kordoc 4.12.0 → 4.12.1

package/README.md

@@ -76,7 +76,7 @@ MCP 등록 대신 스킬(SKILL.md) 형태로 쓰려면:
 *   **📊 복잡한 표(Table) 복원 — 숫자로 잠금**: 병합·중첩 표를 구조 그대로 살립니다. HWPX 코퍼스 291문서 **표 1,673/1,673·셀 27,714/27,714 무손실**(독립 추출기 대조 게이트 1.0), 동일문서 hwpx↔pdf 대조(6쌍 69표)는 **매칭 98.6%·완전일치 65.2%**. 선 없는 PDF 표도 감지해 복원하고, 법령 개정안 PDF의 신구조문대비표도 통째로 살립니다 (v3.16.2).
 *   **🔍 신구대조표 자동 생성**: 두 문서의 차이점을 분석하여 무엇이 바뀌었는지 한눈에 보여줍니다. (HWP와 HWPX 간의 비교도 가능!)
 *   **📝 마크다운을 다시 HWPX로**: AI가 작성한 내용을 다시 보고서 양식(`HWPX`)으로 되돌려줍니다. 이제 복사-붙여넣기 노가다에서 해방되세요.
-*   **🏛️ 정부 표준 공문서 생성 (v4.0)**: 실제 정부 양식 16종 + 실결재 기안문 60건을 전수 디코드·대조해 만든 공문서 엔진. 개조식 보고서(표지·목차 배너·로마숫자 장헤더·쪽번호·결재란), 기안문(별지 제1호서식 두문·결문, "끝." 자동), 공고문·보도자료 프리셋, 항목부호 8단계(1. 가. 1) 가)…) 자동, 공문서 표기법 검수 13룰(`kordoc lint`)까지 — 한글 COM 실렌더로 조판까지 실측 검증했습니다.
+*   **🏛️ 정부 표준 공문서 생성 (v4.0)**: 실제 정부 양식 16종 + 실결재 기안문 60건을 전수 디코드·대조해 만든 공문서 엔진. 개조식 보고서(표지·목차 배너·로마숫자 장헤더·쪽번호·결재란), 기안문(별지 제1호서식 두문·결문, "끝." 자동), 공고문·보도자료 프리셋, 항목부호 8단계(1. 가. 1) 가)…) 자동, 공문서 표기법 검수 19룰(`kordoc lint`)까지 — 한글 COM 실렌더로 조판까지 실측 검증했습니다.
 *   **🔄 서식 보존 무손실 라운드트립 (v3.0)**: 변환된 마크다운을 편집해서 `patchHwpx`(HWPX) / `patchHwp`(HWP 5.x 바이너리)에 넘기면, **원본 서식을 1바이트도 건드리지 않고** 바뀐 문단/표 셀의 텍스트만 원본 안에서 교체합니다. v3.7부터는 **표에 행을 추가/삭제하는 편집**도 원본 서식을 승계하며 반영되고, v3.8부터는 HWP 5.x의 **빈 셀에 값 넣기**도 지원합니다.
 *   **🖼️ 레이아웃 보존 렌더 (v3.10~3.15)**: 한컴이 저장한 조판 캐시 좌표로 원본 레이아웃을 SVG로 재현하고, 캐시가 없는 파일(AI가 만든 HWPX·편집본)은 **순수 TS reflow 엔진**이 직접 조판합니다. 다페이지·표·그리기 도형·검색어 형광펜까지. 서버에 한컴 없이 HWPX 미리보기를 만들 수 있습니다.
 *   **📊 차트 생성 (v3.16)**: 마크다운의 ```chart 펜스(type/cat/계열 라인)가 한컴 네이티브 차트(OOXML chartSpace)로 생성됩니다 — 막대·선·원·도넛·영역·분산·방사형 등 20종, 계열/조각 색 지정 가능.
@@ -89,6 +89,16 @@ MCP 등록 대신 스킬(SKILL.md) 형태로 쓰려면:
 
 ---
 
+## v4.12.1 변경사항
+
+법령 **별지서식**(법제처 별표·서식) 파싱을 HWP·HWPX·PDF 세 포맷에서 맞췄습니다. 법제처 OpenAPI로 서식 300건의 HWP 원본과 PDF를 쌍으로 모으고 rhwp 변환 HWPX를 붙인 900파일 코퍼스를 게이트에 편입해, HWP↔PDF 셀 텍스트 일치 0.79 → 0.92, PDF 표 GT cellExact 0.73 → 0.96 을 실측했습니다.
+
+- **📐 PDF 셀 클립 그리드**: 한컴 PDF는 표 셀마다 클립 사각형을 깔고 글을 찍습니다. 별지서식 외곽 표는 테두리 "없음" 셀이라 괘선만 보는 종전 경로가 표를 놓치고 제목·기입란이 헤딩과 문단으로 흩어졌는데, 이제 클립 사각형 묶음을 **셀 기하가 확정된 표**로 바로 씁니다. 도선사면허 신청서가 HWP 파서와 같은 23열·같은 병합 구조로 나옵니다. 클립의 포함 관계로 층을 나눠, 지정서·증서의 1칸 테두리 틀은 제목행·꼬리행과 한 표의 셀이 되고 안쪽 "발신명의 | 직인" 표는 별도 표로 살립니다.
+- **➔ 심볼 폰트 글리프 복원**: 처리절차 화살표·체크박스를 Wingdings로 찍은 PDF에서 `è` 같은 깨진 라틴 문자가 남던 것을 유니코드(➔ □ ✔ ☑)로 되돌립니다.
+- **📝 표기법 검수 6룰 보강**: 금액 괄호 한글 병기(`MONEY_NO_HANGUL`), 물결표 앞뒤 붙여쓰기(`TILDE_SPACE`), 두음법칙 '년도→연도'(`DUEUM_ERROR`), 외래어 오기 40종(`LOANWORD_ERROR`, 표준 표기 제안), 차별·비하 표현 25종(`DISCRIMINATORY_TERM`, 순화어 제안), 붙임 있는데 "끝." 없음(`END_MARK_MISSING`, `kordoc lint` 전용). 시중 HWPX 공문 도구(pyhwpxlib·hwpx-skill)의 검사 항목과 대조해 비어 있던 축을 채웠습니다.
+- **🏛️ 내부결재 문서의 발신명의 생략**: 수신이 "내부결재"인 기안문은 결문에 발신명의를 적지 않습니다(행정업무규정 시행규칙).
+- **🩹 기입 빈칸 보존**: "년   월   일"·"시   분"이 "년월일"로 붙던 것, 청구서류 1열 틀 표의 기입 항목이 한 줄로 뭉개지던 것을 고쳤습니다.
+
 ## v4.12.0 변경사항
 
 v4.11.0 이 걷어낸 HWP3 유실의 **잔여 전량**을 처리했습니다. rhwp `samples/` 의 HWP3 원본 9종을 한컴 변환본과 문자 멀티셋으로 대조하면, 이제 변환본에만 있는 문자는 그림 대체텍스트를 빼면 **0** 입니다.
@@ -293,7 +303,7 @@ rhwp 최신 devel(v0.8.3~0.8.4)을 훑어 kordoc 의 HWP3 디코더에 반영했
 ## v4.0.1 변경사항
 
 - **✍️ 실무자 QA 3건 수정**: bold 시 폰트가 HY견고딕/Arial Black으로 바뀌던 "정체모를 폰트" 제거, h2 섹션 제목 말머리 `--h2-marker`(box □ / number / none), 개조식 소분류 부호 ― → 실무 관행 하이픈 `-`.
-- **📝 공문서 표기법 검수 13룰**: 날짜·시간·금액·붙임 표기 등을 검수하는 `kordoc lint <file>` + 공문서 생성 시 경고 병기.
+- **📝 공문서 표기법 검수 19룰**: 날짜·시간·금액·붙임·외래어·차별 표현 등을 검수하는 `kordoc lint <file>` + 공문서 생성 시 경고 병기.
 
 ## v4.0.0 변경사항
 
@@ -892,7 +902,7 @@ npx kordoc generate 보고서.md -o 보고서.hwpx --preset 보고서         #
 npx kordoc patch 원본.hwpx 편집.md -o 반영.hwpx      # 서식 보존 라운드트립 패치 (.hwp도 자동 분기)
 npx kordoc seal 신청서.hwpx --image 도장.png --anchor "(인)" -o 날인.hwpx  # 도장/서명 날인
 npx kordoc validate 산출물.hwpx                      # HWPX 구조 검증 (ZIP·필수 파트·XML)
-npx kordoc lint 보고서.md                            # 공문서 표기법 검수 13룰 — 입력은 md/txt('-'=stdin), error 있으면 exit 1
+npx kordoc lint 보고서.md                            # 공문서 표기법 검수 19룰 — 입력은 md/txt('-'=stdin), error 있으면 exit 1
 npx kordoc redact 민원서류.hwpx -o 마스킹.hwpx       # 개인정보 탐지 + 서식 보존 마스킹 (v4.1)
 npx kordoc redact 민원서류.hwpx --mask-char '*' -o 마스킹.hwpx   # 마스크 문자 지정 (기본 ●)
 npx kordoc profile 기관서식.hwpx                     # 표 서식 프로필(JSON) 추출 → generate --profile 로 재현
@@ -910,7 +920,7 @@ npx kordoc watch ./문서 --webhook https://api/hook  # 웹훅 알림
 >
 > `kordoc lint` 는 **텍스트(마크다운/txt)를 검사합니다** — HWPX 를 바로 넘기면 안 됩니다.
 > 생성물을 검수하려면 원고 마크다운을 넘기거나, `kordoc 문서.hwpx | kordoc lint -` 로 파이프하세요.
-> (공문서 생성 시에는 `generate` 가 같은 13룰 경고를 함께 냅니다.)
+> (공문서 생성 시에는 `generate` 가 같은 룰의 경고를 함께 냅니다. `END_MARK_MISSING` 은 완성 원고를 보는 `lint` 에서만 켜집니다.)
 
 ### 실패 계약 — 기계 판독 가능한 실패 JSON (v4.10.0+, #69)
 
@@ -1056,7 +1066,7 @@ codex mcp add kordoc -- npx -y kordoc mcp
 | `renderHwpxToSvg(buffer, options?)` | HWPX → 레이아웃 보존 SVG — 다페이지·형광펜·도형, 캐시 없으면 `reflow` (v3.10~15) |
 | `placeSealHwpx(buffer, seals)` | 도장/서명 이미지를 앵커 문구 위에 부유 배치 (v3.16) |
 | `validateHwpx(buffer)` | HWPX 구조 검증 — ZIP·mimetype·필수 파트·XML 웰폼드 (v3.16) |
-| `lintGongmunText(text)` | 공문서 표기법 검수 13룰 — 텍스트/마크다운 입력 (v4.0.1) |
+| `lintGongmunText(text, { document? })` | 공문서 표기법 검수 19룰 + AI 슬롭 2룰 — 텍스트/마크다운 입력 (v4.0.1, v4.12.1 보강). `document: true` 면 붙임/"끝." 문서 단위 검사 포함 |
 | `redactMarkdown(text, options?)` / `redactText(...)` | 개인정보 탐지 + 마스킹 (기본 룰: 주민번호·전화·이메일·카드·계좌, v4.1) |
 | `blocksToChunks(blocks, options?)` | RAG용 구조 청크 — 헤딩·개조식 위계 breadcrumb + 표 독립 청크 (v4.1) |
 | `blocksToMarkdown(blocks)` | IRBlock[] → Markdown 문자열 |

package/dist/{-B3EUUBPV.js → -BUYIUNW3.js}

@@ -56,7 +56,7 @@ import {
   unknownFontWarnings,
   usesGaejosikMunche,
   validateHwpx
-} from "./chunk-PFUW4I4I.js";
+} from "./chunk-VO37BVAH.js";
 import "./chunk-7FCNOKOR.js";
 import "./chunk-N3DLQ2YH.js";
 import "./chunk-MOL7MDBG.js";
@@ -66,7 +66,7 @@ import {
   buildRangeSplices,
   placeSealHwpx,
   scanSectionXml
-} from "./chunk-TF5DS74F.js";
+} from "./chunk-JD4H6Y33.js";
 import {
   DEFAULT_REDACT_RULES,
   redactMarkdown,
@@ -74,10 +74,10 @@ import {
 } from "./chunk-DZIXKL7E.js";
 import {
   blocksToChunks
-} from "./chunk-YASFVKDT.js";
+} from "./chunk-ST3JBCIL.js";
 import {
   blocksToMarkdown
-} from "./chunk-GHVXLV52.js";
+} from "./chunk-JQBF4PSF.js";
 import {
   SPACE_EM_FIXED,
   SPACE_EM_FONT,
@@ -87,11 +87,11 @@ import {
   renderHwpxToSvg,
   simulateWrap,
   simulateWrapKeepWord
-} from "./chunk-46UDI56N.js";
-import "./chunk-I2OETEEL.js";
+} from "./chunk-F2KL5Z5T.js";
+import "./chunk-EWSH6O6S.js";
 import {
   VERSION
-} from "./chunk-4PKNWBXJ.js";
+} from "./chunk-E37VAV46.js";
 export {
   BUILTIN_TEMPLATES,
   DEFAULT_REDACT_RULES,
@@ -169,4 +169,4 @@ export {
   usesGaejosikMunche,
   validateHwpx
 };
-//# sourceMappingURL=-B3EUUBPV.js.map
+//# sourceMappingURL=-BUYIUNW3.js.map

package/dist/{chunk-BQSWSMRG.cjs → chunk-6MF7JRG3.cjs}

@@ -1,5 +1,5 @@
 "use strict";Object.defineProperty(exports, "__esModule", {value: true}); function _nullishCoalesce(lhs, rhsFn) { if (lhs != null) { return lhs; } else { return rhsFn(); } } function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; }// src/utils.ts
-var VERSION = true ? "4.12.0" : "0.0.0-dev";
+var VERSION = true ? "4.12.1" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -121,4 +121,4 @@ function classifyError(err) {
 
 
 exports.VERSION = VERSION; exports.toArrayBuffer = toArrayBuffer; exports.KordocError = KordocError; exports.sanitizeError = sanitizeError; exports.isPathTraversal = isPathTraversal; exports.normalizeSectionHref = normalizeSectionHref; exports.compareSectionPaths = compareSectionPaths; exports.precheckZipSize = precheckZipSize; exports.stripDtd = stripDtd; exports.sanitizeHref = sanitizeHref; exports.safeMin = safeMin; exports.safeMax = safeMax; exports.classifyError = classifyError;
-//# sourceMappingURL=chunk-BQSWSMRG.cjs.map
+//# sourceMappingURL=chunk-6MF7JRG3.cjs.map

package/dist/chunk-6MF7JRG3.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/chris_gomdori/workspace/kordoc/dist/chunk-6MF7JRG3.cjs","../src/utils.ts"],"names":[],"mappings":"AAAA;ACIO,IAAM,QAAA,EAAkB,KAAA,EAA4C,SAAA,EAAqB,WAAA;AAOzF,SAAS,aAAA,CAAc,GAAA,EAA0B;AACtD,EAAA,GAAA,CAAI,GAAA,CAAI,WAAA,IAAe,EAAA,GAAK,GAAA,CAAI,WAAA,IAAe,GAAA,CAAI,MAAA,CAAO,UAAA,EAAY;AACpE,IAAA,OAAO,GAAA,CAAI,MAAA;AAAA,EACb;AACA,EAAA,OAAO,GAAA,CAAI,MAAA,CAAO,KAAA,CAAM,GAAA,CAAI,UAAA,EAAY,GAAA,CAAI,WAAA,EAAa,GAAA,CAAI,UAAU,CAAA;AACzE;AAMO,IAAM,YAAA,EAAN,MAAA,QAA0B,MAAM;AAAA,EACrC,WAAA,CAAY,OAAA,EAAiB;AAC3B,IAAA,KAAA,CAAM,OAAO,CAAA;AACb,IAAA,IAAA,CAAK,KAAA,EAAO,aAAA;AAAA,EACd;AACF,CAAA;AAMO,SAAS,aAAA,CAAc,GAAA,EAAsB;AAClD,EAAA,GAAA,CAAI,IAAA,WAAe,WAAA,EAAa,OAAO,GAAA,CAAI,OAAA;AAC3C,EAAA,OAAO,0FAAA;AACT;AAMO,SAAS,eAAA,CAAgB,IAAA,EAAuB;AACrD,EAAA,GAAA,CAAI,IAAA,CAAK,QAAA,CAAS,IAAM,CAAA,EAAG,OAAO,IAAA;AAClC,EAAA,MAAM,WAAA,EAAa,IAAA,CAAK,OAAA,CAAQ,KAAA,EAAO,GAAG,CAAA;AAC1C,EAAA,MAAM,SAAA,EAAW,UAAA,CAAW,KAAA,CAAM,GAAG,CAAA;AACrC,EAAA,OAAO,QAAA,CAAS,IAAA,CAAK,CAAA,CAAA,EAAA,GAAK,EAAA,IAAM,IAAI,EAAA,GAAK,UAAA,CAAW,UAAA,CAAW,GAAG,EAAA,GAAK,YAAA,CAAa,IAAA,CAAK,UAAU,CAAA;AACrG;AASO,SAAS,oBAAA,CAAqB,IAAA,EAA6B;AAChE,EAAA,GAAA,CAAI,CAAC,IAAA,EAAM,OAAO,IAAA;AAClB,EAAA,IAAI,WAAA,EAAa,IAAA,CAAK,OAAA,CAAQ,KAAA,EAAO,GAAG,CAAA,CAAE,OAAA,CAAQ,MAAA,EAAQ,EAAE,CAAA;AAC5D,EAAA,GAAA,CAAI,eAAA,CAAgB,UAAU,CAAA,EAAG,OAAO,IAAA;AACxC,EAAA,GAAA,CAAI,sBAAA,CAAuB,IAAA,CAAK,UAAU,CAAA,EAAG,WAAA,EAAa,YAAA,EAAc,UAAA;AACxE,EAAA,OAAO,6BAAA,CAA8B,IAAA,CAAK,UAAU,EAAA,EAAI,WAAA,EAAa,IAAA;AACvE;AAGO,SAAS,mBAAA,CAAoB,CAAA,EAAW,CAAA,EAAmB;AAChE,EAAA,MAAM,GAAA,EAAK,MAAA,kCAAO,CAAA,mBAAE,KAAA,mBAAM,uBAAuB,CAAA,4BAAA,CAAI,CAAC,GAAA,UAAK,MAAA,CAAO,kBAAgB,CAAA;AAClF,EAAA,MAAM,GAAA,EAAK,MAAA,kCAAO,CAAA,qBAAE,KAAA,mBAAM,uBAAuB,CAAA,4BAAA,CAAI,CAAC,GAAA,UAAK,MAAA,CAAO,kBAAgB,CAAA;AAClF,EAAA,OAAO,GAAA,IAAO,GAAA,EAAK,CAAA,CAAE,aAAA,CAAc,CAAC,EAAA,EAAI,GAAA,EAAK,EAAA;AAC/C;AAQO,SAAS,eAAA,CACd,MAAA,EACA,oBAAA,EAAsB,IAAA,EAAM,KAAA,EAAO,IAAA,EACnC,WAAA,EAAa,GAAA,EACsC;AACnD,EAAA,IAAI;AACF,IAAA,MAAM,KAAA,EAAO,IAAI,QAAA,CAAS,MAAM,CAAA;AAChC,IAAA,MAAM,IAAA,EAAM,MAAA,CAAO,UAAA;AAEnB,IAAA,IAAI,WAAA,EAAa,CAAA,CAAA;AACjB,IAAA,IAAA,CAAA,IAAS,EAAA,EAAI,IAAA,EAAM,EAAA,EAAI,EAAA,GAAK,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,IAAA,EAAM,KAAK,CAAA,EAAG,CAAA,EAAA,EAAK;AACzD,MAAA,GAAA,CAAI,IAAA,CAAK,SAAA,CAAU,CAAA,EAAG,IAAI,EAAA,IAAM,SAAA,EAAY;AAAE,QAAA,WAAA,EAAa,CAAA;AAAG,QAAA,KAAA;AAAA,MAAM;AAAA,IACtE;AACA,IAAA,GAAA,CAAI,WAAA,EAAa,CAAA,EAAG,OAAO,EAAE,iBAAA,EAAmB,CAAA,EAAG,UAAA,EAAY,EAAE,CAAA;AAEjE,IAAA,MAAM,WAAA,EAAa,IAAA,CAAK,SAAA,CAAU,WAAA,EAAa,EAAA,EAAI,IAAI,CAAA;AACvD,IAAA,GAAA,CAAI,WAAA,EAAa,UAAA,EAAY;AAC3B,MAAA,MAAM,IAAI,WAAA,CAAY,CAAA,4CAAA,EAAiB,UAAU,CAAA,eAAA,EAAQ,UAAU,CAAA,CAAA,CAAG,CAAA;AAAA,IACxE;AAEA,IAAA,MAAM,OAAA,EAAS,IAAA,CAAK,SAAA,CAAU,WAAA,EAAa,EAAA,EAAI,IAAI,CAAA;AACnD,IAAA,MAAM,SAAA,EAAW,IAAA,CAAK,SAAA,CAAU,WAAA,EAAa,EAAA,EAAI,IAAI,CAAA;AACrD,IAAA,GAAA,CAAI,SAAA,EAAW,OAAA,EAAS,GAAA,EAAK,OAAO,EAAE,iBAAA,EAAmB,CAAA,EAAG,WAAW,CAAA;AAEvE,IAAA,IAAI,kBAAA,EAAoB,CAAA;AACxB,IAAA,IAAI,IAAA,EAAM,QAAA;AACV,IAAA,IAAA,CAAA,IAAS,EAAA,EAAI,CAAA,EAAG,EAAA,EAAI,WAAA,GAAc,IAAA,EAAM,GAAA,GAAM,SAAA,EAAW,MAAA,EAAQ,CAAA,EAAA,EAAK;AACpE,MAAA,GAAA,CAAI,IAAA,CAAK,SAAA,CAAU,GAAA,EAAK,IAAI,EAAA,IAAM,QAAA,EAAY,KAAA;AAC9C,MAAA,kBAAA,GAAqB,IAAA,CAAK,SAAA,CAAU,IAAA,EAAM,EAAA,EAAI,IAAI,CAAA;AAClD,MAAA,MAAM,QAAA,EAAU,IAAA,CAAK,SAAA,CAAU,IAAA,EAAM,EAAA,EAAI,IAAI,CAAA;AAC7C,MAAA,MAAM,SAAA,EAAW,IAAA,CAAK,SAAA,CAAU,IAAA,EAAM,EAAA,EAAI,IAAI,CAAA;AAC9C,MAAA,MAAM,WAAA,EAAa,IAAA,CAAK,SAAA,CAAU,IAAA,EAAM,EAAA,EAAI,IAAI,CAAA;AAChD,MAAA,IAAA,GAAO,GAAA,EAAK,QAAA,EAAU,SAAA,EAAW,UAAA;AAAA,IACnC;AAEA,IAAA,GAAA,CAAI,kBAAA,EAAoB,mBAAA,EAAqB;AAC3C,MAAA,MAAM,IAAI,WAAA,CAAY,CAAA,kDAAA,EAAA,CAAmB,kBAAA,EAAoB,KAAA,EAAO,IAAA,CAAA,CAAM,OAAA,CAAQ,CAAC,CAAC,CAAA,iBAAA,EAAU,oBAAA,EAAsB,KAAA,EAAO,IAAI,CAAA,GAAA,CAAK,CAAA;AAAA,IACtI;AAEA,IAAA,OAAO,EAAE,iBAAA,EAAmB,WAAW,CAAA;AAAA,EACzC,EAAA,MAAA,CAAS,GAAA,EAAK;AACZ,IAAA,GAAA,CAAI,IAAA,WAAe,WAAA,EAAa,MAAM,GAAA;AACtC,IAAA,OAAO,EAAE,iBAAA,EAAmB,CAAA,EAAG,UAAA,EAAY,EAAE,CAAA;AAAA,EAC/C;AACF;AASO,SAAS,QAAA,CAAS,GAAA,EAAqB;AAC5C,EAAA,OAAO,GAAA,CAAI,OAAA,CAAQ,SAAA,EAAW,EAAE,CAAA,CAAE,OAAA,CAAQ,wCAAA,EAA0C,EAAE,CAAA;AACxF;AAGA,IAAM,aAAA,EAAe,8BAAA;AACd,SAAS,YAAA,CAAa,IAAA,EAA6B;AACxD,EAAA,MAAM,QAAA,EAAU,IAAA,CAAK,IAAA,CAAK,CAAA;AAC1B,EAAA,GAAA,CAAI,CAAC,QAAA,GAAW,CAAC,YAAA,CAAa,IAAA,CAAK,OAAO,CAAA,EAAG,OAAO,IAAA;AAGpD,EAAA,OAAO,OAAA,CAAQ,OAAA,CAAQ,KAAA,EAAO,KAAK,CAAA,CAAE,OAAA,CAAQ,KAAA,EAAO,KAAK,CAAA;AAC3D;AAKO,SAAS,OAAA,CAAQ,GAAA,EAAuB;AAC7C,EAAA,IAAI,IAAA,EAAM,QAAA;AACV,EAAA,IAAA,CAAA,IAAS,EAAA,EAAI,CAAA,EAAG,EAAA,EAAI,GAAA,CAAI,MAAA,EAAQ,CAAA,EAAA,EAAK,GAAA,CAAI,GAAA,CAAI,CAAC,EAAA,EAAI,GAAA,EAAK,IAAA,EAAM,GAAA,CAAI,CAAC,CAAA;AAClE,EAAA,OAAO,GAAA;AACT;AAGO,SAAS,OAAA,CAAQ,GAAA,EAAuB;AAC7C,EAAA,IAAI,IAAA,EAAM,CAAA,QAAA;AACV,EAAA,IAAA,CAAA,IAAS,EAAA,EAAI,CAAA,EAAG,EAAA,EAAI,GAAA,CAAI,MAAA,EAAQ,CAAA,EAAA,EAAK,GAAA,CAAI,GAAA,CAAI,CAAC,EAAA,EAAI,GAAA,EAAK,IAAA,EAAM,GAAA,CAAI,CAAC,CAAA;AAClE,EAAA,OAAO,GAAA;AACT;AAOO,SAAS,aAAA,CAAc,GAAA,EAAyB;AACrD,EAAA,GAAA,CAAI,CAAA,CAAE,IAAA,WAAe,KAAA,CAAA,EAAQ,OAAO,aAAA;AACpC,EAAA,MAAM,IAAA,EAAM,GAAA,CAAI,OAAA;AAEhB,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,KAAK,CAAA,EAAG,OAAO,eAAA;AAChC,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,oBAAK,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,iCAAQ,CAAA,EAAG,OAAO,WAAA;AAC1D,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,qBAAqB,CAAA,EAAG,OAAO,oBAAA;AAEhD,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,uBAAuB,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,+BAA+B,CAAA,EAAG,OAAO,kBAAA;AACnG,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,UAAU,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,kDAAe,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,4CAAc,CAAA,EAAG,OAAO,UAAA;AACtG,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,MAAM,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,2BAAO,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,2BAAO,CAAA,EAAG,OAAO,oBAAA;AACnF,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,iCAAQ,CAAA,EAAG,OAAO,iBAAA;AACnC,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,cAAI,EAAA,GAAA,CAAM,GAAA,CAAI,QAAA,CAAS,4BAAQ,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,cAAI,CAAA,CAAA,EAAI,OAAO,aAAA;AACjF,EAAA,GAAA,CAAI,GAAA,CAAI,QAAA,CAAS,0BAAM,EAAA,GAAK,GAAA,CAAI,QAAA,CAAS,kCAAS,CAAA,EAAG,OAAO,WAAA;AAC5D,EAAA,OAAO,aAAA;AACT;ADzEA;AACA;AACE;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACF,weAAC","file":"/Users/chris_gomdori/workspace/kordoc/dist/chunk-6MF7JRG3.cjs","sourcesContent":[null,"/** kordoc 공용 유틸리티 */\n\n/** 빌드 타임에 tsup define으로 주입되는 버전 */\ndeclare const __KORDOC_VERSION__: string\nexport const VERSION: string = typeof __KORDOC_VERSION__ !== \"undefined\" ? __KORDOC_VERSION__ : \"0.0.0-dev\"\n\n/**\n * Node.js Buffer → ArrayBuffer 변환\n * pool Buffer의 공유 ArrayBuffer 문제를 안전하게 처리.\n * offset=0이고 전체 ArrayBuffer를 차지하면 복사 없이 직접 반환.\n */\nexport function toArrayBuffer(buf: Buffer): ArrayBuffer {\n  if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {\n    return buf.buffer as ArrayBuffer\n  }\n  return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength) as ArrayBuffer\n}\n\n/**\n * kordoc 내부 에러 클래스 — 사용자에게 노출해도 안전한 메시지만 포함.\n * MCP 에러 정제에서 instanceof로 판별하여 allowlist 패턴 매칭 없이 안전하게 통과.\n */\nexport class KordocError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = \"KordocError\"\n  }\n}\n\n/**\n * 에러 메시지 정제 — KordocError는 그대로, 나머지는 일반 메시지로 대체.\n * 파일시스템 경로, 스택 트레이스 등 내부 정보 노출 방지.\n */\nexport function sanitizeError(err: unknown): string {\n  if (err instanceof KordocError) return err.message\n  return \"문서 처리 중 오류가 발생했습니다\"\n}\n\n/**\n * ZIP 엔트리 경로의 경로 순회 여부 판별.\n * 백슬래시 정규화, .., 절대경로, Windows 드라이브 문자 모두 차단.\n */\nexport function isPathTraversal(name: string): boolean {\n  if (name.includes(\"\\x00\")) return true\n  const normalized = name.replace(/\\\\/g, \"/\")\n  const segments = normalized.split(\"/\")\n  return segments.some(s => s === \"..\") || normalized.startsWith(\"/\") || /^[A-Za-z]:/.test(normalized)\n}\n\n// ─── HWPX 섹션 href 해석 (parser/roundtrip 공용) ────────────────────\n\n/**\n * manifest href를 본문 섹션 경로(`Contents/sectionN.xml`)로 정규화.\n * 비본문 XML(header/script/settings)·path traversal·드라이브 경로는 null로 거른다.\n * 백슬래시 구분자를 슬래시로 통일해 parser와 roundtrip이 동일한 목록을 만든다.\n */\nexport function normalizeSectionHref(href: string): string | null {\n  if (!href) return null\n  let normalized = href.replace(/\\\\/g, \"/\").replace(/^\\/+/, \"\")\n  if (isPathTraversal(normalized)) return null\n  if (/^[Ss]ection\\d+\\.xml$/.test(normalized)) normalized = \"Contents/\" + normalized\n  return /(?:^|\\/)[Ss]ection\\d+\\.xml$/.test(normalized) ? normalized : null\n}\n\n/** sectionN.xml을 N 숫자 순서로 정렬 (section10 > section2). */\nexport function compareSectionPaths(a: string, b: string): number {\n  const ai = Number(a.match(/[Ss]ection(\\d+)\\.xml$/)?.[1] ?? Number.MAX_SAFE_INTEGER)\n  const bi = Number(b.match(/[Ss]ection(\\d+)\\.xml$/)?.[1] ?? Number.MAX_SAFE_INTEGER)\n  return ai === bi ? a.localeCompare(b) : ai - bi\n}\n\n// ─── ZIP 안전 로딩 (ZIP bomb 방지) ────────────────────\n\n/**\n * ZIP bomb 사전 검사 — Central Directory에서 비압축 합계와 엔트리 수 확인.\n * HWPX/XLSX/DOCX 등 모든 ZIP 기반 포맷에서 공통 사용.\n */\nexport function precheckZipSize(\n  buffer: ArrayBuffer,\n  maxUncompressedSize = 256 * 1024 * 1024, // parser-shared MAX_DECOMPRESS_SIZE 와 동기\n  maxEntries = 500,\n): { totalUncompressed: number; entryCount: number } {\n  try {\n    const data = new DataView(buffer)\n    const len = buffer.byteLength\n    // EOCD 시그니처 역방향 스캔\n    let eocdOffset = -1\n    for (let i = len - 22; i >= Math.max(0, len - 65557); i--) {\n      if (data.getUint32(i, true) === 0x06054b50) { eocdOffset = i; break }\n    }\n    if (eocdOffset < 0) return { totalUncompressed: 0, entryCount: 0 }\n\n    const entryCount = data.getUint16(eocdOffset + 10, true)\n    if (entryCount > maxEntries) {\n      throw new KordocError(`ZIP 엔트리 수 초과: ${entryCount} (최대 ${maxEntries})`)\n    }\n\n    const cdSize = data.getUint32(eocdOffset + 12, true)\n    const cdOffset = data.getUint32(eocdOffset + 16, true)\n    if (cdOffset + cdSize > len) return { totalUncompressed: 0, entryCount }\n\n    let totalUncompressed = 0\n    let pos = cdOffset\n    for (let i = 0; i < entryCount && pos + 46 <= cdOffset + cdSize; i++) {\n      if (data.getUint32(pos, true) !== 0x02014b50) break\n      totalUncompressed += data.getUint32(pos + 24, true)\n      const nameLen = data.getUint16(pos + 28, true)\n      const extraLen = data.getUint16(pos + 30, true)\n      const commentLen = data.getUint16(pos + 32, true)\n      pos += 46 + nameLen + extraLen + commentLen\n    }\n\n    if (totalUncompressed > maxUncompressedSize) {\n      throw new KordocError(`ZIP 비압축 크기 초과: ${(totalUncompressed / 1024 / 1024).toFixed(1)}MB (최대 ${maxUncompressedSize / 1024 / 1024}MB)`)\n    }\n\n    return { totalUncompressed, entryCount }\n  } catch (err) {\n    if (err instanceof KordocError) throw err\n    return { totalUncompressed: 0, entryCount: 0 }\n  }\n}\n\n/**\n * XML 파싱 전 프롤로그 정리 — 선두 BOM 제거 + DOCTYPE 제거(XXE/Billion Laughs 방지).\n *\n * BOM: 일부 OpenXML 라이터가 `[Content_Types].xml`·`.rels` 앞에 EF BB BF를 붙인다.\n * XML 스펙상 적법(엔티티는 BOM으로 시작할 수 있음)이고 엑셀·리브레오피스는 여는데,\n * xmldom은 \"선언이 문서 시작이 아님\" fatalError로 파일 전체를 거부했다 (#63).\n */\nexport function stripDtd(xml: string): string {\n  return xml.replace(/^\\uFEFF/, \"\").replace(/<!DOCTYPE\\s[^[>]*(\\[[\\s\\S]*?\\])?\\s*>/gi, \"\")\n}\n\n/** 하이퍼링크 URL 살균 — javascript: 등 XSS 위험 스킴 차단 */\nconst SAFE_HREF_RE = /^(?:https?:|mailto:|tel:|#)/i\nexport function sanitizeHref(href: string): string | null {\n  const trimmed = href.trim()\n  if (!trimmed || !SAFE_HREF_RE.test(trimmed)) return null\n  // 괄호 percent-encoding — 마크다운 링크 목적지에서 불균형 ')'는 링크를 조기 종료시켜\n  // 문법이 깨진다 (원문 하이퍼링크에 ')'가 박힌 실문서 존재). %28/%29는 의미 동일.\n  return trimmed.replace(/\\(/g, \"%28\").replace(/\\)/g, \"%29\")\n}\n\n// ─── 안전한 min/max (스택 오버플로 방지) ─────────────\n\n/** Math.min(...arr) 대체 — 대형 배열에서 스택 오버플로 방지 */\nexport function safeMin(arr: number[]): number {\n  let min = Infinity\n  for (let i = 0; i < arr.length; i++) if (arr[i] < min) min = arr[i]\n  return min\n}\n\n/** Math.max(...arr) 대체 — 대형 배열에서 스택 오버플로 방지 */\nexport function safeMax(arr: number[]): number {\n  let max = -Infinity\n  for (let i = 0; i < arr.length; i++) if (arr[i] > max) max = arr[i]\n  return max\n}\n\n// ─── 에러 분류 ──────────────────────────────────────\n\nimport type { ErrorCode } from \"./types.js\"\n\n/** 에러를 구조화된 ErrorCode로 분류 — KordocError 메시지 패턴 매칭 */\nexport function classifyError(err: unknown): ErrorCode {\n  if (!(err instanceof Error)) return \"PARSE_ERROR\"\n  const msg = err.message\n  // DRM 을 암호화보다 먼저 — \"DRM 암호화된 HWPX…\" 류 메시지가 ENCRYPTED 로 떨어지지 않도록\n  if (msg.includes(\"DRM\")) return \"DRM_PROTECTED\"\n  if (msg.includes(\"암호화\") || msg.includes(\"암호로 보호\")) return \"ENCRYPTED\"\n  if (msg.includes(\"optional dependency\")) return \"MISSING_DEPENDENCY\"\n  // V8 문자열 한계 초과 — 이미지 base64 총량이 큰 문서의 JSON 직렬화 (#65)\n  if (msg.includes(\"Invalid string length\") || msg.includes(\"Cannot create a string longer\")) return \"OUTPUT_TOO_LARGE\"\n  if (msg.includes(\"ZIP bomb\") || msg.includes(\"ZIP 비압축 크기 초과\") || msg.includes(\"ZIP 엔트리 수 초과\")) return \"ZIP_BOMB\"\n  if (msg.includes(\"bomb\") || msg.includes(\"크기 초과\") || msg.includes(\"압축 해제\")) return \"DECOMPRESSION_BOMB\"\n  if (msg.includes(\"이미지 기반\")) return \"IMAGE_BASED_PDF\"\n  if (msg.includes(\"섹션\") && (msg.includes(\"찾을 수 없\") || msg.includes(\"없음\"))) return \"NO_SECTIONS\"\n  if (msg.includes(\"시그니처\") || msg.includes(\"복구할 수 없\")) return \"CORRUPTED\"\n  return \"PARSE_ERROR\"\n}\n"]}

package/dist/{chunk-7JQ2ITT2.js → chunk-7U34YNFR.js}

@@ -1,6 +1,6 @@
 import {
   KordocError
-} from "./chunk-MJPRWCI5.js";
+} from "./chunk-D2XNLIFU.js";
 
 // src/pdf/formula/models.ts
 import { createHash } from "crypto";
@@ -266,4 +266,4 @@ export {
   ensureFormulaModels,
   ensureSingleModel
 };
-//# sourceMappingURL=chunk-7JQ2ITT2.js.map
+//# sourceMappingURL=chunk-7U34YNFR.js.map

package/dist/{chunk-SMGSCP57.js → chunk-BPKJTQAW.js}

@@ -6,7 +6,7 @@ import {
 import {
   safeMax,
   safeMin
-} from "./chunk-MJPRWCI5.js";
+} from "./chunk-D2XNLIFU.js";
 
 // src/pdf/image-regions.ts
 import { OPS } from "pdfjs-dist/legacy/build/pdf.mjs";
@@ -68,6 +68,8 @@ var MAX_LINE_WIDTH = 5;
 function extractLines(fnArray, argsArray) {
   const horizontals = [];
   const verticals = [];
+  const clipRects = [];
+  let pendingClip = false;
   let lineWidth = 1;
   let ctm = [1, 0, 0, 1, 0, 0];
   const ctmStack = [];
@@ -97,9 +99,12 @@ function extractLines(fnArray, argsArray) {
   }
   function flushPath(isStroke, fromFill = false) {
     if (!isStroke) {
+      if (pendingClip) captureClipRect(currentPath, clipRects);
+      pendingClip = false;
       currentPath = [];
       return;
     }
+    pendingClip = false;
     const effWidth = lineWidth * ctmScale();
     for (const seg of currentPath) {
       classifyAndAdd(seg, effWidth, horizontals, verticals, fromFill);
@@ -227,9 +232,29 @@ function extractLines(fnArray, argsArray) {
       case OPS2.endPath:
         flushPath(false);
         break;
+      case OPS2.clip:
+      case OPS2.eoClip:
+        pendingClip = true;
+        break;
     }
   }
-  return { horizontals, verticals };
+  return { horizontals, verticals, clipRects };
+}
+var CLIP_MIN_W = 4;
+var CLIP_MIN_H = 2;
+function captureClipRect(path, out) {
+  if (path.length < 3 || path.length > 5) return;
+  let x1 = Infinity, y1 = Infinity, x2 = -Infinity, y2 = -Infinity;
+  for (const s of path) {
+    const dx = Math.abs(s.x2 - s.x1), dy = Math.abs(s.y2 - s.y1);
+    if (dx > ORIENTATION_TOL && dy > ORIENTATION_TOL) return;
+    x1 = Math.min(x1, s.x1, s.x2);
+    x2 = Math.max(x2, s.x1, s.x2);
+    y1 = Math.min(y1, s.y1, s.y2);
+    y2 = Math.max(y2, s.y1, s.y2);
+  }
+  if (x2 - x1 < CLIP_MIN_W || y2 - y1 < CLIP_MIN_H) return;
+  out.push({ x1, y1, x2, y2 });
 }
 function classifyAndAdd(seg, lineWidth, horizontals, verticals, fromFill = false) {
   const dx = Math.abs(seg.x2 - seg.x1);
@@ -442,9 +467,9 @@ function mergeParallelLines(lines, dir) {
       const prevEnd = dir === "h" ? prev.x2 : prev.y2;
       const currStart = dir === "h" ? curr.x1 : curr.y1;
       const currEnd = dir === "h" ? curr.x2 : curr.y2;
-      const overlap = Math.min(prevEnd, currEnd) - Math.max(prevStart, currStart);
+      const overlap2 = Math.min(prevEnd, currEnd) - Math.max(prevStart, currStart);
       const minLen = Math.min(prevEnd - prevStart, currEnd - currStart);
-      if (overlap > minLen * 0.3) {
+      if (overlap2 > minLen * 0.3) {
         if (dir === "h") {
           prev.x1 = Math.min(prev.x1, curr.x1);
           prev.x2 = Math.max(prev.x2, curr.x2);
@@ -933,6 +958,164 @@ function linesIntersect(a, b) {
   return v.x1 >= h.x1 - tol && v.x1 <= h.x2 + tol && h.y1 >= v.y1 - tol && h.y1 <= v.y2 + tol;
 }
 
+// src/pdf/clip-cells.ts
+var CLIP_EDGE_TOL = 1.5;
+var CLIP_MIN_GROUP = 2;
+var CLIP_MAX_PAGE_FRAC = 0.75;
+var CLIP_MIN_W2 = 4;
+var CLIP_MIN_H2 = 2;
+var STROKE_NEAR = 2;
+var STROKE_COVER = 0.5;
+var CLIP_MIN_INVISIBLE = 0;
+var overlap = (a1, a2, b1, b2) => Math.min(a2, b2) - Math.max(a1, b1);
+function adjacent(a, b) {
+  if ((Math.abs(a.x2 - b.x1) <= CLIP_EDGE_TOL || Math.abs(b.x2 - a.x1) <= CLIP_EDGE_TOL) && overlap(a.y1, a.y2, b.y1, b.y2) > CLIP_EDGE_TOL) return true;
+  if ((Math.abs(a.y2 - b.y1) <= CLIP_EDGE_TOL || Math.abs(b.y2 - a.y1) <= CLIP_EDGE_TOL) && overlap(a.x1, a.x2, b.x1, b.x2) > CLIP_EDGE_TOL) return true;
+  return false;
+}
+function contains(a, b) {
+  return b.x1 >= a.x1 - CLIP_EDGE_TOL && b.x2 <= a.x2 + CLIP_EDGE_TOL && b.y1 >= a.y1 - CLIP_EDGE_TOL && b.y2 <= a.y2 + CLIP_EDGE_TOL && (b.x2 - b.x1 < a.x2 - a.x1 - CLIP_EDGE_TOL || b.y2 - b.y1 < a.y2 - a.y1 - CLIP_EDGE_TOL);
+}
+function clusterCoords(values) {
+  const sorted = [...values].sort((a, b) => a - b);
+  const out = [];
+  let run = [];
+  for (const v of sorted) {
+    if (run.length > 0 && v - run[run.length - 1] > CLIP_EDGE_TOL) {
+      out.push(run.reduce((s, x) => s + x, 0) / run.length);
+      run = [];
+    }
+    run.push(v);
+  }
+  if (run.length > 0) out.push(run.reduce((s, x) => s + x, 0) / run.length);
+  return out;
+}
+var nearestIndex = (coords, v) => {
+  let best = 0;
+  for (let i = 1; i < coords.length; i++) if (Math.abs(coords[i] - v) < Math.abs(coords[best] - v)) best = i;
+  return best;
+};
+function edgeStroked(lines, dir, pos, a1, a2) {
+  const len = a2 - a1;
+  if (len <= 0) return true;
+  for (const l of lines) {
+    if (dir === "h") {
+      if (Math.abs(l.y1 - pos) <= STROKE_NEAR && overlap(l.x1, l.x2, a1, a2) >= len * STROKE_COVER) return true;
+    } else if (Math.abs(l.x1 - pos) <= STROKE_NEAR && overlap(l.y1, l.y2, a1, a2) >= len * STROKE_COVER) {
+      return true;
+    }
+  }
+  return false;
+}
+function buildClipCellGrids(rects, strokedH, strokedV, pageWidth, pageHeight) {
+  const pageArea = pageWidth * pageHeight;
+  const cells = rects.filter((r) => r.x2 - r.x1 >= CLIP_MIN_W2 && r.y2 - r.y1 >= CLIP_MIN_H2 && (pageArea <= 0 || (r.x2 - r.x1) * (r.y2 - r.y1) < pageArea * CLIP_MAX_PAGE_FRAC));
+  if (cells.length < CLIP_MIN_GROUP) return { grids: [], containers: [] };
+  const parent = new Array(cells.length).fill(-1);
+  for (let i = 0; i < cells.length; i++) {
+    for (let j = 0; j < cells.length; j++) {
+      if (i === j || !contains(cells[j], cells[i])) continue;
+      const cur = parent[i];
+      if (cur < 0 || contains(cells[cur], cells[j])) parent[i] = j;
+    }
+  }
+  const isContainer = new Array(cells.length).fill(false);
+  for (const p of parent) if (p >= 0) isContainer[p] = true;
+  const root = cells.map((_, i) => i);
+  const find = (i) => {
+    while (root[i] !== i) {
+      root[i] = root[root[i]];
+      i = root[i];
+    }
+    return i;
+  };
+  for (let i = 0; i < cells.length; i++) {
+    for (let j = i + 1; j < cells.length; j++) {
+      if (parent[i] !== parent[j]) continue;
+      if (adjacent(cells[i], cells[j])) {
+        const ra = find(i), rb = find(j);
+        if (ra !== rb) root[ra] = rb;
+      }
+    }
+  }
+  const groups = /* @__PURE__ */ new Map();
+  for (let i = 0; i < cells.length; i++) {
+    const r = find(i);
+    const g = groups.get(r);
+    if (g) g.push(i);
+    else groups.set(r, [i]);
+  }
+  const containers = [];
+  for (let i = 0; i < cells.length; i++) {
+    if (!isContainer[i]) continue;
+    const r = cells[i];
+    if (!containers.some((c) => Math.abs(c.x1 - r.x1) <= CLIP_EDGE_TOL && Math.abs(c.x2 - r.x2) <= CLIP_EDGE_TOL && Math.abs(c.y1 - r.y1) <= CLIP_EDGE_TOL && Math.abs(c.y2 - r.y2) <= CLIP_EDGE_TOL)) containers.push(r);
+  }
+  const grids = [];
+  for (const idxs of groups.values()) {
+    if (idxs.length < CLIP_MIN_GROUP) continue;
+    const members = idxs.map((i) => cells[i]);
+    let edges = 0, invisible = 0;
+    for (const r of members) {
+      edges += 4;
+      if (!edgeStroked(strokedH, "h", r.y1, r.x1, r.x2)) invisible++;
+      if (!edgeStroked(strokedH, "h", r.y2, r.x1, r.x2)) invisible++;
+      if (!edgeStroked(strokedV, "v", r.x1, r.y1, r.y2)) invisible++;
+      if (!edgeStroked(strokedV, "v", r.x2, r.y1, r.y2)) invisible++;
+    }
+    if (invisible / edges < CLIP_MIN_INVISIBLE) continue;
+    const colXs = clusterCoords(members.flatMap((r) => [r.x1, r.x2]));
+    const rowYs = clusterCoords(members.flatMap((r) => [r.y1, r.y2])).reverse();
+    const numRows = rowYs.length - 1, numCols = colXs.length - 1;
+    if (numRows < 1 || numCols < 1) continue;
+    const occupied = Array.from({ length: numRows }, () => new Array(numCols).fill(false));
+    const out = [];
+    for (const r of members) {
+      const c0 = nearestIndex(colXs, r.x1), c1 = nearestIndex(colXs, r.x2);
+      const r0 = nearestIndex(rowYs, r.y2), r1 = nearestIndex(rowYs, r.y1);
+      if (c1 <= c0 || r1 <= r0) continue;
+      let clash = false;
+      for (let rr = r0; rr < r1 && !clash; rr++) for (let cc = c0; cc < c1; cc++) if (occupied[rr][cc]) {
+        clash = true;
+        break;
+      }
+      if (clash) continue;
+      for (let rr = r0; rr < r1; rr++) for (let cc = c0; cc < c1; cc++) occupied[rr][cc] = true;
+      out.push({ row: r0, col: c0, rowSpan: r1 - r0, colSpan: c1 - c0, bbox: { x1: r.x1, y1: r.y1, x2: r.x2, y2: r.y2 } });
+    }
+    for (let rr = 0; rr < numRows; rr++) {
+      for (let cc = 0; cc < numCols; cc++) {
+        if (occupied[rr][cc]) continue;
+        out.push({ row: rr, col: cc, rowSpan: 1, colSpan: 1, bbox: { x1: colXs[cc], y1: rowYs[rr + 1], x2: colXs[cc + 1], y2: rowYs[rr] } });
+      }
+    }
+    grids.push({
+      rowYs,
+      colXs,
+      bbox: { x1: colXs[0], y1: rowYs[numRows], x2: colXs[numCols], y2: rowYs[0] },
+      vertexRadius: 1,
+      cells: out
+    });
+  }
+  return { grids, containers };
+}
+function dropGridsInside(lineGrids, clipGrids, containers = []) {
+  if (clipGrids.length === 0 && containers.length === 0) return lineGrids;
+  const area = (b) => Math.max(0, b.x2 - b.x1) * Math.max(0, b.y2 - b.y1);
+  const overlapsHalf = (g, b) => {
+    const ix = Math.min(g.x2, b.x2) - Math.max(g.x1, b.x1);
+    const iy = Math.min(g.y2, b.y2) - Math.max(g.y1, b.y1);
+    if (ix <= 0 || iy <= 0) return false;
+    const ga = area(g);
+    return ga > 0 && ix * iy / ga >= 0.5;
+  };
+  return lineGrids.filter((g) => {
+    if (clipGrids.some((c) => overlapsHalf(g.bbox, c.bbox))) return false;
+    if (containers.some((c) => overlapsHalf(g.bbox, c))) return false;
+    return true;
+  });
+}
+
 // src/pdf/cell-extract.ts
 function extractCells(grid, horizontals, verticals) {
   const { rowYs, colXs } = grid;
@@ -1011,8 +1194,8 @@ function hasVerticalLine(verticals, x, topY, botY, vertexRadius) {
       if (cellH < 0.1) continue;
       const overlapTop = Math.min(v.y2, topY);
       const overlapBot = Math.max(v.y1, botY);
-      const overlap = overlapTop - overlapBot;
-      if (overlap >= cellH * 0.75) return true;
+      const overlap2 = overlapTop - overlapBot;
+      if (overlap2 >= cellH * 0.75) return true;
     }
   }
   return false;
@@ -1025,8 +1208,8 @@ function hasHorizontalLine(horizontals, y, leftX, rightX, vertexRadius) {
       if (cellW < 0.1) continue;
       const overlapLeft = Math.max(h.x1, leftX);
       const overlapRight = Math.min(h.x2, rightX);
-      const overlap = overlapRight - overlapLeft;
-      if (overlap >= cellW * 0.75) return true;
+      const overlap2 = overlapRight - overlapLeft;
+      if (overlap2 >= cellW * 0.75) return true;
     }
   }
   return false;
@@ -1133,9 +1316,9 @@ function mergeSuperscriptRows(lines) {
     const curr = lines[i];
     const a = band(prev);
     const b = band(curr);
-    const overlap = Math.min(a.top, b.top) - Math.max(a.bottom, b.bottom);
-    const prevIsFrag = isFrag(prev) && a.height <= b.height * 0.8 && overlap >= a.height * 0.5;
-    const currIsFrag = isFrag(curr) && b.height <= a.height * 0.8 && overlap >= b.height * 0.5;
+    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;
     if (prevIsFrag || currIsFrag) {
       result[result.length - 1] = [...prev, ...curr];
     } else {
@@ -1354,14 +1537,17 @@ function filterHiddenText(items, pageWidth, pageHeight) {
 function collapseEvenSpacing(text) {
   const tokens = text.split(" ");
   const singleCharCount = tokens.filter((t) => t.length === 1).length;
-  if (tokens.length >= 3 && singleCharCount / tokens.length >= 0.7) {
+  if (tokens.length >= 3 && singleCharCount / tokens.length >= 0.7 && !isDateUnitBlank(tokens)) {
     return tokens.join("");
   }
   return text.replace(
     /(?<![가-힣])[가-힣](?: [가-힣\d]){2,}(?![가-힣])/g,
-    (match) => match.replace(/ /g, "")
+    (match) => isDateUnitBlank(match.split(" ")) ? match : match.replace(/ /g, "")
   );
 }
+function isDateUnitBlank(tokens) {
+  return tokens.every((t) => t.length !== 1 || !/[가-힣]/.test(t) || /[년월일시분초]/.test(t));
+}
 function computeBBox(items, pageNum) {
   let minX = Infinity, minY = Infinity, maxX = -Infinity, maxY = -Infinity;
   for (const i of items) {
@@ -1502,9 +1688,9 @@ function mergeSuperscriptLines(lines) {
     const curr = lines[i];
     const a = band(prev);
     const b = band(curr);
-    const overlap = Math.min(a.top, b.top) - Math.max(a.bottom, b.bottom);
-    const prevIsFrag = isFrag(prev) && a.height <= b.height * 0.8 && overlap >= a.height * 0.5;
-    const currIsFrag = isFrag(curr) && b.height <= a.height * 0.8 && overlap >= b.height * 0.5;
+    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;
     if (prevIsFrag || currIsFrag) {
       result[result.length - 1] = [...prev, ...curr];
     } else {
@@ -1653,8 +1839,8 @@ function detectTableCaptions(blocks) {
       const tblBottom = table.bbox.y;
       const gap = capBottom >= tblTop ? capBottom - tblTop : tblBottom - capTop;
       if (gap > CAPTION_MAX_GAP) return false;
-      const overlap = Math.min(b.bbox.x + b.bbox.width, table.bbox.x + table.bbox.width) - Math.max(b.bbox.x, table.bbox.x);
-      if (overlap < Math.min(b.bbox.width, table.bbox.width) * 0.3) return false;
+      const overlap2 = Math.min(b.bbox.x + b.bbox.width, table.bbox.x + table.bbox.width) - Math.max(b.bbox.x, table.bbox.x);
+      if (overlap2 < Math.min(b.bbox.width, table.bbox.width) * 0.3) return false;
     }
     return true;
   };
@@ -1918,8 +2104,8 @@ function removeHeaderFooterBlocks(blocks, pageHeights, warnings) {
         if (!o.bbox || o.bbox.page !== cb.page) continue;
         const oNorm = o.text?.trim().replace(/\d+/g, "#");
         if (oNorm && repeatedPatterns.has(oNorm)) continue;
-        const overlap = Math.min(cb.y + cb.height, o.bbox.y + o.bbox.height) - Math.max(cb.y, o.bbox.y);
-        if (overlap >= cb.height * 0.5) {
+        const overlap2 = Math.min(cb.y + cb.height, o.bbox.y + o.bbox.height) - Math.max(cb.y, o.bbox.y);
+        if (overlap2 >= cb.height * 0.5) {
           sharesLine = true;
           break;
         }
@@ -2191,9 +2377,9 @@ function mergeOverlappingRows(rows) {
     const curr = rows[i];
     const a = rowBand(prev);
     const b = rowBand(curr);
-    const overlap = Math.min(a.top, b.top) - Math.max(a.bottom, b.bottom);
-    const prevIsFrag = isFragmentRow(prev) && a.height <= b.height * 0.8 && overlap >= a.height * 0.5;
-    const currIsFrag = isFragmentRow(curr) && b.height <= a.height * 0.8 && overlap >= b.height * 0.5;
+    const overlap2 = Math.min(a.top, b.top) - Math.max(a.bottom, b.bottom);
+    const prevIsFrag = isFragmentRow(prev) && a.height <= b.height * 0.8 && overlap2 >= a.height * 0.5;
+    const currIsFrag = isFragmentRow(curr) && b.height <= a.height * 0.8 && overlap2 >= b.height * 0.5;
     if (prevIsFrag || currIsFrag) {
       const baseY = prevIsFrag ? curr.y : prev.y;
       result[result.length - 1] = { y: baseY, items: [...prev.items, ...curr.items] };
@@ -3070,8 +3256,8 @@ function markUnderlineItems(items, horizontals, verticals) {
       if (h <= 0 || item.w <= 0 || !item.text.trim()) continue;
       const below = Math.max(h * UNDER_BELOW_EM, UNDER_BELOW_MIN_PT);
       if (line.y1 < item.y - below || line.y1 > item.y + UNDER_ABOVE_PT) continue;
-      const overlap = Math.min(line.x2, item.x + item.w) - Math.max(line.x1, item.x);
-      if (overlap / item.w < UNDER_MIN_OVERLAP_RATIO) continue;
+      const overlap2 = Math.min(line.x2, item.x + item.w) - Math.max(line.x1, item.x);
+      if (overlap2 / item.w < UNDER_MIN_OVERLAP_RATIO) continue;
       matches.push(item);
     }
     if (matches.length === 0) continue;
@@ -3115,8 +3301,8 @@ function hasBoxTopPair(line, maxH, horizontals) {
     if (dy < maxH * UNDER_BOX_PAIR_MIN_EM || dy > maxH * UNDER_BOX_PAIR_MAX_EM) continue;
     const ow = o.x2 - o.x1;
     if (ow <= 0) continue;
-    const overlap = Math.min(line.x2, o.x2) - Math.max(line.x1, o.x1);
-    if (overlap / Math.max(w, ow) >= UNDER_BOX_PAIR_OVERLAP) return true;
+    const overlap2 = Math.min(line.x2, o.x2) - Math.max(line.x1, o.x1);
+    if (overlap2 / Math.max(w, ow) >= UNDER_BOX_PAIR_OVERLAP) return true;
   }
   return false;
 }
@@ -3127,8 +3313,8 @@ function isRepeatedSpanRule(line, horizontals) {
   for (const o of horizontals) {
     const ow = o.x2 - o.x1;
     if (ow <= 0) continue;
-    const overlap = Math.min(line.x2, o.x2) - Math.max(line.x1, o.x1);
-    if (overlap / Math.max(w, ow) < UNDER_SPAN_OVERLAP_RATIO) continue;
+    const overlap2 = Math.min(line.x2, o.x2) - Math.max(line.x1, o.x1);
+    if (overlap2 / Math.max(w, ow) < UNDER_SPAN_OVERLAP_RATIO) continue;
     if (!ys.some((y) => Math.abs(y - o.y1) < 2)) ys.push(o.y1);
   }
   return ys.length >= UNDER_REPEATED_SPAN_LEVELS;
@@ -3162,7 +3348,10 @@ function wrapUnderlineRuns(items) {
 // src/pdf/page-blocks.ts
 function extractPageBlocksWithLines(items, pageNum, opList, pageWidth, pageHeight, extraLines, detectTables = true) {
   if (items.length === 0) return [];
-  let { horizontals, verticals } = extractLines(opList.fnArray, opList.argsArray);
+  const extracted = extractLines(opList.fnArray, opList.argsArray);
+  let { horizontals, verticals } = extracted;
+  const clipResult = detectTables ? buildClipCellGrids(extracted.clipRects, horizontals, verticals, pageWidth, pageHeight) : { grids: [], containers: [] };
+  const clipGrids = clipResult.grids;
   if (extraLines) {
     horizontals = horizontals.concat(extraLines.horizontals);
     verticals = verticals.concat(extraLines.verticals);
@@ -3176,7 +3365,8 @@ function extractPageBlocksWithLines(items, pageNum, opList, pageWidth, pageHeigh
   wrapStrikethroughRuns(items);
   markUnderlineItems(items, horizontals, verticals);
   wrapUnderlineRuns(items);
-  const grids = detectTables ? buildTableGrids(horizontals, verticals) : [];
+  const lineGrids = detectTables ? buildTableGrids(horizontals, verticals) : [];
+  const grids = [...clipGrids, ...dropGridsInside(lineGrids, clipGrids, clipResult.containers)];
   if (grids.length > 0) {
     return extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals);
   }
@@ -3198,8 +3388,8 @@ function markStrikethroughItems(items, horizontals) {
       if (line.lineWidth > h * STRIKE_MAX_THICKNESS_RATIO) continue;
       const centerY = item.y + h * 0.4;
       if (Math.abs(line.y1 - centerY) > h * STRIKE_CENTER_TOLERANCE) continue;
-      const overlap = Math.min(line.x2, item.x + item.w) - Math.max(line.x1, item.x);
-      if (overlap / item.w < STRIKE_MIN_OVERLAP_RATIO) continue;
+      const overlap2 = Math.min(line.x2, item.x + item.w) - Math.max(line.x1, item.x);
+      if (overlap2 / item.w < STRIKE_MIN_OVERLAP_RATIO) continue;
       matches.push(item);
     }
     if (matches.length === 0) continue;
@@ -3276,12 +3466,13 @@ function isProseBoxGrid(grid, verticals, table) {
 function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
   const blocks = [];
   const usedItems = /* @__PURE__ */ new Set();
-  const sortedGrids = [...grids].sort((a, b) => b.bbox.y2 - a.bbox.y2);
+  const gridArea = (g) => (g.bbox.x2 - g.bbox.x1) * (g.bbox.y2 - g.bbox.y1);
+  const sortedGrids = [...grids].sort((a, b) => (b.cells ? 1 : 0) - (a.cells ? 1 : 0) || (a.cells && b.cells ? gridArea(a) - gridArea(b) : 0) || b.bbox.y2 - a.bbox.y2);
   for (const grid of sortedGrids) {
     const numGridRows = grid.rowYs.length - 1;
     const numGridCols = grid.colXs.length - 1;
-    if (numGridRows === 1 && numGridCols >= 2) continue;
-    if (numGridCols === 1 && numGridRows >= 2) continue;
+    if (!grid.cells && numGridRows === 1 && numGridCols >= 2) continue;
+    if (!grid.cells && numGridCols === 1 && numGridRows >= 2) continue;
     const tableItems = [];
     const pad = 3;
     const gridW = grid.bbox.x2 - grid.bbox.x1;
@@ -3293,7 +3484,7 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
       tableItems.push(item);
       usedItems.add(item);
     }
-    const cells = extractCells(grid, horizontals, verticals);
+    const cells = grid.cells ?? extractCells(grid, horizontals, verticals);
     if (cells.length === 0) continue;
     const textItems = tableItems.map((i) => ({
       text: i.text,
@@ -3356,7 +3547,7 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
     };
     const hasContent = finalGrid.some((row) => row.some((cell) => cell.text.trim() !== ""));
     if (!hasContent) continue;
-    if (isProseBoxGrid(grid, verticals, irTable)) {
+    if (!grid.cells && isProseBoxGrid(grid, verticals, irTable)) {
       for (const it of tableItems) usedItems.delete(it);
       continue;
     }
@@ -3367,7 +3558,7 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
       width: grid.bbox.x2 - grid.bbox.x1,
       height: grid.bbox.y2 - grid.bbox.y1
     };
-    if (shouldDemoteTable(irTable)) {
+    if (!grid.cells && shouldDemoteTable(irTable)) {
       const demoted = demoteTableToText(irTable);
       if (demoted) {
         const text = numGridRows === 1 ? "\n" + demoted + "\n" : demoted;
@@ -3686,4 +3877,4 @@ export {
   extractPageBlocksWithLines,
   mergeCrossPageTables
 };
-//# sourceMappingURL=chunk-SMGSCP57.js.map
+//# sourceMappingURL=chunk-BPKJTQAW.js.map