdori/workspace/kordoc/dist/chunk-DCZVOIEO.cjs","../src/page-range.ts"],"names":[],"mappings":"AAAA;ACSO,SAAS,cAAA,CAAe,IAAA,EAAyB,QAAA,EAA+B;AACrF,EAAA,MAAM,OAAA,kBAAS,IAAI,GAAA,CAAY,CAAA;AAC/B,EAAA,GAAA,CAAI,SAAA,GAAY,CAAA,EAAG,OAAO,MAAA;AAE1B,EAAA,GAAA,CAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,IAAA,CAAA,MAAW,EAAA,GAAK,IAAA,EAAM;AACpB,MAAA,MAAM,KAAA,EAAO,IAAA,CAAK,KAAA,CAAM,CAAC,CAAA;AACzB,MAAA,GAAA,CAAI,KAAA,GAAQ,EAAA,GAAK,KAAA,GAAQ,QAAA,EAAU,MAAA,CAAO,GAAA,CAAI,IAAI,CAAA;AAAA,IACpD;AACA,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,GAAA,CAAI,OAAO,KAAA,IAAS,SAAA,GAAY,IAAA,CAAK,IAAA,CAAK,EAAA,IAAM,EAAA,EAAI,OAAO,MAAA;AAE3D,EAAA,MAAM,MAAA,EAAQ,IAAA,CAAK,KAAA,CAAM,GAAG,CAAA;AAC5B,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,MAAM,QAAA,EAAU,IAAA,CAAK,IAAA,CAAK,CAAA;AAC1B,IAAA,GAAA,CAAI,CAAC,OAAA,EAAS,QAAA;AAEd,IAAA,MAAM,WAAA,EAAa,OAAA,CAAQ,KAAA,CAAM,qBAAqB,CAAA;AACtD,IAAA,GAAA,CAAI,UAAA,EAAY;AACd,MAAA,MAAM,MAAA,EAAQ,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,QAAA,CAAS,UAAA,CAAW,CAAC,CAAA,EAAG,EAAE,CAAC,CAAA;AACrD,MAAA,MAAM,IAAA,EAAM,IAAA,CAAK,GAAA,CAAI,QAAA,EAAU,QAAA,CAAS,UAAA,CAAW,CAAC,CAAA,EAAG,EAAE,CAAC,CAAA;AAC1D,MAAA,IAAA,CAAA,IAAS,EAAA,EAAI,KAAA,EAAO,EAAA,GAAK,GAAA,EAAK,CAAA,EAAA,EAAK,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA;AAAA,IACjD,EAAA,KAAO;AACL,MAAA,MAAM,KAAA,EAAO,QAAA,CAAS,OAAA,EAAS,EAAE,CAAA;AACjC,MAAA,GAAA,CAAI,CAAC,KAAA,CAAM,IAAI,EAAA,GAAK,KAAA,GAAQ,EAAA,GAAK,KAAA,GAAQ,QAAA,EAAU,MAAA,CAAO,GAAA,CAAI,IAAI,CAAA;AAAA,IACpE;AAAA,EACF;AAEA,EAAA,OAAO,MAAA;AACT;ADZA;AACA;AACE;AACF,wCAAC","file":"/Users/chris_gomdori/workspace/kordoc/dist/chunk-DCZVOIEO.cjs","sourcesContent":[null,"/** 페이지/섹션 범위 파싱 유틸리티 */\n\n/**\n * 페이지 범위 지정을 1-based Set<number>로 변환.\n *\n * @param spec - [1,2,3] 또는 \"1-3\" 또는 \"1,3,5-7\"\n * @param maxPages - 최대 페이지 수 (클램핑 상한)\n * @returns 1-based 페이지 번호 Set\n */\nexport function parsePageRange(spec: number[] | string, maxPages: number): Set<number> {\n  const result = new Set<number>()\n  if (maxPages <= 0) return result\n\n  if (Array.isArray(spec)) {\n    for (const n of spec) {\n      const page = Math.round(n)\n      if (page >= 1 && page <= maxPages) result.add(page)\n    }\n    return result\n  }\n\n  if (typeof spec !== \"string\" || spec.trim() === \"\") return result\n\n  const parts = spec.split(\",\")\n  for (const part of parts) {\n    const trimmed = part.trim()\n    if (!trimmed) continue\n\n    const rangeMatch = trimmed.match(/^(\\d+)\\s*-\\s*(\\d+)$/)\n    if (rangeMatch) {\n      const start = Math.max(1, parseInt(rangeMatch[1], 10))\n      const end = Math.min(maxPages, parseInt(rangeMatch[2], 10))\n      for (let i = start; i <= end; i++) result.add(i)\n    } else {\n      const page = parseInt(trimmed, 10)\n      if (!isNaN(page) && page >= 1 && page <= maxPages) result.add(page)\n    }\n  }\n\n  return result\n}\n"]}
+{"version":3,"sources":["/Users/mong-e/workspace/kordoc/dist/chunk-DCZVOIEO.cjs","../src/page-range.ts"],"names":[],"mappings":"AAAA;ACSO,SAAS,cAAA,CAAe,IAAA,EAAyB,QAAA,EAA+B;AACrF,EAAA,MAAM,OAAA,kBAAS,IAAI,GAAA,CAAY,CAAA;AAC/B,EAAA,GAAA,CAAI,SAAA,GAAY,CAAA,EAAG,OAAO,MAAA;AAE1B,EAAA,GAAA,CAAI,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,EAAG;AACvB,IAAA,IAAA,CAAA,MAAW,EAAA,GAAK,IAAA,EAAM;AACpB,MAAA,MAAM,KAAA,EAAO,IAAA,CAAK,KAAA,CAAM,CAAC,CAAA;AACzB,MAAA,GAAA,CAAI,KAAA,GAAQ,EAAA,GAAK,KAAA,GAAQ,QAAA,EAAU,MAAA,CAAO,GAAA,CAAI,IAAI,CAAA;AAAA,IACpD;AACA,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,GAAA,CAAI,OAAO,KAAA,IAAS,SAAA,GAAY,IAAA,CAAK,IAAA,CAAK,EAAA,IAAM,EAAA,EAAI,OAAO,MAAA;AAE3D,EAAA,MAAM,MAAA,EAAQ,IAAA,CAAK,KAAA,CAAM,GAAG,CAAA;AAC5B,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,MAAM,QAAA,EAAU,IAAA,CAAK,IAAA,CAAK,CAAA;AAC1B,IAAA,GAAA,CAAI,CAAC,OAAA,EAAS,QAAA;AAEd,IAAA,MAAM,WAAA,EAAa,OAAA,CAAQ,KAAA,CAAM,qBAAqB,CAAA;AACtD,IAAA,GAAA,CAAI,UAAA,EAAY;AACd,MAAA,MAAM,MAAA,EAAQ,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,QAAA,CAAS,UAAA,CAAW,CAAC,CAAA,EAAG,EAAE,CAAC,CAAA;AACrD,MAAA,MAAM,IAAA,EAAM,IAAA,CAAK,GAAA,CAAI,QAAA,EAAU,QAAA,CAAS,UAAA,CAAW,CAAC,CAAA,EAAG,EAAE,CAAC,CAAA;AAC1D,MAAA,IAAA,CAAA,IAAS,EAAA,EAAI,KAAA,EAAO,EAAA,GAAK,GAAA,EAAK,CAAA,EAAA,EAAK,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA;AAAA,IACjD,EAAA,KAAO;AACL,MAAA,MAAM,KAAA,EAAO,QAAA,CAAS,OAAA,EAAS,EAAE,CAAA;AACjC,MAAA,GAAA,CAAI,CAAC,KAAA,CAAM,IAAI,EAAA,GAAK,KAAA,GAAQ,EAAA,GAAK,KAAA,GAAQ,QAAA,EAAU,MAAA,CAAO,GAAA,CAAI,IAAI,CAAA;AAAA,IACpE;AAAA,EACF;AAEA,EAAA,OAAO,MAAA;AACT;ADZA;AACA;AACE;AACF,wCAAC","file":"/Users/mong-e/workspace/kordoc/dist/chunk-DCZVOIEO.cjs","sourcesContent":[null,"/** 페이지/섹션 범위 파싱 유틸리티 */\n\n/**\n * 페이지 범위 지정을 1-based Set<number>로 변환.\n *\n * @param spec - [1,2,3] 또는 \"1-3\" 또는 \"1,3,5-7\"\n * @param maxPages - 최대 페이지 수 (클램핑 상한)\n * @returns 1-based 페이지 번호 Set\n */\nexport function parsePageRange(spec: number[] | string, maxPages: number): Set<number> {\n  const result = new Set<number>()\n  if (maxPages <= 0) return result\n\n  if (Array.isArray(spec)) {\n    for (const n of spec) {\n      const page = Math.round(n)\n      if (page >= 1 && page <= maxPages) result.add(page)\n    }\n    return result\n  }\n\n  if (typeof spec !== \"string\" || spec.trim() === \"\") return result\n\n  const parts = spec.split(\",\")\n  for (const part of parts) {\n    const trimmed = part.trim()\n    if (!trimmed) continue\n\n    const rangeMatch = trimmed.match(/^(\\d+)\\s*-\\s*(\\d+)$/)\n    if (rangeMatch) {\n      const start = Math.max(1, parseInt(rangeMatch[1], 10))\n      const end = Math.min(maxPages, parseInt(rangeMatch[2], 10))\n      for (let i = start; i <= end; i++) result.add(i)\n    } else {\n      const page = parseInt(trimmed, 10)\n      if (!isNaN(page) && page >= 1 && page <= maxPages) result.add(page)\n    }\n  }\n\n  return result\n}\n"]}

package/dist/{chunk-MOOVF6CT.js → chunk-DOLK5KIS.js}

@@ -4,7 +4,7 @@ import {
 } from "./chunk-J2RZ444H.js";
 import {
   extractPageBlocksWithLines
-} from "./chunk-X77U36TV.js";
+} from "./chunk-GFTAPAJS.js";
 
 // src/ocr/ruling-lines.ts
 var INK_LUMA_MAX = 190;
@@ -557,4 +557,4 @@ export {
   runPdfOcr,
   ocrItemsToBlocks
 };
-//# sourceMappingURL=chunk-MOOVF6CT.js.map
+//# sourceMappingURL=chunk-DOLK5KIS.js.map

package/dist/{chunk-2L7MIKOM.js → chunk-E2K4SNMD.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 
 // src/utils.ts
-var VERSION = true ? "4.2.2" : "0.0.0-dev";
+var VERSION = true ? "4.2.4" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -121,4 +121,4 @@ export {
   safeMax,
   classifyError
 };
-//# sourceMappingURL=chunk-2L7MIKOM.js.map
+//# sourceMappingURL=chunk-E2K4SNMD.js.map

package/dist/{chunk-X77U36TV.js → chunk-GFTAPAJS.js}

@@ -4,7 +4,7 @@ import {
   HEADING_RATIO_H3,
   safeMax,
   safeMin
-} from "./chunk-GUYIXVV7.js";
+} from "./chunk-OBX3VSIZ.js";
 
 // src/pdf/image-regions.ts
 import { OPS } from "pdfjs-dist/legacy/build/pdf.mjs";
@@ -3340,4 +3340,4 @@ export {
   extractPageBlocksWithLines,
   mergeCrossPageTables
 };
-//# sourceMappingURL=chunk-X77U36TV.js.map
+//# sourceMappingURL=chunk-GFTAPAJS.js.map

package/dist/{chunk-ZF7QL4IY.js → chunk-GJO3ORFW.js}

@@ -10,4 +10,4 @@ export {
   HEADING_RATIO_H2,
   HEADING_RATIO_H3
 };
-//# sourceMappingURL=chunk-ZF7QL4IY.js.map
+//# sourceMappingURL=chunk-GJO3ORFW.js.map

package/dist/chunk-GJO3ORFW.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/types.ts"],"sourcesContent":["/** kordoc 공통 타입 정의 */\n\n// ─── 중간 표현 (Intermediate Representation) ─────────\n\nexport interface CellContext {\n  text: string\n  colSpan: number\n  rowSpan: number\n  /** HWP5 셀 열 주소 (0-based) — 병합 테이블 배치용 */\n  colAddr?: number\n  /** HWP5 셀 행 주소 (0-based) — 병합 테이블 배치용 */\n  rowAddr?: number\n}\n\n/** 블록 타입 — v2.0에서 heading, list, image, separator 추가 */\nexport type IRBlockType = \"paragraph\" | \"table\" | \"heading\" | \"list\" | \"image\" | \"separator\"\n\n/** 인라인 강조 run-span — 문단 텍스트를 서식 단위로 쪼갠 조각 (텍스트 연결 = block.text) */\nexport interface IRSpan {\n  text: string\n  bold?: boolean\n  italic?: boolean\n  strike?: boolean\n  code?: boolean\n}\n\nexport interface IRBlock {\n  type: IRBlockType\n  text?: string\n  table?: IRTable\n  /** 헤딩 레벨 (1-6), type=\"heading\"일 때 사용 */\n  level?: number\n  /** 원본 페이지 번호 (1-based) */\n  pageNumber?: number\n  /** 바운딩 박스 — PDF에서만 제공 */\n  bbox?: BoundingBox\n  /** 텍스트 스타일 정보 (선택) */\n  style?: InlineStyle\n  /** 리스트 타입, type=\"list\"일 때 사용 */\n  listType?: \"ordered\" | \"unordered\"\n  /** 중첩 리스트 아이템 */\n  children?: IRBlock[]\n  /** 하이퍼링크 URL */\n  href?: string\n  /** 각주/미주 텍스트 (인라인 삽입용) */\n  footnoteText?: string\n  /** 이미지 데이터 (type=\"image\"일 때) */\n  imageData?: ImageData\n  /** 인라인 강조 run-span (선택 — kordoc 생성 hwpx 왕복 채널 + 외래 실속성 볼드/이탤릭).\n   *  존재하면 마크다운 변환이 **·*·` 마커를 재방출한다 */\n  spans?: IRSpan[]\n  /** 인용문 문단 (선택 — kordoc 생성 hwpx 왕복 채널) — 마크다운 변환이 \"> \" 접두 재방출 */\n  quote?: boolean\n  /**\n   * 문단 들여쓰기(HWPUNIT, 선택) — HWPX paraPr `<hh:margin>` 자식요소형 hc:left\n   * (+양수 hc:intent 첫줄분). 마크다운 방출엔 쓰지 않는 관찰 슬롯 — gongmun 리스트\n   * depth 재유도·양식 분석 등 소비자 몫 (v4.0.4)\n   */\n  indent?: number\n  /**\n   * gongmun 리스트 단계(1~7, 선택) — indent를 levelIndent 단위로 역산한 소비 결과\n   * (v4.0.5). md 리스트 문법과 충돌하는 부호('- '·'1) ') 문단에만 채워지며,\n   * 마크다운 변환이 2칸/단계 선행 공백을 방출해 재생성 시 depth가 복원된다\n   */\n  listDepth?: number\n}\n\n/** 추출된 이미지 바이너리 데이터 */\nexport interface ImageData {\n  /** 이미지 바이너리 */\n  data: Uint8Array\n  /** MIME 타입 (image/png, image/jpeg, image/gif, image/bmp, image/wmf, image/emf) */\n  mimeType: string\n  /** 원본 파일명 (있는 경우) */\n  filename?: string\n}\n\n/** 바운딩 박스 — PDF 포인트 단위 (72pt = 1인치) */\nexport interface BoundingBox {\n  page: number\n  x: number\n  y: number\n  width: number\n  height: number\n}\n\n/** 인라인 텍스트 스타일 */\nexport interface InlineStyle {\n  bold?: boolean\n  italic?: boolean\n  /** 취소선 — 법령 개정문 등의 삭제 표시. 판정은 취소선 모양 whitelist (비트만 믿으면 오탐) */\n  strike?: boolean\n  fontSize?: number\n  fontName?: string\n}\n\nexport interface IRTable {\n  rows: number\n  cols: number\n  cells: IRCell[][]\n  /** 첫 행을 헤더로 렌더링할지 여부 (현재: rows > 1이면 true — 의미적 감지가 아닌 레이아웃 힌트) */\n  hasHeader: boolean\n  /** 표 캡션 (예: \"표 1. 부서별 예산\") — v3.0 */\n  caption?: string\n}\n\nexport interface IRCell {\n  text: string\n  colSpan: number\n  rowSpan: number\n  /**\n   * 셀 내부 블록 콘텐츠 — v3.0.\n   * 중첩 표·이미지·다중 문단을 구조 그대로, 문서(원문) 순서대로 보존한다.\n   * 표와 텍스트가 한 줄에 번갈아 놓인 셀도 배치 순서를 따른다 (v4.2.3, #49).\n   * blocks가 있으면 text는 blocks의 평탄화 텍스트(하위 호환용)다.\n   */\n  blocks?: IRBlock[]\n  /** 제목 셀 여부 (HWP5 width_ref bit2 / HWPX header 속성) — v3.0 */\n  isHeader?: boolean\n}\n\n// ─── 메타데이터 ─────────────────────────────────────\n\n/** 문서 메타데이터 — 각 포맷에서 추출 가능한 필드만 채워짐 */\nexport interface DocumentMetadata {\n  /** 문서 제목 */\n  title?: string\n  /** 작성자 */\n  author?: string\n  /** 작성 프로그램 (예: \"한글 2020\", \"Adobe Acrobat\") */\n  creator?: string\n  /** 생성일시 (ISO 8601) */\n  createdAt?: string\n  /** 수정일시 (ISO 8601) */\n  modifiedAt?: string\n  /** 페이지/섹션 수 */\n  pageCount?: number\n  /** 문서 포맷 버전 (예: HWP \"5.1.0.1\") */\n  version?: string\n  /** 설명 */\n  description?: string\n  /** 키워드 */\n  keywords?: string[]\n}\n\n// ─── 파싱 옵션 ──────────────────────────────────────\n\n/** 파싱 옵션 — parse() 함수에 전달 */\nexport interface ParseOptions {\n  /**\n   * 파싱할 페이지/섹션 범위 (1-based).\n   * - 배열: [1, 2, 3]\n   * - 문자열: \"1-3\", \"1,3,5-7\"\n   *\n   * PDF: 정확한 페이지 단위. HWP/HWPX: 섹션 단위 근사치.\n   */\n  pages?: number[] | string\n  /** 이미지 기반 PDF OCR (선택).\n   *  - `true`: 내장 엔진(PP-OCRv5 korean, ~18MB 자동 다운로드)으로 OCR 필요 판정\n   *    페이지만 인식 (스캔 페이지·글꼴 매핑 깨진 페이지). 정상 페이지는 파싱 결과 유지.\n   *  - `\"force\"`: 전 페이지를 내장 엔진으로 강제 OCR.\n   *  - 함수: 사용자 제공 OcrProvider (Claude Vision·Tesseract 등) — 판정은 `true`와 동일. */\n  ocr?: boolean | \"force\" | OcrProvider\n  /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */\n  onProgress?: (current: number, total: number) => void\n  /** PDF 머리글/바닥글 자동 제거 */\n  removeHeaderFooter?: boolean\n  /** 표 오른쪽 끝의 빈 열(서식 문서의 입력란) 보존 (#47).\n   *  기본 false: 마크다운 가독성을 위해 후행 빈 열을 트림.\n   *  양식 인식 경로(parse_form·fill)는 내부적으로 항상 켠다. */\n  keepTrailingEmptyCols?: boolean\n  /** 원본 파일 경로 (DRM COM fallback에 필요, 내부 전용) */\n  filePath?: string\n  /**\n   * PDF 수식 OCR 활성화 (기본 false).\n   *\n   * 활성화 시 각 PDF 페이지를 이미지로 렌더링 → YOLOv8 기반 수식 영역 검출 →\n   * TrOCR 기반 LaTeX 인식. 감지된 수식은 `$...$` (inline) / `$$...$$` (display) 로\n   * 블록 텍스트에 삽입된다.\n   *\n   * 필수 optional 의존성: `onnxruntime-node`, `@huggingface/transformers`,\n   * `@hyzyla/pdfium`, `sharp`. 미설치 시 parse 에 실패하지 않고 **경고만** 남기고\n   * 수식 인식은 skip 한다 (일반 텍스트 추출은 정상 동작).\n   *\n   * 모델(~155MB) 은 첫 사용 시 HuggingFace 에서 자동 다운로드 되어\n   * `~/.cache/kordoc/models/pix2text/` 에 SHA-256 검증과 함께 저장된다.\n   */\n  formulaOcr?: boolean\n  /**\n   * 레이아웃 표 페이지 반복 헤더(러닝 헤더) 정리 휴리스틱 활성화 (기본 false).\n   *\n   * 구형 HWP5 문서가 페이지마다 재삽입한 짧은 번호매김 러닝 헤더\n   * (\"2. 과제 구축 내용\")를 최초 1회만 남기고 이후 중복을 제거한다.\n   * 다만 페이지 위치 정보가 없는 HWP5 특성상 이 휴리스틱은 정당하게 반복되는\n   * 번호매김 문단(예: 붙임/별지별 재번호 \"1. 목적\")도 삭제할 수 있어 opt-in 으로\n   * 둔다. 활성 시 제거가 발생하면 HIDDEN_TEXT_FILTERED 경고를 남긴다.\n   */\n  dedupeRunningHeaders?: boolean\n  /**\n   * 추출된 이미지를 마크다운에 base64 data URI 로 인라인 (기본 false).\n   *\n   * 활성화 시 `![image](image_001.bmp)` 참조를 `![image](data:image/png;base64,...)` 로\n   * 치환한다. 임베드된 BMP 는 PNG 로 무손실 압축 후 인라인하여 용량을 크게 줄인다.\n   * 별도 이미지 파일 없이 자체 완결형 마크다운이 되어 MCP/AI 에이전트 소비에 적합하다.\n   * (현재 HWP5 경로 지원)\n   */\n  inlineImages?: boolean\n}\n\n// ─── 파싱 경고 ──────────────────────────────────────\n\n/** 파싱 중 스킵/실패한 요소 보고 */\nexport interface ParseWarning {\n  /** 관련 페이지 번호 (알 수 있는 경우) */\n  page?: number\n  /** 경고 메시지 */\n  message: string\n  /** 구조화된 경고 코드 */\n  code: WarningCode\n}\n\nexport type WarningCode =\n  | \"SKIPPED_IMAGE\"\n  | \"SKIPPED_OLE\"\n  | \"TRUNCATED_TABLE\"\n  | \"OCR_FALLBACK\"\n  | \"UNSUPPORTED_ELEMENT\"\n  | \"BROKEN_ZIP_RECOVERY\"\n  | \"HIDDEN_TEXT_FILTERED\"\n  | \"MALFORMED_XML\"\n  | \"PARTIAL_PARSE\"\n  | \"LENIENT_CFB_RECOVERY\"\n  | \"NEEDS_OCR\"\n  | \"OCR_FAILED\"\n  | \"OCR_APPLIED\"\n  | \"COM_EMPTY\"\n  | \"DRM_COM_FALLBACK\"\n\n/** 문서 구조 (헤딩 트리) */\nexport interface OutlineItem {\n  level: number\n  text: string\n  pageNumber?: number\n}\n\n// ─── 에러 코드 ──────────────────────────────────────\n\n/** 구조화된 에러 코드 — 프로그래밍적 에러 핸들링용 */\nexport type ErrorCode =\n  | \"EMPTY_INPUT\"\n  | \"UNSUPPORTED_FORMAT\"\n  | \"ENCRYPTED\"\n  | \"DRM_PROTECTED\"\n  | \"CORRUPTED\"\n  | \"DECOMPRESSION_BOMB\"\n  | \"ZIP_BOMB\"\n  | \"IMAGE_BASED_PDF\"\n  | \"NO_SECTIONS\"\n  | \"PARSE_ERROR\"\n  | \"MISSING_DEPENDENCY\"\n\n// ─── 파싱 결과 (discriminated union) ────────────────\n\nexport type FileType = \"hwpx\" | \"hwp\" | \"hwp3\" | \"hwpml\" | \"pdf\" | \"xlsx\" | \"xls\" | \"docx\" | \"image\" | \"unknown\"\n\ninterface ParseResultBase {\n  fileType: FileType\n  /** 페이지/섹션 수 — PDF: 실제 페이지 수, HWP/HWPX: 섹션 수, XLSX: 시트 수 */\n  pageCount?: number\n  /** 이미지 기반 PDF 여부 (텍스트 추출 불가) */\n  isImageBased?: boolean\n}\n\nexport interface ParseSuccess extends ParseResultBase {\n  success: true\n  /** 추출된 마크다운 텍스트 */\n  markdown: string\n  /**\n   * 중간 표현 블록 (구조화된 데이터 접근용).\n   * 블록은 문서(원문) 읽기 순서를 따른다 — 한 문단 안에 글자취급(treatAsChar)\n   * 표와 텍스트가 섞여 있어도 배치 순서대로 방출된다 (v4.2.3, #50).\n   */\n  blocks: IRBlock[]\n  /** 문서 메타데이터 */\n  metadata?: DocumentMetadata\n  /** 문서 구조 (헤딩 트리) — v2.0 */\n  outline?: OutlineItem[]\n  /** 파싱 중 발생한 경고 — v2.0 */\n  warnings?: ParseWarning[]\n  /** 추출된 이미지 목록 — 마크다운에서 파일명으로 참조됨 */\n  images?: ExtractedImage[]\n  /** 페이지별 텍스트 품질 신호 — PDF에서만 제공 */\n  pageQuality?: PageQuality[]\n  /** 문서 단위 품질 요약 — PDF에서만 제공 */\n  qualitySummary?: DocumentQualitySummary\n}\n\n/** 페이지별 텍스트 품질 신호 (PDF 전용). 자세한 설명은 src/pdf/quality.ts */\nexport interface PageQuality {\n  page: number\n  textChars: number\n  hangulRatio: number\n  controlCharRatio: number\n  replacementCharRatio: number\n  puaRatio: number\n  needsOcr: boolean\n  ocrReason?: \"low_text\" | \"high_pua\" | \"high_control\" | \"high_replacement\" | \"garbled_hangul\"\n}\n\n/** 문서 단위 품질 요약 (PDF 전용). */\nexport interface DocumentQualitySummary {\n  totalPages: number\n  totalTextChars: number\n  avgHangulRatio: number\n  avgControlCharRatio: number\n  avgReplacementCharRatio: number\n  avgPuaRatio: number\n  lowTextPageCount: number\n  highPuaPageCount: number\n  needsOcr: boolean\n  ocrCandidatePages: number[]\n}\n\n/** 추출된 이미지 — ParseSuccess.images에 포함 */\nexport interface ExtractedImage {\n  /** 마크다운에서 참조되는 파일명 (예: image_001.png) */\n  filename: string\n  /** 이미지 바이너리 */\n  data: Uint8Array\n  /** MIME 타입 */\n  mimeType: string\n}\n\nexport interface ParseFailure extends ParseResultBase {\n  success: false\n  /** 오류 메시지 */\n  error: string\n  /** 구조화된 에러 코드 */\n  code?: ErrorCode\n}\n\nexport type ParseResult = ParseSuccess | ParseFailure\n\n// ─── 문서 비교 (Diff) ───────────────────────────────\n\nexport type DiffChangeType = \"added\" | \"removed\" | \"modified\" | \"unchanged\"\n\nexport interface BlockDiff {\n  type: DiffChangeType\n  /** 원본 블록 (added이면 undefined) */\n  before?: IRBlock\n  /** 변경 후 블록 (removed이면 undefined) */\n  after?: IRBlock\n  /** modified 테이블의 셀 단위 diff */\n  cellDiffs?: CellDiff[][]\n  /** 유사도 (0-1) */\n  similarity?: number\n}\n\nexport interface CellDiff {\n  type: DiffChangeType\n  before?: string\n  after?: string\n}\n\nexport interface DiffResult {\n  stats: { added: number; removed: number; modified: number; unchanged: number }\n  diffs: BlockDiff[]\n}\n\n// ─── 라운드트립 패치 (v3.0) ─────────────────────────\n\n/** 패치 중 매핑 실패/미지원으로 건너뛴 항목 — silent 실패 금지 */\nexport interface PatchSkip {\n  /** 건너뛴 사유 */\n  reason: string\n  /** 원본 쪽 내용 요약 (최대 80자) */\n  before?: string\n  /** 편집 쪽 내용 요약 (최대 80자) */\n  after?: string\n  /**\n   * 부분 적용 표시 — 변경이 적용은 됐지만(applied 계상) 편집 원형 그대로는\n   * 아님 (예: 셀 내 줄 추가를 마지막 문단에 병합). 완전 미적용 skip과 구분.\n   */\n  partial?: boolean\n}\n\n/** patchHwpx 옵션 */\nexport interface PatchOptions {\n  /** 패치 후 재파싱 자동 검증 (기본 true) */\n  verify?: boolean\n}\n\n/** patchHwpx / patchBlocks 결과 */\nexport interface PatchResult {\n  success: boolean\n  /** 패치된 HWPX (success=true) */\n  data?: Uint8Array\n  /** 적용된 변경 수 */\n  applied: number\n  /** 매핑 실패 항목 (이유 포함) */\n  skipped: PatchSkip[]\n  /**\n   * 무손실 검증 (patchHwpx/patchHwp 전용): 패치본 재파싱 vs 편집 마크다운의\n   * 잔차 diff — modified/added/removed가 0이어야 의도가 전부 반영된 것.\n   * session.patchBlocks는 이 필드를 채우지 않는다 (changes 참조).\n   */\n  verification?: DiffResult\n  /**\n   * 변경 가시화 (session.patchBlocks 전용): 패치 전 → 후 문서 diff —\n   * 적용된 편집 수만큼 modified가 나오는 것이 정상. verification과 의미가\n   * 정반대이므로 혼용 금지.\n   */\n  changes?: DiffResult\n  /** 실패 사유 (success=false) */\n  error?: string\n}\n\n// ─── 양식 인식 ──────────────────────────────────────\n\nexport interface FormField {\n  label: string\n  value: string\n  /** 0-based 소스 행 */\n  row: number\n  /** 0-based 소스 열 */\n  col: number\n  /** 매칭된 입력 키(정규화) — 모호 라벨 거부 가드(fillWithUniqueGuard)의 집계 기준 */\n  key?: string\n}\n\nexport interface FormResult {\n  fields: FormField[]\n  /** 양식 확신도 (0-1) */\n  confidence: number\n}\n\n// ─── OCR 프로바이더 ─────────────────────────────────\n\n/** 사용자 제공 OCR 함수 — 페이지 이미지를 받아 텍스트 반환 */\nexport type OcrProvider = (\n  pageImage: Uint8Array,\n  pageNumber: number,\n  mimeType: \"image/png\"\n) => Promise<string>\n\n// ─── Watch 모드 ─────────────────────────────────────\n\nexport interface WatchOptions {\n  dir: string\n  outDir?: string\n  webhook?: string\n  format?: \"markdown\" | \"json\"\n  pages?: string\n  silent?: boolean\n}\n\n// ─── 헤딩 감지 공통 임계값 ──────────────────────────\n\n/** 폰트 크기 비율 → heading level (전 파서 공통) */\nexport const HEADING_RATIO_H1 = 1.5\nexport const HEADING_RATIO_H2 = 1.3\nexport const HEADING_RATIO_H3 = 1.15\n\n// ─── 내부 파서 반환 타입 ─────────────────────────────\n\n/** 내부 파서가 index.ts에 반환하는 공통 타입 (HWP5/HWPX/PDF/XLSX/DOCX) */\nexport interface InternalParseResult {\n  markdown: string\n  blocks: IRBlock[]\n  metadata?: DocumentMetadata\n  outline?: OutlineItem[]\n  warnings?: ParseWarning[]\n  images?: ExtractedImage[]\n  /** PDF 전용: 이미지 기반 PDF 여부 */\n  isImageBased?: boolean\n  /** PDF 전용: 페이지별 품질 신호 */\n  pageQuality?: PageQuality[]\n  /** PDF 전용: 문서 단위 품질 요약 */\n  qualitySummary?: DocumentQualitySummary\n}\n"],"mappings":";;;AA4cO,IAAM,mBAAmB;AACzB,IAAM,mBAAmB;AACzB,IAAM,mBAAmB;","names":[]}

package/dist/chunk-GS7T56RP.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/chris_gomdori/workspace/kordoc/dist/chunk-GS7T56RP.cjs","../node_modules/tsup/assets/cjs_shims.js"],"names":[],"mappings":"AAAA;ACKA,IAAM,iBAAA,EAAmB,CAAA,EAAA,GACvB,OAAO,SAAA,IAAa,YAAA,EAChB,IAAI,GAAA,CAAI,CAAA,KAAA,EAAQ,UAAU,CAAA,CAAA;AAK8B;ADT4B;AACA;AACA;AACA","file":"/Users/chris_gomdori/workspace/kordoc/dist/chunk-GS7T56RP.cjs","sourcesContent":[null,"// Shim globals in cjs bundle\n// There's a weird bug that esbuild will always inject importMetaUrl\n// if we export it as `const importMetaUrl = ... __filename ...`\n// But using a function will not cause this issue\n\nconst getImportMetaUrl = () => \n  typeof document === \"undefined\" \n    ? new URL(`file:${__filename}`).href \n    : (document.currentScript && document.currentScript.tagName.toUpperCase() === 'SCRIPT') \n      ? document.currentScript.src \n      : new URL(\"main.js\", document.baseURI).href;\n\nexport const importMetaUrl = /* @__PURE__ */ getImportMetaUrl()\n"]}
+{"version":3,"sources":["/Users/mong-e/workspace/kordoc/dist/chunk-GS7T56RP.cjs","../node_modules/tsup/assets/cjs_shims.js"],"names":[],"mappings":"AAAA;ACKA,IAAM,iBAAA,EAAmB,CAAA,EAAA,GACvB,OAAO,SAAA,IAAa,YAAA,EAChB,IAAI,GAAA,CAAI,CAAA,KAAA,EAAQ,UAAU,CAAA,CAAA;AAK8B;ADT4B;AACA;AACA;AACA","file":"/Users/mong-e/workspace/kordoc/dist/chunk-GS7T56RP.cjs","sourcesContent":[null,"// Shim globals in cjs bundle\n// There's a weird bug that esbuild will always inject importMetaUrl\n// if we export it as `const importMetaUrl = ... __filename ...`\n// But using a function will not cause this issue\n\nconst getImportMetaUrl = () => \n  typeof document === \"undefined\" \n    ? new URL(`file:${__filename}`).href \n    : (document.currentScript && document.currentScript.tagName.toUpperCase() === 'SCRIPT') \n      ? document.currentScript.src \n      : new URL(\"main.js\", document.baseURI).href;\n\nexport const importMetaUrl = /* @__PURE__ */ getImportMetaUrl()\n"]}

package/dist/{chunk-FHYLLKIP.js → chunk-H6O65EPK.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   KordocError
-} from "./chunk-2L7MIKOM.js";
+} from "./chunk-E2K4SNMD.js";
 
 // src/hwpx/parser-shared.ts
 import { DOMParser } from "@xmldom/xmldom";
@@ -273,4 +273,4 @@ export {
   DEFAULT_CHAR,
   parseRenderStyles
 };
-//# sourceMappingURL=chunk-FHYLLKIP.js.map
+//# sourceMappingURL=chunk-H6O65EPK.js.map

package/dist/chunk-H7OWEYH2.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/chris_gomdori/workspace/kordoc/dist/chunk-H7OWEYH2.cjs","../src/pdf/formula/models.ts"],"names":[],"mappings":"AAAA;ACUA,gCAA2B;AAC3B,wBAAiC;AACjC,uCAA4C;AAC5C;AACA,wBAAwB;AACxB,4BAA8B;AAC9B,4CAAyB;AACzB,gCAAyB;AAWlB,IAAM,UAAA,EAAuB;AAAA,EAClC,IAAA,EAAM,cAAA;AAAA,EACN,QAAA,EAAU,UAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,kBAAA,EAA+B;AAAA,EAC1C,IAAA,EAAM,sBAAA;AAAA,EACN,QAAA,EAAU,oBAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,kBAAA,EAA+B;AAAA,EAC1C,IAAA,EAAM,sBAAA;AAAA,EACN,QAAA,EAAU,oBAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,cAAA,EAA2B;AAAA,EACtC,IAAA,EAAM,wBAAA;AAAA,EACN,QAAA,EAAU,gBAAA;AAAA,EACV,GAAA,EAAK,4EAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,mBAAA,EAA+C;AAAA,EAC1D,SAAA;AAAA,EACA,iBAAA;AAAA,EACA,iBAAA;AAAA,EACA;AACF,CAAA;AAOO,SAAS,YAAA,CAAa,MAAA,EAAwB;AACnD,EAAA,MAAM,SAAA,EAAW,OAAA,CAAQ,GAAA,CAAI,kBAAA;AAC7B,EAAA,GAAA,CAAI,SAAA,GAAY,QAAA,CAAS,IAAA,CAAK,CAAA,EAAG;AAC/B,IAAA,OAAO,wBAAA,QAAK,EAAU,MAAM,CAAA;AAAA,EAC9B;AACA,EAAA,OAAO,wBAAA,yBAAK,CAAQ,EAAG,QAAA,EAAU,QAAA,EAAU,QAAA,EAAU,MAAM,CAAA;AAC7D;AAEO,SAAS,mBAAA,CAAA,EAA8B;AAC5C,EAAA,OAAO,YAAA,CAAa,UAAU,CAAA;AAChC;AAaA,MAAA,SAAsB,gBAAA,CAAiB,GAAA,EAAa,KAAA,EAAyD;AAC3G,EAAA,MAAM,OAAA,EAAwB,CAAC,CAAA;AAC/B,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,MAAM,UAAA,EAAY,wBAAA,GAAK,EAAK,IAAA,CAAK,QAAQ,CAAA;AACzC,IAAA,IAAI,OAAA,EAAS,KAAA;AACb,IAAA,IAAI;AACF,MAAA,MAAM,EAAA,EAAI,MAAM,4BAAA,SAAc,CAAA;AAC9B,MAAA,OAAA,EAAS,CAAA,CAAE,MAAA,CAAO,EAAA,GAAK,CAAA,CAAE,KAAA,EAAO,CAAA;AAAA,IAClC,EAAA,WAAQ;AACN,MAAA,OAAA,EAAS,KAAA;AAAA,IACX;AACA,IAAA,GAAA,CAAI,CAAC,MAAA,EAAQ;AACX,MAAA,MAAA,CAAO,IAAA,CAAK,EAAE,IAAA,EAAM,SAAA,EAAW,MAAA,EAAQ,KAAA,EAAO,QAAA,EAAU,MAAM,CAAC,CAAA;AAC/D,MAAA,QAAA;AAAA,IACF;AACA,IAAA,IAAI;AACF,MAAA,MAAM,OAAA,EAAS,MAAM,YAAA,CAAa,SAAS,CAAA;AAC3C,MAAA,GAAA,CAAI,OAAA,IAAW,IAAA,CAAK,MAAA,EAAQ;AAC1B,QAAA,MAAA,CAAO,IAAA,CAAK,EAAE,IAAA,EAAM,SAAA,EAAW,MAAA,EAAQ,IAAA,EAAM,QAAA,EAAU,KAAK,CAAC,CAAA;AAAA,MAC/D,EAAA,KAAO;AACL,QAAA,MAAA,CAAO,IAAA,CAAK;AAAA,UACV,IAAA;AAAA,UACA,SAAA;AAAA,UACA,MAAA,EAAQ,IAAA;AAAA,UACR,QAAA,EAAU,KAAA;AAAA,UACV,aAAA,EAAe,CAAA,0BAAA,EAA6B,IAAA,CAAK,MAAM,CAAA,MAAA,EAAS,MAAM,CAAA;AAAA,QAAA;AACvE,MAAA;AACH,IAAA;AAEA,MAAA;AAAY,QAAA;AACV,QAAA;AACA,QAAA;AACQ,QAAA;AACE,QAAA;AACgD,MAAA;AAC3D,IAAA;AACH,EAAA;AAEF,EAAA;AACF;AAGA;AACE,EAAA;AACF;AAoBA;AACE,EAAA;AAEA,EAAA;AACE,IAAA;AAEA,IAAA;AACE,sBAAA;AAAa,QAAA;AACX,QAAA;AACY,QAAA;AACL,QAAA;AACA,QAAA;AACE,MAAA;AAEX,MAAA;AAAA,IAAA;AAIF,IAAA;AACE,MAAA;AAAsB,IAAA;AAChB,IAAA;AAIR,IAAA;AAAgD,EAAA;AAEpD;AAGA;AACE,EAAA;AACF;AAGA;AACE,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACE,oBAAA;AACA,IAAA;AAAA,EAAA;AAEF,EAAA;AACE,IAAA;AAAsB,EAAA;AAChB,EAAA;AACR,EAAA;AACF;AAEA;AACE,EAAA;AACE,IAAA;AACA,IAAA;AAAwC,EAAA;AAExC,IAAA;AAAO,EAAA;AAET,EAAA;AACE,IAAA;AACA,IAAA;AAAkB,EAAA;AAElB,IAAA;AAAO,EAAA;AAEX;AAEA;AAME,EAAA;AACA,EAAA;AAEA,EAAA;AAAmC,IAAA;AACxB;AAAA,MAAA;AAEO,IAAA;AAChB,EAAA;AAEF,EAAA;AACE,IAAA;AAAU,MAAA;AACiE,IAAA;AAC3E,EAAA;AAGF,EAAA;AACA,EAAA;AACA,EAAA;AAEA,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACE,MAAA;AACA,sBAAA;AAAa,QAAA;AACX,QAAA;AACA,QAAA;AACA,QAAA;AACO,MAAA;AACR,IAAA;AAEH,IAAA;AAAyB,EAAA;AAEzB,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAA+D,EAAA;AAGjE,kBAAA;AAAa,IAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,EAAA;AAIT,EAAA;AACA,EAAA;AACE,IAAA;AAAoC,EAAA;AAEpC,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAAiE,EAAA;AAGnE,EAAA;AACE,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAAU,MAAA;AAC4D,IAAA;AACtE,EAAA;AAGF,EAAA;AACA,kBAAA;AAAa,IAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,EAAA;AAEX;AAEA;AACE,EAAA;AACA,EAAA;AACA,EAAA;AACE,IAAA;AACE,MAAA;AAAc,IAAA;AAEhB,EAAA;AAEF,EAAA;AACF;ADrFA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/Users/chris_gomdori/workspace/kordoc/dist/chunk-H7OWEYH2.cjs","sourcesContent":[null,"/**\n * 수식 OCR 모델 자동 다운로드 + SHA-256 검증\n *\n * 캐시 위치: `~/.cache/kordoc/models/pix2text/`\n * 총 ~155MB (MFD 44MB + MFR encoder 87MB + MFR decoder 30MB + tokenizer 40KB)\n *\n * HF URL + SHA 는 Docufinder archive/rust-formula-ocr-wip 에서 검증 완료됨.\n * 모델 버전 갱신 시 URL + SHA 를 함께 갱신해야 한다.\n */\n\nimport { createHash } from \"crypto\"\nimport { createReadStream } from \"fs\"\nimport { mkdir, stat, unlink, rename } from \"fs/promises\"\nimport { createWriteStream } from \"fs\"\nimport { homedir } from \"os\"\nimport { join, dirname } from \"path\"\nimport { pipeline } from \"stream/promises\"\nimport { Readable } from \"stream\"\n\nexport interface ModelSpec {\n  name: string\n  filename: string\n  url: string\n  sha256: string\n  sizeMb: number\n}\n\n/** 모델 스펙 (SHA-256 은 실제 다운로드로 검증됨 — 변경 금지) */\nexport const MFD_MODEL: ModelSpec = {\n  name: \"Pix2Text MFD\",\n  filename: \"mfd.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfd/resolve/main/mfd-v20240618.onnx\",\n  sha256: \"51a8854743b17ae654729af8db82a630c1ccfa06debf4856c8b28055f87d02c1\",\n  sizeMb: 42,\n}\n\nexport const MFR_ENCODER_MODEL: ModelSpec = {\n  name: \"Pix2Text MFR encoder\",\n  filename: \"encoder_model.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/encoder_model.onnx\",\n  sha256: \"bd8d5c322792e9ec45793af5569e9748f82a3d728a9e00213dbfc56c1486f37d\",\n  sizeMb: 87,\n}\n\nexport const MFR_DECODER_MODEL: ModelSpec = {\n  name: \"Pix2Text MFR decoder\",\n  filename: \"decoder_model.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/decoder_model.onnx\",\n  sha256: \"fd0f92d7a012f3dae41e1ac79421aea0ea888b5a66cb3f9a004e424f82f3daed\",\n  sizeMb: 30,\n}\n\nexport const MFR_TOKENIZER: ModelSpec = {\n  name: \"Pix2Text MFR tokenizer\",\n  filename: \"tokenizer.json\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/tokenizer.json\",\n  sha256: \"3e2ab757277d22639bec28c9d7972e352d3d1dba223051fa674002dc5ab64df3\",\n  sizeMb: 1,\n}\n\nexport const ALL_FORMULA_MODELS: ReadonlyArray<ModelSpec> = [\n  MFD_MODEL,\n  MFR_ENCODER_MODEL,\n  MFR_DECODER_MODEL,\n  MFR_TOKENIZER,\n]\n\n/**\n * 모델 캐시 하위 디렉토리 경로. 환경 변수 `KORDOC_MODEL_CACHE` 가 설정되면\n * 해당 경로를 우선 사용, 없으면 `~/.cache/kordoc/models/<subdir>/`.\n * (수식 OCR 외 텍스트 OCR 등 다른 모델군도 공용 — subdir 로 격리)\n */\nexport function getModelsDir(subdir: string): string {\n  const override = process.env.KORDOC_MODEL_CACHE\n  if (override && override.trim()) {\n    return join(override, subdir)\n  }\n  return join(homedir(), \".cache\", \"kordoc\", \"models\", subdir)\n}\n\nexport function getFormulaModelsDir(): string {\n  return getModelsDir(\"pix2text\")\n}\n\nexport interface ModelStatus {\n  spec: ModelSpec\n  localPath: string\n  exists: boolean\n  /** SHA 검증까지 끝났는가 (exists && valid) */\n  verified: boolean\n  /** 검증 실패 시 이유 */\n  invalidReason?: string\n}\n\n/** 지정 모델군의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getModelStatusIn(dir: string, specs: ReadonlyArray<ModelSpec>): Promise<ModelStatus[]> {\n  const result: ModelStatus[] = []\n  for (const spec of specs) {\n    const localPath = join(dir, spec.filename)\n    let exists = false\n    try {\n      const s = await stat(localPath)\n      exists = s.isFile() && s.size > 0\n    } catch {\n      exists = false\n    }\n    if (!exists) {\n      result.push({ spec, localPath, exists: false, verified: false })\n      continue\n    }\n    try {\n      const actual = await sha256OfFile(localPath)\n      if (actual === spec.sha256) {\n        result.push({ spec, localPath, exists: true, verified: true })\n      } else {\n        result.push({\n          spec,\n          localPath,\n          exists: true,\n          verified: false,\n          invalidReason: `SHA256 mismatch: expected ${spec.sha256}, got ${actual}`,\n        })\n      }\n    } catch (e) {\n      result.push({\n        spec,\n        localPath,\n        exists: true,\n        verified: false,\n        invalidReason: `SHA compute failed: ${(e as Error).message}`,\n      })\n    }\n  }\n  return result\n}\n\n/** 모든 수식 모델의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getFormulaModelStatus(): Promise<ModelStatus[]> {\n  return getModelStatusIn(getFormulaModelsDir(), ALL_FORMULA_MODELS)\n}\n\nexport interface DownloadProgress {\n  spec: ModelSpec\n  /** 현재까지 다운로드한 바이트 */\n  downloaded: number\n  /** 전체 바이트 (알 수 있으면) */\n  total: number | null\n  /** \"download\" | \"verify\" | \"done\" | \"skip\" */\n  phase: \"download\" | \"verify\" | \"done\" | \"skip\" | \"error\"\n  message?: string\n}\n\nexport type ProgressHandler = (p: DownloadProgress) => void\n\n/**\n * 지정 모델군을 다운로드/검증한다.\n * - 이미 있고 SHA 일치 → skip\n * - 없음 → 다운로드 후 SHA 검증. 실패 시 파일 삭제.\n */\nexport async function ensureModelsIn(dir: string, specs: ReadonlyArray<ModelSpec>, onProgress?: ProgressHandler): Promise<void> {\n  await mkdir(dir, { recursive: true })\n\n  for (const spec of specs) {\n    const localPath = join(dir, spec.filename)\n\n    if (await isExistingValid(localPath, spec.sha256)) {\n      onProgress?.({\n        spec,\n        downloaded: 0,\n        total: null,\n        phase: \"skip\",\n        message: \"이미 존재 + SHA 일치\",\n      })\n      continue\n    }\n\n    // 기존 파일 있지만 SHA 불일치 → 삭제\n    try {\n      await unlink(localPath)\n    } catch {\n      // 없을 수 있음\n    }\n\n    await downloadToFile(spec, localPath, onProgress)\n  }\n}\n\n/** 필요한 모든 수식 모델을 다운로드/검증한다. */\nexport async function ensureFormulaModels(onProgress?: ProgressHandler): Promise<void> {\n  return ensureModelsIn(getFormulaModelsDir(), ALL_FORMULA_MODELS, onProgress)\n}\n\n/** 단일 모델만 확인/다운로드 (진단 UI 에서 개별 상태 새로고침용) */\nexport async function ensureSingleModel(spec: ModelSpec, onProgress?: ProgressHandler): Promise<void> {\n  const dir = getFormulaModelsDir()\n  await mkdir(dir, { recursive: true })\n  const localPath = join(dir, spec.filename)\n  if (await isExistingValid(localPath, spec.sha256)) {\n    onProgress?.({ spec, downloaded: 0, total: null, phase: \"skip\" })\n    return\n  }\n  try {\n    await unlink(localPath)\n  } catch {}\n  await downloadToFile(spec, localPath, onProgress)\n}\n\nasync function isExistingValid(localPath: string, sha256Expected: string): Promise<boolean> {\n  try {\n    const s = await stat(localPath)\n    if (!s.isFile() || s.size === 0) return false\n  } catch {\n    return false\n  }\n  try {\n    const actual = await sha256OfFile(localPath)\n    return actual === sha256Expected\n  } catch {\n    return false\n  }\n}\n\nasync function downloadToFile(\n  spec: ModelSpec,\n  localPath: string,\n  onProgress?: ProgressHandler,\n): Promise<void> {\n  // 먼저 .part 로 받고 검증 후 rename — 중단된 다운로드가 \"정상 파일\"로 오인되는 걸 방지\n  const partPath = `${localPath}.part`\n  await mkdir(dirname(localPath), { recursive: true })\n\n  const resp = await fetch(spec.url, {\n    headers: {\n      // HF CDN 은 UA 없으면 가끔 403 을 뱉는다\n      \"User-Agent\": \"kordoc-formula-ocr/1.0 (+https://github.com/chrisryugj/kordoc)\",\n    },\n  })\n  if (!resp.ok || !resp.body) {\n    throw new Error(\n      `${spec.name} 다운로드 실패: HTTP ${resp.status} ${resp.statusText} (${spec.url})`,\n    )\n  }\n\n  const lenHeader = resp.headers.get(\"content-length\")\n  const total = lenHeader ? Number.parseInt(lenHeader, 10) : null\n  let downloaded = 0\n\n  const ws = createWriteStream(partPath)\n  try {\n    const reader = Readable.fromWeb(resp.body as unknown as import(\"stream/web\").ReadableStream)\n    reader.on(\"data\", (chunk: Buffer | Uint8Array) => {\n      downloaded += chunk.length\n      onProgress?.({\n        spec,\n        downloaded,\n        total,\n        phase: \"download\",\n      })\n    })\n    await pipeline(reader, ws)\n  } catch (e) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(`${spec.name} 스트리밍 실패: ${(e as Error).message}`)\n  }\n\n  onProgress?.({\n    spec,\n    downloaded,\n    total,\n    phase: \"verify\",\n  })\n\n  // SHA 검증\n  let actual: string\n  try {\n    actual = await sha256OfFile(partPath)\n  } catch (e) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(`${spec.name} SHA 계산 실패: ${(e as Error).message}`)\n  }\n\n  if (actual !== spec.sha256) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(\n      `${spec.name} SHA256 mismatch: expected ${spec.sha256}, got ${actual} — 모델 URL 이 오염되었거나 전송 중 손상되었습니다.`,\n    )\n  }\n\n  await rename(partPath, localPath)\n  onProgress?.({\n    spec,\n    downloaded,\n    total,\n    phase: \"done\",\n  })\n}\n\nasync function sha256OfFile(p: string): Promise<string> {\n  const h = createHash(\"sha256\")\n  const stream = createReadStream(p)\n  await pipeline(stream, async function* (src) {\n    for await (const chunk of src) {\n      h.update(chunk)\n      // yield 안 함 (소비만)\n    }\n  })\n  return h.digest(\"hex\")\n}\n"]}
+{"version":3,"sources":["/Users/mong-e/workspace/kordoc/dist/chunk-H7OWEYH2.cjs","../src/pdf/formula/models.ts"],"names":[],"mappings":"AAAA;ACUA,gCAA2B;AAC3B,wBAAiC;AACjC,uCAA4C;AAC5C;AACA,wBAAwB;AACxB,4BAA8B;AAC9B,4CAAyB;AACzB,gCAAyB;AAWlB,IAAM,UAAA,EAAuB;AAAA,EAClC,IAAA,EAAM,cAAA;AAAA,EACN,QAAA,EAAU,UAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,kBAAA,EAA+B;AAAA,EAC1C,IAAA,EAAM,sBAAA;AAAA,EACN,QAAA,EAAU,oBAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,kBAAA,EAA+B;AAAA,EAC1C,IAAA,EAAM,sBAAA;AAAA,EACN,QAAA,EAAU,oBAAA;AAAA,EACV,GAAA,EAAK,gFAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,cAAA,EAA2B;AAAA,EACtC,IAAA,EAAM,wBAAA;AAAA,EACN,QAAA,EAAU,gBAAA;AAAA,EACV,GAAA,EAAK,4EAAA;AAAA,EACL,MAAA,EAAQ,kEAAA;AAAA,EACR,MAAA,EAAQ;AACV,CAAA;AAEO,IAAM,mBAAA,EAA+C;AAAA,EAC1D,SAAA;AAAA,EACA,iBAAA;AAAA,EACA,iBAAA;AAAA,EACA;AACF,CAAA;AAOO,SAAS,YAAA,CAAa,MAAA,EAAwB;AACnD,EAAA,MAAM,SAAA,EAAW,OAAA,CAAQ,GAAA,CAAI,kBAAA;AAC7B,EAAA,GAAA,CAAI,SAAA,GAAY,QAAA,CAAS,IAAA,CAAK,CAAA,EAAG;AAC/B,IAAA,OAAO,wBAAA,QAAK,EAAU,MAAM,CAAA;AAAA,EAC9B;AACA,EAAA,OAAO,wBAAA,yBAAK,CAAQ,EAAG,QAAA,EAAU,QAAA,EAAU,QAAA,EAAU,MAAM,CAAA;AAC7D;AAEO,SAAS,mBAAA,CAAA,EAA8B;AAC5C,EAAA,OAAO,YAAA,CAAa,UAAU,CAAA;AAChC;AAaA,MAAA,SAAsB,gBAAA,CAAiB,GAAA,EAAa,KAAA,EAAyD;AAC3G,EAAA,MAAM,OAAA,EAAwB,CAAC,CAAA;AAC/B,EAAA,IAAA,CAAA,MAAW,KAAA,GAAQ,KAAA,EAAO;AACxB,IAAA,MAAM,UAAA,EAAY,wBAAA,GAAK,EAAK,IAAA,CAAK,QAAQ,CAAA;AACzC,IAAA,IAAI,OAAA,EAAS,KAAA;AACb,IAAA,IAAI;AACF,MAAA,MAAM,EAAA,EAAI,MAAM,4BAAA,SAAc,CAAA;AAC9B,MAAA,OAAA,EAAS,CAAA,CAAE,MAAA,CAAO,EAAA,GAAK,CAAA,CAAE,KAAA,EAAO,CAAA;AAAA,IAClC,EAAA,WAAQ;AACN,MAAA,OAAA,EAAS,KAAA;AAAA,IACX;AACA,IAAA,GAAA,CAAI,CAAC,MAAA,EAAQ;AACX,MAAA,MAAA,CAAO,IAAA,CAAK,EAAE,IAAA,EAAM,SAAA,EAAW,MAAA,EAAQ,KAAA,EAAO,QAAA,EAAU,MAAM,CAAC,CAAA;AAC/D,MAAA,QAAA;AAAA,IACF;AACA,IAAA,IAAI;AACF,MAAA,MAAM,OAAA,EAAS,MAAM,YAAA,CAAa,SAAS,CAAA;AAC3C,MAAA,GAAA,CAAI,OAAA,IAAW,IAAA,CAAK,MAAA,EAAQ;AAC1B,QAAA,MAAA,CAAO,IAAA,CAAK,EAAE,IAAA,EAAM,SAAA,EAAW,MAAA,EAAQ,IAAA,EAAM,QAAA,EAAU,KAAK,CAAC,CAAA;AAAA,MAC/D,EAAA,KAAO;AACL,QAAA,MAAA,CAAO,IAAA,CAAK;AAAA,UACV,IAAA;AAAA,UACA,SAAA;AAAA,UACA,MAAA,EAAQ,IAAA;AAAA,UACR,QAAA,EAAU,KAAA;AAAA,UACV,aAAA,EAAe,CAAA,0BAAA,EAA6B,IAAA,CAAK,MAAM,CAAA,MAAA,EAAS,MAAM,CAAA;AAAA,QAAA;AACvE,MAAA;AACH,IAAA;AAEA,MAAA;AAAY,QAAA;AACV,QAAA;AACA,QAAA;AACQ,QAAA;AACE,QAAA;AACgD,MAAA;AAC3D,IAAA;AACH,EAAA;AAEF,EAAA;AACF;AAGA;AACE,EAAA;AACF;AAoBA;AACE,EAAA;AAEA,EAAA;AACE,IAAA;AAEA,IAAA;AACE,sBAAA;AAAa,QAAA;AACX,QAAA;AACY,QAAA;AACL,QAAA;AACA,QAAA;AACE,MAAA;AAEX,MAAA;AAAA,IAAA;AAIF,IAAA;AACE,MAAA;AAAsB,IAAA;AAChB,IAAA;AAIR,IAAA;AAAgD,EAAA;AAEpD;AAGA;AACE,EAAA;AACF;AAGA;AACE,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACE,oBAAA;AACA,IAAA;AAAA,EAAA;AAEF,EAAA;AACE,IAAA;AAAsB,EAAA;AAChB,EAAA;AACR,EAAA;AACF;AAEA;AACE,EAAA;AACE,IAAA;AACA,IAAA;AAAwC,EAAA;AAExC,IAAA;AAAO,EAAA;AAET,EAAA;AACE,IAAA;AACA,IAAA;AAAkB,EAAA;AAElB,IAAA;AAAO,EAAA;AAEX;AAEA;AAME,EAAA;AACA,EAAA;AAEA,EAAA;AAAmC,IAAA;AACxB;AAAA,MAAA;AAEO,IAAA;AAChB,EAAA;AAEF,EAAA;AACE,IAAA;AAAU,MAAA;AACiE,IAAA;AAC3E,EAAA;AAGF,EAAA;AACA,EAAA;AACA,EAAA;AAEA,EAAA;AACA,EAAA;AACE,IAAA;AACA,IAAA;AACE,MAAA;AACA,sBAAA;AAAa,QAAA;AACX,QAAA;AACA,QAAA;AACA,QAAA;AACO,MAAA;AACR,IAAA;AAEH,IAAA;AAAyB,EAAA;AAEzB,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAA+D,EAAA;AAGjE,kBAAA;AAAa,IAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,EAAA;AAIT,EAAA;AACA,EAAA;AACE,IAAA;AAAoC,EAAA;AAEpC,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAAiE,EAAA;AAGnE,EAAA;AACE,IAAA;AACE,MAAA;AAAqB,IAAA;AACf,IAAA;AACR,IAAA;AAAU,MAAA;AAC4D,IAAA;AACtE,EAAA;AAGF,EAAA;AACA,kBAAA;AAAa,IAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,EAAA;AAEX;AAEA;AACE,EAAA;AACA,EAAA;AACA,EAAA;AACE,IAAA;AACE,MAAA;AAAc,IAAA;AAEhB,EAAA;AAEF,EAAA;AACF;ADrFA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/Users/mong-e/workspace/kordoc/dist/chunk-H7OWEYH2.cjs","sourcesContent":[null,"/**\n * 수식 OCR 모델 자동 다운로드 + SHA-256 검증\n *\n * 캐시 위치: `~/.cache/kordoc/models/pix2text/`\n * 총 ~155MB (MFD 44MB + MFR encoder 87MB + MFR decoder 30MB + tokenizer 40KB)\n *\n * HF URL + SHA 는 Docufinder archive/rust-formula-ocr-wip 에서 검증 완료됨.\n * 모델 버전 갱신 시 URL + SHA 를 함께 갱신해야 한다.\n */\n\nimport { createHash } from \"crypto\"\nimport { createReadStream } from \"fs\"\nimport { mkdir, stat, unlink, rename } from \"fs/promises\"\nimport { createWriteStream } from \"fs\"\nimport { homedir } from \"os\"\nimport { join, dirname } from \"path\"\nimport { pipeline } from \"stream/promises\"\nimport { Readable } from \"stream\"\n\nexport interface ModelSpec {\n  name: string\n  filename: string\n  url: string\n  sha256: string\n  sizeMb: number\n}\n\n/** 모델 스펙 (SHA-256 은 실제 다운로드로 검증됨 — 변경 금지) */\nexport const MFD_MODEL: ModelSpec = {\n  name: \"Pix2Text MFD\",\n  filename: \"mfd.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfd/resolve/main/mfd-v20240618.onnx\",\n  sha256: \"51a8854743b17ae654729af8db82a630c1ccfa06debf4856c8b28055f87d02c1\",\n  sizeMb: 42,\n}\n\nexport const MFR_ENCODER_MODEL: ModelSpec = {\n  name: \"Pix2Text MFR encoder\",\n  filename: \"encoder_model.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/encoder_model.onnx\",\n  sha256: \"bd8d5c322792e9ec45793af5569e9748f82a3d728a9e00213dbfc56c1486f37d\",\n  sizeMb: 87,\n}\n\nexport const MFR_DECODER_MODEL: ModelSpec = {\n  name: \"Pix2Text MFR decoder\",\n  filename: \"decoder_model.onnx\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/decoder_model.onnx\",\n  sha256: \"fd0f92d7a012f3dae41e1ac79421aea0ea888b5a66cb3f9a004e424f82f3daed\",\n  sizeMb: 30,\n}\n\nexport const MFR_TOKENIZER: ModelSpec = {\n  name: \"Pix2Text MFR tokenizer\",\n  filename: \"tokenizer.json\",\n  url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/tokenizer.json\",\n  sha256: \"3e2ab757277d22639bec28c9d7972e352d3d1dba223051fa674002dc5ab64df3\",\n  sizeMb: 1,\n}\n\nexport const ALL_FORMULA_MODELS: ReadonlyArray<ModelSpec> = [\n  MFD_MODEL,\n  MFR_ENCODER_MODEL,\n  MFR_DECODER_MODEL,\n  MFR_TOKENIZER,\n]\n\n/**\n * 모델 캐시 하위 디렉토리 경로. 환경 변수 `KORDOC_MODEL_CACHE` 가 설정되면\n * 해당 경로를 우선 사용, 없으면 `~/.cache/kordoc/models/<subdir>/`.\n * (수식 OCR 외 텍스트 OCR 등 다른 모델군도 공용 — subdir 로 격리)\n */\nexport function getModelsDir(subdir: string): string {\n  const override = process.env.KORDOC_MODEL_CACHE\n  if (override && override.trim()) {\n    return join(override, subdir)\n  }\n  return join(homedir(), \".cache\", \"kordoc\", \"models\", subdir)\n}\n\nexport function getFormulaModelsDir(): string {\n  return getModelsDir(\"pix2text\")\n}\n\nexport interface ModelStatus {\n  spec: ModelSpec\n  localPath: string\n  exists: boolean\n  /** SHA 검증까지 끝났는가 (exists && valid) */\n  verified: boolean\n  /** 검증 실패 시 이유 */\n  invalidReason?: string\n}\n\n/** 지정 모델군의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getModelStatusIn(dir: string, specs: ReadonlyArray<ModelSpec>): Promise<ModelStatus[]> {\n  const result: ModelStatus[] = []\n  for (const spec of specs) {\n    const localPath = join(dir, spec.filename)\n    let exists = false\n    try {\n      const s = await stat(localPath)\n      exists = s.isFile() && s.size > 0\n    } catch {\n      exists = false\n    }\n    if (!exists) {\n      result.push({ spec, localPath, exists: false, verified: false })\n      continue\n    }\n    try {\n      const actual = await sha256OfFile(localPath)\n      if (actual === spec.sha256) {\n        result.push({ spec, localPath, exists: true, verified: true })\n      } else {\n        result.push({\n          spec,\n          localPath,\n          exists: true,\n          verified: false,\n          invalidReason: `SHA256 mismatch: expected ${spec.sha256}, got ${actual}`,\n        })\n      }\n    } catch (e) {\n      result.push({\n        spec,\n        localPath,\n        exists: true,\n        verified: false,\n        invalidReason: `SHA compute failed: ${(e as Error).message}`,\n      })\n    }\n  }\n  return result\n}\n\n/** 모든 수식 모델의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getFormulaModelStatus(): Promise<ModelStatus[]> {\n  return getModelStatusIn(getFormulaModelsDir(), ALL_FORMULA_MODELS)\n}\n\nexport interface DownloadProgress {\n  spec: ModelSpec\n  /** 현재까지 다운로드한 바이트 */\n  downloaded: number\n  /** 전체 바이트 (알 수 있으면) */\n  total: number | null\n  /** \"download\" | \"verify\" | \"done\" | \"skip\" */\n  phase: \"download\" | \"verify\" | \"done\" | \"skip\" | \"error\"\n  message?: string\n}\n\nexport type ProgressHandler = (p: DownloadProgress) => void\n\n/**\n * 지정 모델군을 다운로드/검증한다.\n * - 이미 있고 SHA 일치 → skip\n * - 없음 → 다운로드 후 SHA 검증. 실패 시 파일 삭제.\n */\nexport async function ensureModelsIn(dir: string, specs: ReadonlyArray<ModelSpec>, onProgress?: ProgressHandler): Promise<void> {\n  await mkdir(dir, { recursive: true })\n\n  for (const spec of specs) {\n    const localPath = join(dir, spec.filename)\n\n    if (await isExistingValid(localPath, spec.sha256)) {\n      onProgress?.({\n        spec,\n        downloaded: 0,\n        total: null,\n        phase: \"skip\",\n        message: \"이미 존재 + SHA 일치\",\n      })\n      continue\n    }\n\n    // 기존 파일 있지만 SHA 불일치 → 삭제\n    try {\n      await unlink(localPath)\n    } catch {\n      // 없을 수 있음\n    }\n\n    await downloadToFile(spec, localPath, onProgress)\n  }\n}\n\n/** 필요한 모든 수식 모델을 다운로드/검증한다. */\nexport async function ensureFormulaModels(onProgress?: ProgressHandler): Promise<void> {\n  return ensureModelsIn(getFormulaModelsDir(), ALL_FORMULA_MODELS, onProgress)\n}\n\n/** 단일 모델만 확인/다운로드 (진단 UI 에서 개별 상태 새로고침용) */\nexport async function ensureSingleModel(spec: ModelSpec, onProgress?: ProgressHandler): Promise<void> {\n  const dir = getFormulaModelsDir()\n  await mkdir(dir, { recursive: true })\n  const localPath = join(dir, spec.filename)\n  if (await isExistingValid(localPath, spec.sha256)) {\n    onProgress?.({ spec, downloaded: 0, total: null, phase: \"skip\" })\n    return\n  }\n  try {\n    await unlink(localPath)\n  } catch {}\n  await downloadToFile(spec, localPath, onProgress)\n}\n\nasync function isExistingValid(localPath: string, sha256Expected: string): Promise<boolean> {\n  try {\n    const s = await stat(localPath)\n    if (!s.isFile() || s.size === 0) return false\n  } catch {\n    return false\n  }\n  try {\n    const actual = await sha256OfFile(localPath)\n    return actual === sha256Expected\n  } catch {\n    return false\n  }\n}\n\nasync function downloadToFile(\n  spec: ModelSpec,\n  localPath: string,\n  onProgress?: ProgressHandler,\n): Promise<void> {\n  // 먼저 .part 로 받고 검증 후 rename — 중단된 다운로드가 \"정상 파일\"로 오인되는 걸 방지\n  const partPath = `${localPath}.part`\n  await mkdir(dirname(localPath), { recursive: true })\n\n  const resp = await fetch(spec.url, {\n    headers: {\n      // HF CDN 은 UA 없으면 가끔 403 을 뱉는다\n      \"User-Agent\": \"kordoc-formula-ocr/1.0 (+https://github.com/chrisryugj/kordoc)\",\n    },\n  })\n  if (!resp.ok || !resp.body) {\n    throw new Error(\n      `${spec.name} 다운로드 실패: HTTP ${resp.status} ${resp.statusText} (${spec.url})`,\n    )\n  }\n\n  const lenHeader = resp.headers.get(\"content-length\")\n  const total = lenHeader ? Number.parseInt(lenHeader, 10) : null\n  let downloaded = 0\n\n  const ws = createWriteStream(partPath)\n  try {\n    const reader = Readable.fromWeb(resp.body as unknown as import(\"stream/web\").ReadableStream)\n    reader.on(\"data\", (chunk: Buffer | Uint8Array) => {\n      downloaded += chunk.length\n      onProgress?.({\n        spec,\n        downloaded,\n        total,\n        phase: \"download\",\n      })\n    })\n    await pipeline(reader, ws)\n  } catch (e) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(`${spec.name} 스트리밍 실패: ${(e as Error).message}`)\n  }\n\n  onProgress?.({\n    spec,\n    downloaded,\n    total,\n    phase: \"verify\",\n  })\n\n  // SHA 검증\n  let actual: string\n  try {\n    actual = await sha256OfFile(partPath)\n  } catch (e) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(`${spec.name} SHA 계산 실패: ${(e as Error).message}`)\n  }\n\n  if (actual !== spec.sha256) {\n    try {\n      await unlink(partPath)\n    } catch {}\n    throw new Error(\n      `${spec.name} SHA256 mismatch: expected ${spec.sha256}, got ${actual} — 모델 URL 이 오염되었거나 전송 중 손상되었습니다.`,\n    )\n  }\n\n  await rename(partPath, localPath)\n  onProgress?.({\n    spec,\n    downloaded,\n    total,\n    phase: \"done\",\n  })\n}\n\nasync function sha256OfFile(p: string): Promise<string> {\n  const h = createHash(\"sha256\")\n  const stream = createReadStream(p)\n  await pipeline(stream, async function* (src) {\n    for await (const chunk of src) {\n      h.update(chunk)\n      // yield 안 함 (소비만)\n    }\n  })\n  return h.digest(\"hex\")\n}\n"]}

package/dist/{chunk-DSL6EVFJ.js → chunk-I53FOXZR.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   sanitizeHref
-} from "./chunk-2L7MIKOM.js";
+} from "./chunk-E2K4SNMD.js";
 
 // src/shared/pua.ts
 var BMP_SYMBOL_MAP = {
@@ -613,4 +613,4 @@ export {
   dedupeRunningHeaders,
   blocksToMarkdown
 };
-//# sourceMappingURL=chunk-DSL6EVFJ.js.map
+//# sourceMappingURL=chunk-I53FOXZR.js.map

package/dist/{chunk-GHMVNMUN.js → chunk-LWNFAIYG.js}

@@ -7,11 +7,11 @@ import {
   createXmlParser,
   findChildByLocalName,
   parseRenderStyles
-} from "./chunk-FHYLLKIP.js";
+} from "./chunk-H6O65EPK.js";
 import {
   KordocError,
   precheckZipSize
-} from "./chunk-2L7MIKOM.js";
+} from "./chunk-E2K4SNMD.js";
 
 // src/render/svg-render.ts
 import JSZip from "jszip";
@@ -1281,4 +1281,4 @@ export {
   fitRatioForFewerLines,
   renderHwpxToSvg
 };
-//# sourceMappingURL=chunk-GHMVNMUN.js.map
+//# sourceMappingURL=chunk-LWNFAIYG.js.map

package/dist/{chunk-EFF2JPPA.cjs → chunk-LXUZUUUG.cjs}

@@ -4,7 +4,7 @@
 
 
 
-var _chunkKKZATKQWcjs = require('./chunk-KKZATKQW.cjs');
+var _chunkBLNOGMIMcjs = require('./chunk-BLNOGMIM.cjs');
 
 // src/pdf/image-regions.ts
 var _pdfmjs = require('pdfjs-dist/legacy/build/pdf.mjs');
@@ -1536,9 +1536,9 @@ function detectHeadings(blocks, medianFontSize) {
     if (/^\d+$/.test(text)) continue;
     const ratio = block.style.fontSize / medianFontSize;
     let level = 0;
-    if (ratio >= _chunkKKZATKQWcjs.HEADING_RATIO_H1) level = 1;
-    else if (ratio >= _chunkKKZATKQWcjs.HEADING_RATIO_H2) level = 2;
-    else if (ratio >= _chunkKKZATKQWcjs.HEADING_RATIO_H3) level = 3;
+    if (ratio >= _chunkBLNOGMIMcjs.HEADING_RATIO_H1) level = 1;
+    else if (ratio >= _chunkBLNOGMIMcjs.HEADING_RATIO_H2) level = 2;
+    else if (ratio >= _chunkBLNOGMIMcjs.HEADING_RATIO_H3) level = 3;
     if (level > 0) {
       block.type = "heading";
       block.level = level;
@@ -2658,14 +2658,14 @@ function isProseSpread(items) {
   for (let i = 1; i < sorted.length; i++) {
     gaps.push(sorted[i].x - (sorted[i - 1].x + sorted[i - 1].w));
   }
-  const maxGap = _chunkKKZATKQWcjs.safeMax.call(void 0, gaps);
+  const maxGap = _chunkBLNOGMIMcjs.safeMax.call(void 0, gaps);
   const avgLen = items.reduce((s, i) => s + i.text.length, 0) / items.length;
   return maxGap < 40 && avgLen < 5;
 }
 function detectColumns(yLines) {
   const allItems = yLines.flat();
   if (allItems.length === 0) return null;
-  const pageWidth = _chunkKKZATKQWcjs.safeMax.call(void 0, allItems.map((i) => i.x + i.w)) - _chunkKKZATKQWcjs.safeMin.call(void 0, allItems.map((i) => i.x));
+  const pageWidth = _chunkBLNOGMIMcjs.safeMax.call(void 0, allItems.map((i) => i.x + i.w)) - _chunkBLNOGMIMcjs.safeMin.call(void 0, allItems.map((i) => i.x));
   if (pageWidth < 100) return null;
   let bigoLineIdx = -1;
   for (let i = 0; i < yLines.length; i++) {
@@ -3113,7 +3113,7 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
     let finalTextBlocks = [];
     if (remaining.length > 0) {
       const allY = remaining.map((i) => i.y);
-      const pageH = _chunkKKZATKQWcjs.safeMax.call(void 0, allY) - _chunkKKZATKQWcjs.safeMin.call(void 0, allY);
+      const pageH = _chunkBLNOGMIMcjs.safeMax.call(void 0, allY) - _chunkBLNOGMIMcjs.safeMin.call(void 0, allY);
       const groups = xyCutOrder(remaining, Math.max(15, pageH * 0.03));
       const textBlocks = [];
       for (const group of groups) {
@@ -3301,7 +3301,7 @@ function extractPageBlocksFallback(items, pageNum, fullPage = false) {
       blocks.push({ type: "paragraph", text: tableText, pageNumber: pageNum, bbox, style: dominantStyle(items) });
     } else {
       const allY = items.map((i) => i.y);
-      const pageHeight = _chunkKKZATKQWcjs.safeMax.call(void 0, allY) - _chunkKKZATKQWcjs.safeMin.call(void 0, allY);
+      const pageHeight = _chunkBLNOGMIMcjs.safeMax.call(void 0, allY) - _chunkBLNOGMIMcjs.safeMin.call(void 0, allY);
       const gapThreshold = Math.max(15, pageHeight * 0.03);
       const orderedGroups = proseCutX !== null ? splitTwoColumnProse(items, proseCutX) : xyCutOrder(items, gapThreshold);
       for (const group of orderedGroups) {
@@ -3340,4 +3340,4 @@ function extractPageBlocksFallback(items, pageNum, fullPage = false) {
 
 
 exports.extractImageRegions = extractImageRegions; exports.filterHiddenText = filterHiddenText; exports.collapseEvenSpacing = collapseEvenSpacing; exports.normalizeItems = normalizeItems; exports.computeMedianFontSizeFromFreq = computeMedianFontSizeFromFreq; exports.detectHeadings = detectHeadings; exports.detectMarkerHeadings = detectMarkerHeadings; exports.detectTableCaptions = detectTableCaptions; exports.detectKoreanListBlocks = detectKoreanListBlocks; exports.removeHeaderFooterBlocks = removeHeaderFooterBlocks; exports.extractPageBlocksWithLines = extractPageBlocksWithLines; exports.mergeCrossPageTables = mergeCrossPageTables;
-//# sourceMappingURL=chunk-EFF2JPPA.cjs.map
+//# sourceMappingURL=chunk-LXUZUUUG.cjs.map