refactor(board): 처리 머신 보드 나스+맥미니 2노드 재구성

2026-07-02 컷오버 반영 — GPU 서버 퇴역, 맥북 night-drain 보류(06-29 결정).

- 레인 2개: 나스(추출/마크다운/청크·임베딩 등 DS 본체 Docker 스테이지),
  맥미니(분류/요약/심층분석 — 단일 생성 LLM 허브 + bge-m3/리랭크)
- summarize 풀 분리(summarize_by_machine·ai_model_version 조인 SQL) 제거
  — FE 유일 소비자 확인 후 응답 스키마에서 정리 (5쿼리 -> 4쿼리)
- 맥북 전제 UI 제거: 요약 오프로드 분담막대·요약 합류 칩·번다운 합류
  변곡점 마커·잠듦 문구·전역 스트립 맥북 칩(맥미니 칩으로 대체)
- deferred_pending = LLM 백오프 신호로 맥미니 카드 귀속 (기능 보존)
- 번다운 차트·정직 ETA·실패 드로어·백그라운드 작업 등 머신 무관 기능 보존
- background_jobs 머신 귀속 기본값 gpu -> nas
- 단위테스트 2노드 기준 재작성 (27 passed)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Caddyfile

@@ -19,6 +19,14 @@ http://document.hyungi.net {
         Referrer-Policy strict-origin-when-cross-origin
         -Server
     }
+
+    # 2노드 이관(2026-07-02): 업로드 100MB 한도 집행을 edge(home-caddy)에서 DS 내부로 재홈.
+    # 인그레스가 DSM 리버스 프록시(한도 GUI 미노출)로 바뀌어도 413 단일 소스 유지.
+    # config.yaml upload.max_bytes(100000000)와 정합.
+    request_body {
+        max_size 100MB
+    }
+
     encode {
         gzip
         match {

app/Dockerfile

@@ -11,8 +11,8 @@ RUN apt-get update && \
       ffmpeg && \
     apt-get clean && rm -rf /var/lib/apt/lists/*
 
-COPY requirements.txt .
-RUN pip install --no-cache-dir -r requirements.txt
+COPY requirements.txt requirements.lock ./
+RUN pip install --no-cache-dir -r requirements.lock
 
 COPY . .
 

app/ai/client.py

@@ -290,23 +290,43 @@ class AIClient:
         return response.json()["embedding"]
 
     async def rerank(self, query: str, texts: list[str]) -> list[dict]:
-        """TEI bge-reranker-v2-m3 호출 (Phase 1.3).
+        """리랭커 호출 — ai.models.rerank.protocol 로 백엔드 분기 (2노드 이관 2026-07-02).
 
-        TEI POST /rerank API:
+        공통 반환 계약: [{"index": int, "score": float}, ...] (score 내림차순)
+
+        "tei" (기본, 무회귀) — TEI POST /rerank:
             request:  {"query": str, "texts": [str, ...]}
             response: [{"index": int, "score": float}, ...] (정렬됨)
+        "llamacpp" — llama.cpp POST /v1/rerank (bge-reranker GGUF, 맥미니 :8807):
+            request:  {"model": str, "query": str, "documents": [str, ...]}
+            response: {"results": [{"index": int, "relevance_score": float}, ...]}
+            → normalize_llamacpp_rerank 로 TEI 형태 정규화.
 
+        미지원 protocol = ValueError (명시 실패 — silent fallback 금지).
         timeout은 self.ai.rerank.timeout (config.yaml).
         호출자(rerank_service)가 asyncio.Semaphore + try/except로 감쌈.
         """
+        protocol = getattr(self.ai.rerank, "protocol", "tei") or "tei"
         timeout = float(self.ai.rerank.timeout) if self.ai.rerank.timeout else 5.0
-        response = await self._http.post(
-            self.ai.rerank.endpoint,
-            json={"query": query, "texts": texts},
-            timeout=timeout,
-        )
-        response.raise_for_status()
-        return response.json()
+        if protocol == "tei":
+            response = await self._http.post(
+                self.ai.rerank.endpoint,
+                json={"query": query, "texts": texts},
+                timeout=timeout,
+            )
+            response.raise_for_status()
+            return response.json()
+        if protocol == "llamacpp":
+            from ai.rerank_protocol import normalize_llamacpp_rerank
+
+            response = await self._http.post(
+                self.ai.rerank.endpoint,
+                json={"model": self.ai.rerank.model, "query": query, "documents": texts},
+                timeout=timeout,
+            )
+            response.raise_for_status()
+            return normalize_llamacpp_rerank(response.json())
+        raise ValueError(f"unknown rerank protocol: {protocol}")
 
     async def _call_chat(self, model_config, prompt: str) -> str:
         """OpenAI 호환 API 호출 (R6: 무동의 클라우드 폴백 제거).

app/ai/rerank_protocol.py

@@ -0,0 +1,24 @@
+"""rerank 백엔드 응답 정규화 — 2노드 이관 (2026-07-02, main-server-retirement-1 P1-4).
+
+TEI(/rerank)와 llama.cpp(/v1/rerank)는 요청/응답 스키마가 다르다.
+소비자(rerank_service)는 TEI 형태 [{"index": int, "score": float}]를 기대하므로
+llama.cpp 응답을 여기서 정규화한다. 순수 함수(stdlib only) — 단위 테스트 대상.
+"""
+
+
+def normalize_llamacpp_rerank(payload: dict) -> list[dict]:
+    """llama.cpp /v1/rerank 응답을 TEI 형태로 정규화.
+
+    입력:  {"results": [{"index": int, "relevance_score": float}, ...], ...}
+    반환:  [{"index": int, "score": float}, ...] (score 내림차순 — TEI '정렬됨' 계약 유지)
+
+    index/relevance_score 가 없는 항목은 버린다 (소비자 측 idx/sc None 가드와 동일 방어).
+    """
+    results = payload.get("results") or []
+    normalized = [
+        {"index": r["index"], "score": float(r["relevance_score"])}
+        for r in results
+        if r.get("index") is not None and r.get("relevance_score") is not None
+    ]
+    normalized.sort(key=lambda r: -r["score"])
+    return normalized

app/api/documents.py

@@ -28,7 +28,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 from starlette.requests import ClientDisconnect
 
 from ai.client import AIClient, _load_prompt, parse_json_response
-from core.auth import get_current_user
+from core.auth import get_current_user, get_egress_class
 from core.config import settings
 from core.database import async_session, get_session
 from core.utils import file_hash
@@ -742,11 +742,31 @@ async def get_document(
     doc_id: int,
     user: Annotated[User, Depends(get_current_user)],
     session: Annotated[AsyncSession, Depends(get_session)],
+    egress_class: Annotated[str, Depends(get_egress_class)],
 ):
-    """문서 단건 조회. 본문(extracted_text)·canonical markdown 동봉."""
+    """문서 단건 조회. 본문(extracted_text)·canonical markdown 동봉.
+
+    cloud egress(갭2): egress=cloud 토큰(예: Claude/MCP)은 search 와 동일한 cloud-eligibility
+    게이트를 통과한 문서만 열람 가능 — id 직접 fetch 로 비공개/인프라/개인/restricted 문서를
+    우회 열람하는 경로를 차단한다. 부적격은 404(존재 자체 비노출). local 토큰=게이트 미발동(무회귀).
+    """
+    from sqlalchemy import text as sql_text
+    from services.search.retrieval_service import cloud_eligible_doc_sql
+
     doc = await session.get(Document, doc_id)
     if not doc or doc.deleted_at is not None:
         raise HTTPException(status_code=404, detail="문서를 찾을 수 없습니다")
+    if egress_class == "cloud":
+        eligible = (
+            await session.execute(
+                sql_text(
+                    "SELECT 1 FROM documents WHERE id = :doc_id AND deleted_at IS NULL"
+                    + cloud_eligible_doc_sql("")
+                ).bindparams(doc_id=doc_id)
+            )
+        ).first()
+        if eligible is None:
+            raise HTTPException(status_code=404, detail="문서를 찾을 수 없습니다")
     return DocumentDetailResponse.model_validate(doc)
 
 
@@ -1028,6 +1048,19 @@ async def get_document_image_raw(
             DocumentImage.image_key == image_key,
         )
     )
+    if img is None:
+        # clause-KB: 절-문서는 부모 표준 이미지를 공유(md_content=부모 슬라이스) → parent_id 폴백.
+        from sqlalchemy import text as sql_text
+        _par = (await session.execute(
+            sql_text("SELECT parent_id FROM documents WHERE id = :id").bindparams(id=doc_id)
+        )).scalar()
+        if _par is not None:
+            img = await session.scalar(
+                select(DocumentImage).where(
+                    DocumentImage.document_id == _par,
+                    DocumentImage.image_key == image_key,
+                )
+            )
     if img is None:
         raise HTTPException(status_code=404, detail="이미지를 찾을 수 없습니다")
 
@@ -1801,3 +1834,305 @@ async def analyze_document(
             error_code=error_code,
             source=source,
         )
+
+
+# ─── ASME 절-지식베이스: 유기적 책 네비 (clause-KB, doc_kind='clause' 자식 문서 기반) ───
+class ClauseTocItem(BaseModel):
+    id: int
+    clause_code: str | None = None
+    clause_part: str | None = None
+    clause_order: int | None = None
+    title: str | None = None
+
+
+class ClauseBookResponse(BaseModel):
+    parent_id: int
+    parent_title: str | None = None
+    clauses: list[ClauseTocItem]
+
+
+@router.get("/{doc_id}/clauses", response_model=ClauseBookResponse)
+async def get_document_clauses(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+):
+    """부모 표준 doc 의 절-문서 목차(유기적 책 TOC). doc_kind='clause' 자식을 clause_order 순 반환.
+
+    절-문서는 in_corpus=false + doc_kind='clause'(검색 제외)라 일반 목록/검색엔 안 뜨지만,
+    이 책-내 네비는 부모 표준에서 자식 절로 진입하는 전용 경로다(ASME 2025판=한 권의 책).
+    """
+    from sqlalchemy import text as sql_text
+
+    parent = await session.get(Document, doc_id)
+    if not parent or parent.deleted_at is not None:
+        raise HTTPException(status_code=404, detail="문서를 찾을 수 없습니다")
+    rows = (
+        await session.execute(
+            sql_text(
+                """
+                SELECT id, clause_code, clause_part, clause_order, title
+                FROM documents
+                WHERE parent_id = :pid AND doc_kind = 'clause' AND deleted_at IS NULL
+                ORDER BY clause_order
+                """
+            ).bindparams(pid=doc_id)
+        )
+    ).mappings().all()
+    return ClauseBookResponse(
+        parent_id=doc_id,
+        parent_title=parent.title,
+        clauses=[ClauseTocItem(**dict(r)) for r in rows],
+    )
+
+
+class BacklinkRef(BaseModel):
+    code: str
+    doc_id: int | None = None   # 해소된 절-문서(같은 부모) — dangling 이면 None
+    title: str | None = None
+    anchor: str | None = None
+    ctx: str | None = None
+
+
+class BacklinksResponse(BaseModel):
+    doc_id: int
+    clause_code: str | None = None
+    parent_id: int | None = None
+    prev: ClauseTocItem | None = None
+    next: ClauseTocItem | None = None
+    forward: list[BacklinkRef]   # 이 절이 참조하는 절들
+    back: list[BacklinkRef]      # 이 절을 참조하는 절들
+
+
+@router.get("/{doc_id}/backlinks", response_model=BacklinksResponse)
+async def get_document_backlinks(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+):
+    """절-문서의 양방향 백링크 + 같은 부모 내 이전/다음 절(유기적 책 흐름)."""
+    from sqlalchemy import text as sql_text
+
+    doc = await session.get(Document, doc_id)
+    if not doc or doc.deleted_at is not None:
+        raise HTTPException(status_code=404, detail="문서를 찾을 수 없습니다")
+
+    _meta = (await session.execute(sql_text(
+        "SELECT parent_id, clause_code, clause_order FROM documents WHERE id = :id"
+    ).bindparams(id=doc_id))).mappings().first()
+    _parent_id = _meta["parent_id"] if _meta else None
+    _clause_code = _meta["clause_code"] if _meta else None
+    _clause_order = _meta["clause_order"] if _meta else None
+    forward = (
+        await session.execute(
+            sql_text(
+                """
+                SELECT cl.dst_code AS code, cl.dst_doc_id AS doc_id, cl.anchor, cl.ctx, d.title
+                FROM clause_links cl
+                LEFT JOIN documents d ON d.id = cl.dst_doc_id
+                WHERE cl.src_doc_id = :id
+                ORDER BY cl.char_off NULLS LAST
+                LIMIT 300
+                """
+            ).bindparams(id=doc_id)
+        )
+    ).mappings().all()
+    back = (
+        await session.execute(
+            sql_text(
+                """
+                SELECT s.clause_code AS code, cl.src_doc_id AS doc_id, s.title, cl.ctx
+                FROM clause_links cl
+                JOIN documents s ON s.id = cl.src_doc_id
+                WHERE cl.dst_doc_id = :id
+                ORDER BY s.clause_order NULLS LAST
+                LIMIT 300
+                """
+            ).bindparams(id=doc_id)
+        )
+    ).mappings().all()
+
+    prev = nxt = None
+    if _parent_id is not None and _clause_order is not None:
+        prow = (
+            await session.execute(
+                sql_text(
+                    """
+                    SELECT id, clause_code, clause_part, clause_order, title FROM documents
+                    WHERE parent_id = :pid AND doc_kind='clause' AND deleted_at IS NULL
+                      AND clause_order < :ord
+                    ORDER BY clause_order DESC LIMIT 1
+                    """
+                ).bindparams(pid=_parent_id, ord=_clause_order)
+            )
+        ).mappings().first()
+        nrow = (
+            await session.execute(
+                sql_text(
+                    """
+                    SELECT id, clause_code, clause_part, clause_order, title FROM documents
+                    WHERE parent_id = :pid AND doc_kind='clause' AND deleted_at IS NULL
+                      AND clause_order > :ord
+                    ORDER BY clause_order ASC LIMIT 1
+                    """
+                ).bindparams(pid=_parent_id, ord=_clause_order)
+            )
+        ).mappings().first()
+        prev = ClauseTocItem(**dict(prow)) if prow else None
+        nxt = ClauseTocItem(**dict(nrow)) if nrow else None
+
+    return BacklinksResponse(
+        doc_id=doc_id,
+        clause_code=_clause_code,
+        parent_id=_parent_id,
+        prev=prev,
+        next=nxt,
+        forward=[BacklinkRef(**dict(r)) for r in forward],
+        back=[BacklinkRef(**dict(r)) for r in back],
+    )
+
+
+# ─── 관련 문서 (유사도, on-demand pgvector KNN — 저부하·무저장) ───
+class RelatedItem(BaseModel):
+    id: int
+    title: str | None = None
+    ai_domain: str | None = None
+    material_type: str | None = None
+    year: int | None = None
+    sim: float | None = None
+
+
+class RelatedResponse(BaseModel):
+    doc_id: int
+    related: list[RelatedItem]
+
+
+@router.get("/{doc_id}/related", response_model=RelatedResponse)
+async def get_related_documents(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    limit: int = 8,
+    same_type: bool = True,
+):
+    """문서-레벨 임베딩 코사인 최근접 = '관련 문서'. on-demand(저장/배치 없음).
+
+    인용그래프가 부적합한 코퍼스(업계 기술기사=인용망 부재)의 대안 연결 레이어.
+    same_type=true면 같은 material_type 내, false면 전 코퍼스. doc_kind='clause'(절-문서)는 제외.
+    """
+    from sqlalchemy import text as sql_text
+
+    lim = max(1, min(limit, 30))
+    type_clause = "AND d.material_type = src.material_type" if same_type else ""
+    rows = (
+        await session.execute(
+            sql_text(
+                f"""
+                WITH src AS (
+                    SELECT embedding, material_type FROM documents WHERE id = :id
+                )
+                SELECT d.id, d.title, d.ai_domain, d.material_type, d.facet_year AS year,
+                       round((1 - (d.embedding <=> (SELECT embedding FROM src)))::numeric, 3) AS sim
+                FROM documents d, src
+                WHERE d.doc_kind = 'standard' AND d.deleted_at IS NULL
+                  AND d.id <> :id AND d.embedding IS NOT NULL
+                  AND (SELECT embedding FROM src) IS NOT NULL
+                  {type_clause}
+                ORDER BY d.embedding <=> (SELECT embedding FROM src)
+                LIMIT :lim
+                """
+            ).bindparams(id=doc_id, lim=lim)
+        )
+    ).mappings().all()
+    return RelatedResponse(
+        doc_id=doc_id,
+        related=[RelatedItem(**{k: r[k] for k in ("id", "title", "ai_domain", "material_type", "year")}, sim=float(r["sim"]) if r["sim"] is not None else None) for r in rows],
+    )
+
+
+# ─── 절 공부도구 (노트/형광펜/암기카드) — clause_study ───
+class StudyItem(BaseModel):
+    id: int
+    kind: str
+    payload: dict = {}
+    created_at: datetime | None = None
+
+
+class StudyListResponse(BaseModel):
+    doc_id: int
+    items: list[StudyItem]
+
+
+class StudyCreate(BaseModel):
+    kind: str           # note | highlight | card
+    payload: dict = {}
+
+
+def _parse_payload(p):
+    import json
+    if isinstance(p, str):
+        try:
+            return json.loads(p)
+        except Exception:
+            return {}
+    return p or {}
+
+
+@router.get("/{doc_id}/study", response_model=StudyListResponse)
+async def list_study(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+):
+    """절-문서의 공부도구 항목(노트/형광펜/암기카드) 목록."""
+    from sqlalchemy import text as sql_text
+    rows = (
+        await session.execute(
+            sql_text("SELECT id, kind, payload, created_at FROM clause_study "
+                     "WHERE doc_id = :id ORDER BY created_at DESC").bindparams(id=doc_id)
+        )
+    ).mappings().all()
+    return StudyListResponse(
+        doc_id=doc_id,
+        items=[StudyItem(id=r["id"], kind=r["kind"], payload=_parse_payload(r["payload"]),
+                         created_at=r["created_at"]) for r in rows],
+    )
+
+
+@router.post("/{doc_id}/study", response_model=StudyItem, status_code=201)
+async def add_study(
+    doc_id: int,
+    body: StudyCreate,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+):
+    """노트/형광펜/암기카드 1건 추가."""
+    import json
+    from sqlalchemy import text as sql_text
+    if body.kind not in ("note", "highlight", "card"):
+        raise HTTPException(status_code=400, detail="kind 는 note/highlight/card")
+    row = (
+        await session.execute(
+            sql_text("INSERT INTO clause_study(doc_id, kind, payload) "
+                     "VALUES (:d, :k, cast(:p AS jsonb)) RETURNING id, kind, payload, created_at")
+            .bindparams(d=doc_id, k=body.kind, p=json.dumps(body.payload, ensure_ascii=False))
+        )
+    ).mappings().first()
+    await session.commit()
+    return StudyItem(id=row["id"], kind=row["kind"], payload=_parse_payload(row["payload"]),
+                     created_at=row["created_at"])
+
+
+@router.delete("/{doc_id}/study/{study_id}", status_code=204)
+async def delete_study(
+    doc_id: int,
+    study_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+):
+    from sqlalchemy import text as sql_text
+    await session.execute(
+        sql_text("DELETE FROM clause_study WHERE id = :s AND doc_id = :d")
+        .bindparams(s=study_id, d=doc_id)
+    )
+    await session.commit()

app/api/ingest_study.py

@@ -23,6 +23,7 @@ from datetime import datetime, timezone
 from fastapi import APIRouter, Depends, Header, HTTPException
 from pydantic import BaseModel
 from sqlalchemy import select
+from sqlalchemy.exc import IntegrityError
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from core.config import settings
@@ -66,6 +67,22 @@ class IngestBody(BaseModel):
     attempts: list[IngestAttempt]
 
 
+def _already_ingested(rows) -> dict:
+    """이미 적재된 세션들의 캐시 요약(멱등 응답). 최초 멱등체크 + 동시경합 흡수 양쪽에서 사용."""
+    return {
+        "status": "already_ingested",
+        "sessions": [
+            {
+                "topic_id": s.study_topic_id,
+                "correct": s.correct_count,
+                "wrong": s.wrong_count,
+                "unsure": s.unsure_count,
+            }
+            for s in rows
+        ],
+    }
+
+
 def _parse_answered_at(s: str | None, now: datetime) -> datetime:
     if not s:
         return now
@@ -98,18 +115,7 @@ async def ingest_attempts(
         )
     ).scalars().all()
     if existing:
-        return {
-            "status": "already_ingested",
-            "sessions": [
-                {
-                    "topic_id": s.study_topic_id,
-                    "correct": s.correct_count,
-                    "wrong": s.wrong_count,
-                    "unsure": s.unsure_count,
-                }
-                for s in existing
-            ],
-        }
+        return _already_ingested(existing)
 
     # pub_id → source_id(내부 질문 id) 해소. deleted tombstone 제외.
     pub_ids = list({a.question_pub_id for a in body.attempts})
@@ -156,73 +162,92 @@ async def ingest_attempts(
     if not by_topic:
         raise HTTPException(status_code=404, detail="해소된 attempt 없음")
 
-    summaries = []
-    for topic_id, items in by_topic.items():
-        qids = [q.id for (_, q) in items]
-        qs = StudyQuizSession(
-            user_id=user_id,
-            study_topic_id=topic_id,
-            question_ids=qids,
-            subject_distribution={},
-            status="done",
-            cursor=len(qids),
-            source="viewer",
-            client_session_uuid=body.client_session_uuid,
-            finished_at=now,
-            created_at=now,
-            updated_at=now,
-        )
-        session.add(qs)
-        await session.flush()  # qs.id
+    try:
+        summaries = []
+        for topic_id, items in by_topic.items():
+            qids = [q.id for (_, q) in items]
+            qs = StudyQuizSession(
+                user_id=user_id,
+                study_topic_id=topic_id,
+                question_ids=qids,
+                subject_distribution={},
+                status="done",
+                cursor=len(qids),
+                source="viewer",
+                client_session_uuid=body.client_session_uuid,
+                finished_at=now,
+                created_at=now,
+                updated_at=now,
+            )
+            session.add(qs)
+            await session.flush()  # qs.id
 
-        c = w = u = 0
-        for a, q in items:
-            try:
-                sel, is_corr, outcome = derive_outcome(a.selected_choice, a.is_unsure, q.correct_choice)
-            except ValueError:
-                skipped.append(a.question_pub_id)  # 선택 없고 unsure 아님 = 무효 → skip
-                continue
-            if outcome == "correct":
-                c += 1
-            elif outcome == "wrong":
-                w += 1
-            elif outcome == "unsure":
-                u += 1
-            session.add(
-                StudyQuestionAttempt(
-                    user_id=user_id,
-                    study_question_id=q.id,
-                    study_topic_id=topic_id,
-                    selected_choice=sel,
-                    correct_choice=q.correct_choice,
-                    is_correct=is_corr,
-                    outcome=outcome,
-                    quiz_session_id=qs.id,
-                    answered_at=_parse_answered_at(a.answered_at, now),
+            c = w = u = 0
+            for a, q in items:
+                try:
+                    sel, is_corr, outcome = derive_outcome(a.selected_choice, a.is_unsure, q.correct_choice)
+                except ValueError:
+                    skipped.append(a.question_pub_id)  # 선택 없고 unsure 아님 = 무효 → skip
+                    continue
+                if outcome == "correct":
+                    c += 1
+                elif outcome == "wrong":
+                    w += 1
+                elif outcome == "unsure":
+                    u += 1
+                session.add(
+                    StudyQuestionAttempt(
+                        user_id=user_id,
+                        study_question_id=q.id,
+                        study_topic_id=topic_id,
+                        selected_choice=sel,
+                        correct_choice=q.correct_choice,
+                        is_correct=is_corr,
+                        outcome=outcome,
+                        quiz_session_id=qs.id,
+                        answered_at=_parse_answered_at(a.answered_at, now),
+                    )
+                )
+            qs.correct_count, qs.wrong_count, qs.unsure_count = c, w, u
+            await session.flush()
+
+            # finalize 무수정 재생(progress/SR/pattern + 4-A/4-B enqueue). 그 후 멱등 마커.
+            summary = await finalize_session(
+                session, user_id=user_id, study_topic_id=topic_id, quiz_session_id=qs.id
+            )
+            qs.finalized_at = now
+            summaries.append(
+                {
+                    "topic_id": topic_id,
+                    "quiz_session_id": qs.id,
+                    "correct": summary.correct,
+                    "wrong": summary.wrong,
+                    "unsure": summary.unsure,
+                    "newly_correct": summary.newly_correct,
+                    "relapsed": summary.relapsed,
+                    "recovered": summary.recovered,
+                }
+            )
+
+        await session.commit()
+    except IntegrityError:
+        # 동시 같은 client_session_uuid 경합 — 상대가 먼저 commit → (client_session_uuid,
+        # study_topic_id) uq(mig376) 위반. 데이터는 안전(원자 1-tx 전체 롤백 → SR 이중 advance
+        # 없음). 승자 결과로 graceful 수렴(500 대신 already_ingested). uuid 경합이 아닌 진짜
+        # 무결성 오류면 재조회가 비어 → re-raise 로 표면화.
+        await session.rollback()
+        winner = (
+            await session.execute(
+                select(StudyQuizSession).where(
+                    StudyQuizSession.client_session_uuid == body.client_session_uuid
                 )
             )
-        qs.correct_count, qs.wrong_count, qs.unsure_count = c, w, u
-        await session.flush()
+        ).scalars().all()
+        if not winner:
+            raise
+        logger.info("study_ingest uuid=%s 동시경합 흡수 → already_ingested", body.client_session_uuid)
+        return _already_ingested(winner)
 
-        # finalize 무수정 재생(progress/SR/pattern + 4-A/4-B enqueue). 그 후 멱등 마커.
-        summary = await finalize_session(
-            session, user_id=user_id, study_topic_id=topic_id, quiz_session_id=qs.id
-        )
-        qs.finalized_at = now
-        summaries.append(
-            {
-                "topic_id": topic_id,
-                "quiz_session_id": qs.id,
-                "correct": summary.correct,
-                "wrong": summary.wrong,
-                "unsure": summary.unsure,
-                "newly_correct": summary.newly_correct,
-                "relapsed": summary.relapsed,
-                "recovered": summary.recovered,
-            }
-        )
-
-    await session.commit()
     logger.info(
         "study_ingest uuid=%s user=%s sessions=%s skipped=%s",
         body.client_session_uuid, user_id, len(summaries), len(skipped),

app/api/queue_overview.py

@@ -37,8 +37,8 @@ class CurrentItem(BaseModel):
 
 
 class MachineCard(BaseModel):
-    """머신 카드 — stage 귀속 합산 + 완료 실적(summarize 는 풀 분리) + state."""
-    key: Literal["gpu", "macmini", "macbook"]
+    """머신 카드 — stage 귀속 합산 + 완료 실적 + state (나스/맥미니 2노드)."""
+    key: Literal["nas", "macmini"]
     label: str
     state: Literal["active", "deferred", "idle"]
     stages: list[str]
@@ -59,20 +59,6 @@ class SummarizeEta(BaseModel):
     eta_minutes: int | None
 
 
-class MachineDone(BaseModel):
-    """머신 1대의 summarize 완료 실적 (분담 표시용)."""
-    done_1h: int
-    done_today: int
-
-
-class SummarizeByMachine(BaseModel):
-    """summarize 풀의 머신별 완료 실적 분담 — 보드 레인의 '맥미니 vs 맥북'
-    오프로드 가시화용. rows_to_summarize_split 이 이미 계산하던 값의 노출
-    (ds-board-merged A-1, 신규 수집 SQL 0)."""
-    macmini: MachineDone
-    macbook: MachineDone
-
-
 class TrendBucket(BaseModel):
     """summarize 24h 추이 버킷 — hour 는 KST "HH:00" 라벨."""
     hour: str
@@ -122,7 +108,6 @@ class QueueOverviewResponse(BaseModel):
     machines: list[MachineCard]
     stages: list[StageRow]
     summarize_eta: SummarizeEta
-    summarize_by_machine: SummarizeByMachine
     trend_24h: list[TrendBucket]
     totals: Totals
     background_jobs: list[BackgroundJobItem] = []

app/api/search.py

@@ -15,7 +15,7 @@ from fastapi.responses import JSONResponse
 from pydantic import BaseModel
 from sqlalchemy.ext.asyncio import AsyncSession
 
-from core.auth import get_current_user
+from core.auth import get_current_user, get_egress_class
 from core.database import get_session
 from core.utils import setup_logger
 from models.user import User
@@ -139,6 +139,7 @@ def _build_search_debug(pr: PipelineResult) -> SearchDebug:
 async def search(
     q: str,
     user: Annotated[User, Depends(get_current_user)],
+    egress_class: Annotated[str, Depends(get_egress_class)],
     session: Annotated[AsyncSession, Depends(get_session)],
     background_tasks: BackgroundTasks,
     mode: str = Query("hybrid", pattern="^(fts|trgm|vector|hybrid)$"),
@@ -211,6 +212,8 @@ async def search(
         None, description="안전 자료실 C-1: 관할 필터 (KR/US/EU/JP/GB/INT)"),
     year_from: int | None = Query(None, ge=1900, le=2100, description="published_date 연도 하한 (NULL=created_at fallback)"),
     year_to: int | None = Query(None, ge=1900, le=2100, description="published_date 연도 상한"),
+    domain_bucket: str | None = Query(None, description="377: domain_bucket 스코프 CSV (Safety,Engineering,Law,Philosophy,Programming,General,News). domain_bucket = ANY"),
+    exclude_bucket: str | None = Query(None, description="377: domain_bucket 제외 CSV (예: News). 지식질의 시 News 기본제외용"),
     facets: bool = Query(False, description="안전 자료실 C-1 후속: top-K 결과 분류 축 분포(material_type/jurisdiction/version_status)를 응답 facets 에 집계. 미지정=계산/노출 0"),
 ):
     """문서 검색 — FTS + ILIKE + 벡터 결합 (Phase 3.1 이후 run_search wrapper)"""
@@ -221,6 +224,9 @@ async def search(
             jurisdiction=jurisdiction,
             year_from=year_from,
             year_to=year_to,
+            domain_buckets=[b.strip() for b in domain_bucket.split(",") if b.strip()] if domain_bucket else None,
+            exclude_buckets=[b.strip() for b in exclude_bucket.split(",") if b.strip()] if exclude_bucket else None,
+            cloud_egress=(egress_class == "cloud"),
         )
         pr = await run_search(
             session,

app/api/study_concepts.py

@@ -0,0 +1,94 @@
+"""study_concepts API — 이론공부 홈(오늘의 개념 · 진도 · 회독 SR). prefix = /api/study.
+
+문제풀이 표면 무접촉. 개념문서(가스기사 태그) 읽기 집계 + 회독 SR write 만. 단일 토픽(가스기사=4).
+경로: GET /curriculum · GET /today-concepts · POST /concepts/{doc_id}/read.
+"""
+
+from __future__ import annotations
+
+from typing import Annotated
+
+from fastapi import APIRouter, Depends, HTTPException
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from core.auth import get_current_user
+from core.database import get_session
+from models.user import User
+from services.study import concept_curriculum as cc
+from services.study import concept_links as cl
+
+router = APIRouter()
+
+# 가스기사 단일 토픽 운영(현행). 다토픽 확장 시 쿼리 파라미터로 승격.
+DEFAULT_TOPIC_ID = 4
+
+
+@router.get("/curriculum")
+async def get_curriculum(
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    topic_id: int = DEFAULT_TOPIC_ID,
+):
+    """과목별 회독 진도 + 개념/문항 복습 due 요약."""
+    return await cc.curriculum(session, user.id, topic_id)
+
+
+@router.get("/today-concepts")
+async def get_today_concepts(
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    topic_id: int = DEFAULT_TOPIC_ID,
+    limit: int = 6,
+):
+    """오늘 공부할 개념(재복습 → 미독 빈출순)."""
+    return await cc.today_concepts(session, user.id, topic_id, limit)
+
+
+@router.get("/concepts/weakness-map")
+async def get_weakness_map(
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    topic_id: int = DEFAULT_TOPIC_ID,
+    limit: int = 12,
+):
+    """개념 약점 지도 — 링크된 기출 정답률로 약점 개념(정답률<60%) 우선(이론↔문제)."""
+    name = await cc._topic_name(session, topic_id)
+    if not name:
+        return {"weak": [], "weak_total": 0, "evaluated_total": 0}
+    return await cl.weakness_map(session, user.id, name, limit)
+
+
+@router.get("/concepts/{doc_id}")
+async def get_concept_detail(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    topic_id: int = DEFAULT_TOPIC_ID,
+):
+    """개념 리더 재료 — 구조 파싱(요약/본문/빈출/관련) + 백링크 해소 + 회독/SR + 이전/다음."""
+    detail = await cc.concept_detail(session, user.id, topic_id, doc_id)
+    if detail is None:
+        raise HTTPException(status_code=404, detail="concept not found")
+    return detail
+
+
+@router.get("/concepts/{doc_id}/questions")
+async def get_concept_questions(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    limit: int = 20,
+):
+    """개념 관련 기출 + 내 정답률 (이론↔문제 브리지)."""
+    return await cl.related_questions(session, user.id, doc_id, limit)
+
+
+@router.post("/concepts/{doc_id}/read")
+async def post_concept_read(
+    doc_id: int,
+    user: Annotated[User, Depends(get_current_user)],
+    session: Annotated[AsyncSession, Depends(get_session)],
+    topic_id: int = DEFAULT_TOPIC_ID,
+):
+    """개념 회독 처리 → 회독 플래그 + SR 입고/전진."""
+    return await cc.mark_read(session, user.id, topic_id, doc_id)

app/core/auth.py

@@ -31,11 +31,11 @@ def hash_password(password: str) -> str:
     return bcrypt.hashpw(password.encode(), bcrypt.gensalt()).decode()
 
 
-def create_access_token(subject: str, expires_minutes: int | None = None) -> str:
+def create_access_token(subject: str, expires_minutes: int | None = None, egress: str = "local") -> str:
     minutes = expires_minutes if expires_minutes is not None else ACCESS_TOKEN_EXPIRE_MINUTES
     now = datetime.now(timezone.utc)
     expire = now + timedelta(minutes=minutes)
-    payload = {"sub": subject, "exp": expire, "iat": int(now.timestamp()), "type": "access"}
+    payload = {"sub": subject, "exp": expire, "iat": int(now.timestamp()), "type": "access", "egress": egress}
     return jwt.encode(payload, settings.jwt_secret, algorithm=ALGORITHM)
 
 
@@ -100,6 +100,15 @@ def verify_totp(code: str, secret: str | None = None) -> bool:
     return totp.verify(code)
 
 
+async def get_egress_class(
+    credentials: Annotated[HTTPAuthorizationCredentials, Depends(security)],
+) -> str:
+    """토큰 egress claim -> 'cloud'|'local' (갭2 cloud-egress allowlist). claim 부재=local
+    (비파괴; 기존 토큰=신뢰/로컬). 쿼리파라미터 아님 -> 호출자가 끌 수 없음(우회 차단)."""
+    payload = decode_token(credentials.credentials)
+    return (payload or {}).get("egress", "local")
+
+
 async def get_current_user(
     credentials: Annotated[HTTPAuthorizationCredentials, Depends(security)],
     session: Annotated[AsyncSession, Depends(get_session)],

app/core/config.py

@@ -35,6 +35,12 @@ class AIModelConfig(BaseModel):
     # OpenAI 호환 분기(mlx)만 적용 — Anthropic 분기는 미적용(별 범위).
     repetition_penalty: float | None = None
     top_k: int | None = None
+    # 2노드 이관 (2026-07-02): rerank 백엔드 프로토콜 판별자.
+    # "tei" = TEI POST /rerank {"query","texts"} → [{"index","score"}] (기본, 무회귀)
+    # "llamacpp" = llama.cpp POST /v1/rerank {"model","query","documents"}
+    #              → {"results":[{"index","relevance_score"}]} (맥미니 :8807)
+    # 미지원 값 = client.rerank 가 ValueError (silent fallback 금지). rerank 블록 외 무시.
+    protocol: str = "tei"
 
 
 class DeepSummaryBacklogConfig(BaseModel):
@@ -145,6 +151,12 @@ class Settings(BaseModel):
     # STT (faster-whisper, §3)
     stt_endpoint: str = "http://stt-service:3300"
 
+    # 2노드 이관 (2026-07-02): GPU CUDA 서비스(Surya OCR / faster-whisper) 폐기 대응 명시 게이트.
+    # false = 해당 경로 명시 비활성 — OCR 은 _call_ocr 이 경고 로그 후 None(기존 soft-fail 의미론),
+    # STT 는 터미널 skip + extract_meta 기록. silent 저품질 fallback 아님 (로그/메타로 가시).
+    ocr_enabled: bool = True
+    stt_enabled: bool = True
+
     # §3 file_watcher: Roon 음원 경로 (prefix match 로 skip).
     # 빈 문자열이면 skip 없음. 예: "/documents/PKM/../Music/roon-library" 또는
     # NFS 경유 별도 마운트된 Roon 라이브러리.
@@ -224,6 +236,8 @@ def load_settings() -> Settings:
     kordoc_endpoint = os.getenv("KORDOC_ENDPOINT", "http://kordoc-service:3100")
     ocr_endpoint = os.getenv("OCR_ENDPOINT", "http://ocr-service:3200")
     stt_endpoint = os.getenv("STT_ENDPOINT", "http://stt-service:3300")
+    ocr_enabled = os.getenv("OCR_ENABLED", "true").lower() in ("1", "true", "yes")
+    stt_enabled = os.getenv("STT_ENABLED", "true").lower() in ("1", "true", "yes")
     roon_library_path = os.getenv("ROON_LIBRARY_PATH", "")
 
     # ADDITIONAL_WATCH_TARGETS — 쉼표 구분 (공백 제거)
@@ -343,6 +357,8 @@ def load_settings() -> Settings:
         kordoc_endpoint=kordoc_endpoint,
         ocr_endpoint=ocr_endpoint,
         stt_endpoint=stt_endpoint,
+        ocr_enabled=ocr_enabled,
+        stt_enabled=stt_enabled,
         roon_library_path=roon_library_path,
         additional_watch_targets=additional_watch_targets,
         taxonomy=taxonomy,

app/main.py

@@ -33,6 +33,7 @@ from api.study_sessions import router as study_sessions_router
 from api.study_topics import router as study_topics_router
 from api.study_reminders import router as study_reminders_router
 from api.study_cards import router as study_cards_router
+from api.study_concepts import router as study_concepts_router
 from api.video import router as video_router
 from core.config import settings
 from core.database import async_session, engine, init_db
@@ -249,6 +250,8 @@ app.include_router(study_reminders_router, prefix="/api/study-reminders", tags=[
 app.include_router(study_cards_router, prefix="/api/study-cards", tags=["study-cards"])
 # Phase 1: 학습 진행 상태 (review-complete + review-queue). prefix=/api/study-topics 안에 정의됨.
 app.include_router(study_question_progress_router, prefix="/api", tags=["study-progress"])
+# 이론공부 홈: 오늘의 개념·진도·회독 SR (개념문서 소비 표면, 문제풀이 무접촉).
+app.include_router(study_concepts_router, prefix="/api/study", tags=["study-theory"])
 
 # TODO: Phase 5에서 추가
 # app.include_router(tasks.router, prefix="/api/tasks", tags=["tasks"])

app/models/published.py

@@ -56,5 +56,9 @@ class PublishOutbox(Base):
         DateTime(timezone=True), default=datetime.now, nullable=False
     )
     processed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
+    # mig378: 행별 격리 재시도/terminal. attempts=savepoint 실패 누적, failed_at=MAX 초과 terminal
+    # (set 시 워커 select 에서 제외 → head-of-line block 방지).
+    attempts: Mapped[int] = mapped_column(SmallInteger, nullable=False, default=0)
+    failed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
 
     # 미처리 부분 인덱스 idx(id) WHERE processed_at IS NULL = mig372.

app/models/study_concept_progress.py

@@ -0,0 +1,46 @@
+"""study_concept_progress — 사용자 × 개념문서 단위 간격반복(SR) 진행 (이론공부 홈).
+
+문제 SR(study_question_progress)의 개념(이론)판. '개념문서' = documents 한 건(가스기사 태그).
+회독(첫 read) → 복습 큐 진입, 이후 회독마다 sr_schedule 산술(1·3·7·14·졸업) 공용 전진.
+concept_doc_id 는 documents.id 를 가리키나 FK 미설정 — hot 테이블(documents) 락 회피(clause_study 선례).
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import BigInteger, DateTime, ForeignKey, SmallInteger, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column
+
+from core.database import Base
+
+
+class StudyConceptProgress(Base):
+    __tablename__ = "study_concept_progress"
+    __table_args__ = (
+        UniqueConstraint(
+            "user_id", "concept_doc_id", name="uq_concept_progress_user_doc"
+        ),
+    )
+
+    id: Mapped[int] = mapped_column(BigInteger, primary_key=True)
+    user_id: Mapped[int] = mapped_column(
+        BigInteger, ForeignKey("users.id", ondelete="CASCADE"), nullable=False
+    )
+    study_topic_id: Mapped[int] = mapped_column(
+        BigInteger, ForeignKey("study_topics.id", ondelete="CASCADE"), nullable=False
+    )
+    # documents.id 참조 — FK 없음(락 회피). 개념문서 삭제 시 고아 행은 read 집계에서 자연 제외.
+    concept_doc_id: Mapped[int] = mapped_column(BigInteger, nullable=False)
+
+    # 복습 큐 (sr_schedule 공용): stage 0~3 = 1·3·7·14일, 4 = 졸업(due_at NULL)
+    review_stage: Mapped[int | None] = mapped_column(SmallInteger)
+    due_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
+    last_read_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
+
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), default=datetime.now, nullable=False
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), default=datetime.now, onupdate=datetime.now, nullable=False
+    )

app/policy/prompt_render.py

@@ -36,6 +36,8 @@ KNOWN_4B_TASKS = {
 }
 KNOWN_26B_TASKS = {
     "p3c_deep_summary",
+    # presegment PR2 — 거대문서 map-reduce 의 reduce 단계 (요약들의 요약)
+    "p3c_deep_summary_reduce",
     "p4b_synthesis",
 }
 

app/prompts/policy/p3c_deep_summary_reduce.txt

@@ -0,0 +1,44 @@
+[System]
+너는 긴 문서·문서 묶음 분석가다. 이 문서는 한 번에 처리하기에 너무 커서, 원문을 순서대로 유닛으로 나눠 각 유닛을 먼저 요약했다(map 단계). 아래 "유닛 요약"들은 원문 순서 그대로이며 문서 전체를 빠짐없이 커버한다. 너는 이를 종합해 문서 전체의 최종 분석을 작성한다(reduce 단계).
+
+subject_description: {subject_description}
+
+{forbidden_block}
+
+envelope 를 읽는 순서:
+1. risk_flags 를 먼저 본다. 어떤 위험 때문에 올라온 것인지 파악.
+2. synthesis_directives 를 system 지시로 간주하여 반드시 준수.
+3. distilled_context 는 "참고 요지"일 뿐, 근거는 유닛 요약에서 재확인.
+
+작성 규칙:
+- TL;DR (1문장, 최대 60자)
+- 핵심 (bullets 5개, 각 30~80자)
+- 상세 (2~4 문단, 각 3~5문장) — 유닛(섹션) 순서의 논리 흐름을 보전하며 문서 전체를 관통하는 서술. 특정 유닛만 편식하지 말 것.
+- 유닛 요약에 없는 정보 금지 (hallucination 금지). 숫자·조문·인용은 유닛 요약에 있는 것만 사용.
+- 유닛 요약의 "불일치(...)" 줄들은 중복 제거해 inconsistencies 로 보전 — 임의로 버리지 않는다.
+- synthesis_directives 의 문구 규칙 ("원인은 ~" 금지 등) 반드시 준수.
+- multi_reference_synthesis flag 있으면 레퍼런스별 입장 분리 기술, 종합 권고 금지.
+
+출력 (JSON only):
+{{
+  "mode": "single|bundle",
+  "tldr": "...",
+  "bullets": ["..."],
+  "detail": "...\\n\\n...",
+  "bundle_flow": ["..."] | null,
+  "inconsistencies": ["..."] | null,
+  "entities_confirmed": {{
+    "people": [{{"name": "...", "evidence": "..."}}],
+    "orgs": [...],
+    "projects": [...]
+  }},
+  "directives_applied": ["..."],
+  "confidence": 0.0~1.0
+}}
+
+[User]
+Envelope:
+{{escalation_envelope_json}}
+
+유닛 요약 (총 {{unit_count}}개, 원문 순서 — 각 블록 = 원문 한 구간의 요약):
+{{unit_summaries}}

app/requirements.lock

@@ -0,0 +1,104 @@
+# requirements.lock — 라이브 fastapi 컨테이너 pip freeze 스냅샷 (2026-07-02, 101 pkgs, CVE-clear known-good)
+# 재생성: docker exec hyungi_document_server-fastapi-1 pip freeze > app/requirements.lock (헤더 재부착)
+# requirements.txt = 사람이 편집하는 floor 사양(>=) / 본 lock = Dockerfile 이 실제 설치하는 정본(==)
+annotated-doc==0.0.4
+annotated-types==0.7.0
+anthropic==0.109.1
+anyio==4.13.0
+APScheduler==3.11.2
+asyncpg==0.31.0
+babel==2.18.0
+bcrypt==5.0.0
+beautifulsoup4==4.15.0
+caldav==3.2.1
+certifi==2026.5.20
+cffi==2.0.0
+chardet==7.4.3
+charset-normalizer==3.4.7
+click==8.4.1
+cobble==0.1.4
+courlan==1.4.0
+cryptography==48.0.1
+cssselect==1.4.0
+dateparser==1.4.0
+defusedxml==0.7.1
+distro==1.9.0
+dnspython==2.8.0
+docstring_parser==0.18.0
+ecdsa==0.19.2
+et_xmlfile==2.0.0
+fastapi==0.136.3
+feedparser==6.0.12
+flatbuffers==25.12.19
+greenlet==3.5.1
+h11==0.16.0
+htmldate==1.10.0
+httpcore==1.0.9
+httptools==0.8.0
+httpx==0.28.1
+icalendar==7.1.2
+icalendar-searcher==1.0.6
+idna==3.18
+jh2==5.0.13
+Jinja2==3.1.6
+jiter==0.15.0
+jusText==3.0.2
+lxml==6.1.1
+lxml_html_clean==0.4.5
+magika==0.6.3
+mammoth==1.11.0
+Markdown==3.10.2
+markdownify==1.2.2
+markitdown==0.1.6
+MarkupSafe==3.0.3
+niquests==3.19.1
+numpy==2.4.6
+olefile==0.47
+onnxruntime==1.26.0
+openpyxl==3.1.5
+packaging==26.2
+pandas==3.0.3
+pgvector==0.4.2
+pillow==12.2.0
+protobuf==7.35.0
+pyasn1==0.6.3
+pycparser==3.0
+pydantic==2.13.4
+pydantic_core==2.46.4
+pyhwp==0.1b15
+PyMuPDF==1.27.2.3
+pyotp==2.9.0
+python-dateutil==2.9.0.post0
+python-dotenv==1.2.2
+python-jose==3.5.0
+python-multipart==0.0.32
+python-pptx==1.0.2
+pytz==2026.2
+PyYAML==6.0.3
+qh3==1.9.2
+readability-lxml==0.8.4.1
+recurring-ical-events==3.8.2
+regex==2026.5.9
+requests==2.34.2
+rsa==4.9.1
+sgmllib3k==1.0.0
+six==1.17.0
+sniffio==1.3.1
+soupsieve==2.8.4
+SQLAlchemy==2.0.50
+starlette==1.2.1
+tld==0.13.2
+trafilatura==2.1.0
+typing-inspection==0.4.2
+typing_extensions==4.15.0
+tzdata==2026.2
+tzlocal==5.3.1
+urllib3==2.7.0
+urllib3-future==2.21.902
+uvicorn==0.49.0
+uvloop==0.22.1
+wassima==2.1.1
+watchfiles==1.2.0
+websockets==16.0
+x-wr-timezone==2.0.1
+xlsxwriter==3.2.9

app/services/queue_overview.py

@@ -3,19 +3,16 @@
 GET /api/queue/overview 의 집계 로직. 모든 수치는 기존 processing_queue /
 documents 컬럼에서 라이브 계산 — 신규 테이블/마이그레이션 0 (HARD 제약).
 
-구조: SQL 수집부(build_overview 내부 5쿼리)와 판정부(순수 함수)를 분리.
+구조: SQL 수집부(build_overview 내부 4쿼리)와 판정부(순수 함수)를 분리.
 판정부(rows_to_* / build_machines / build_summarize_eta / build_trend /
 build_totals / compute_eta_minutes)는 DB 없이 단위테스트 가능.
 
-귀속 규칙 (단일 진실):
-- stage→machine 정적 맵: gpu = extract/embed/chunk/markdown/preview/thumbnail/
-  fulltext/stt · macmini = classify/summarize · macbook = deep_summary
-  (단, settings.ai.deep 부재 시 deep_summary 도 macmini 귀속).
-- summarize 는 풀(pool): pending/processing/failed 는 macmini 귀속이되, 완료
-  실적(done_*)은 documents.ai_model_version 조인으로 분리 — 'qwen-macbook'
-  이면 macbook 실적, 아니면 macmini 실적.
-- deferred_pending(payload.deferred_until 미래)은 macbook 카드 귀속
-  (보류 = 맥북 불가 신호).
+귀속 규칙 (단일 진실 — 2026-07-02 컷오버 후 나스+맥미니 2노드):
+- stage→machine 정적 맵: nas = extract/embed/chunk/markdown/preview/thumbnail/
+  fulltext/stt (DS 본체 Docker — 임베딩·리랭크 모델 콜은 맥미니로 나감) ·
+  macmini = classify/summarize/deep_summary (단일 생성 LLM 허브).
+- deferred_pending(payload.deferred_until 미래)은 LLM 백오프 신호 —
+  summarize/deep_summary 소속인 macmini 카드 귀속.
 """
 
 from datetime import datetime, timedelta
@@ -25,42 +22,33 @@ from zoneinfo import ZoneInfo
 from sqlalchemy import bindparam, text
 from sqlalchemy.ext.asyncio import AsyncSession
 
-from core.config import settings
-
 KST = ZoneInfo("Asia/Seoul")
 
-# 내부 판별용 alias — 응답에 raw 모델명 노출 금지, 머신 label 만 노출.
-_MACBOOK_MODEL_ALIAS = "qwen-macbook"
-
 # stage→machine 정적 맵 재료 (선언 순서 = 카드 stages 표시 순서)
-_GPU_STAGES = (
+_NAS_STAGES = (
     "extract", "embed", "chunk", "markdown",
     "preview", "thumbnail", "fulltext", "stt",
 )
-_MACMINI_STAGES = ("classify", "summarize")
-_MACBOOK_STAGES = ("deep_summary",)
-_STAGE_ORDER = _GPU_STAGES + _MACMINI_STAGES + _MACBOOK_STAGES
+_MACMINI_STAGES = ("classify", "summarize", "deep_summary")
+_STAGE_ORDER = _NAS_STAGES + _MACMINI_STAGES
 
-_MACHINE_KEYS = ("gpu", "macmini", "macbook")
+_MACHINE_KEYS = ("nas", "macmini")
 _MACHINE_LABELS = {
-    "gpu": "GPU 서버",
+    "nas": "나스",
     "macmini": "맥미니",
-    "macbook": "맥북 M5 Max",
 }
 
 # 머신 카드당 current 표시 상한
 _CURRENT_LIMIT = 2
 
 
-def stage_machine_map(deep_enabled: bool) -> dict[str, str]:
-    """stage → machine key 맵. deep 슬롯 부재 시 deep_summary 는 macmini 귀속."""
+def stage_machine_map() -> dict[str, str]:
+    """stage → machine key 맵 (정적 — 나스/맥미니 2노드)."""
     mapping: dict[str, str] = {}
-    for s in _GPU_STAGES:
-        mapping[s] = "gpu"
+    for s in _NAS_STAGES:
+        mapping[s] = "nas"
     for s in _MACMINI_STAGES:
         mapping[s] = "macmini"
-    for s in _MACBOOK_STAGES:
-        mapping[s] = "macbook" if deep_enabled else "macmini"
     return mapping
 
 
@@ -90,23 +78,6 @@ def rows_to_stage_stats(rows) -> dict[str, dict]:
     return stats
 
 
-def rows_to_summarize_split(rows) -> dict[str, dict]:
-    """summarize 완료 실적 분리 쿼리 행 → {"macbook"|"macmini": {done_*}}.
-
-    is_macbook = documents.ai_model_version 이 'qwen-macbook' 인지 (내부 판별 전용).
-    """
-    split = {
-        "macbook": {"done_1h": 0, "done_today": 0, "done_15m": 0},
-        "macmini": {"done_1h": 0, "done_today": 0, "done_15m": 0},
-    }
-    for row in rows:
-        key = "macbook" if row[0] else "macmini"
-        split[key]["done_1h"] += int(row[1] or 0)
-        split[key]["done_today"] += int(row[2] or 0)
-        split[key]["done_15m"] += int(row[3] or 0)
-    return split
-
-
 def display_title(row: dict) -> str:
     """표시용 제목 — title > original_filename > file_path basename > 문서 id."""
     if row.get("title"):
@@ -120,13 +91,10 @@ def display_title(row: dict) -> str:
 
 def build_machines(
     stage_stats: dict[str, dict],
-    summarize_split: dict[str, dict],
     current_rows: list[dict],
-    *,
-    deep_enabled: bool,
 ) -> list[dict]:
-    """머신 카드 3장 (gpu / macmini / macbook) 구성 — 귀속 규칙의 판정부."""
-    smap = stage_machine_map(deep_enabled)
+    """머신 카드 2장 (nas / macmini) 구성 — 귀속 규칙의 판정부."""
+    smap = stage_machine_map()
 
     def g(stage: str, field: str) -> int:
         return stage_stats.get(stage, {}).get(field, 0)
@@ -149,29 +117,23 @@ def build_machines(
         pending = sum(g(s, "pending") for s in stages)
         processing = sum(g(s, "processing") for s in stages)
         failed = sum(g(s, "failed") for s in stages)
+        done_1h = sum(g(s, "done_1h") for s in stages)
+        done_today = sum(g(s, "done_today") for s in stages)
+        done_15m = sum(g(s, "done_15m") for s in stages)
 
-        # 완료 실적: summarize 는 풀이라 stage 합산에서 제외하고 split 로 귀속
-        done_1h = sum(g(s, "done_1h") for s in stages if s != "summarize")
-        done_today = sum(g(s, "done_today") for s in stages if s != "summarize")
-        done_15m = sum(g(s, "done_15m") for s in stages if s != "summarize")
-        if key in summarize_split:
-            done_1h += summarize_split[key]["done_1h"]
-            done_today += summarize_split[key]["done_today"]
-            done_15m += summarize_split[key]["done_15m"]
-
-        # 보류 백오프 = 맥북 불가 신호 → macbook 카드 귀속 (deep 슬롯 유무 무관)
+        # 보류 백오프 = LLM 불가 신호 → LLM stage 소속인 macmini 카드 귀속
         deferred_pending = (
             g("summarize", "deferred_pending") + g("deep_summary", "deferred_pending")
-            if key == "macbook" else 0
+            if key == "macmini" else 0
         )
 
         # state 판정 — 우선순위: 가동 > 보류 > 대기 (사용자 피드백 2026-06-11).
         # 일하고 있으면(처리 중 또는 최근 15분 완료) 백오프 잔여가 있어도 "가동" —
         # 보류 건수는 카드의 deferred_pending 라인이 따로 보여준다. "보류" 칩은
-        # 실제로 일이 멈춰 있고 백오프만 쌓인 상태(sleep/불가 지속)에서만.
+        # 실제로 일이 멈춰 있고 백오프만 쌓인 상태(LLM 허브 불가 지속)에서만.
         if processing > 0 or done_15m > 0:
             state = "active"
-        elif key == "macbook" and deferred_pending > 0:
+        elif deferred_pending > 0:
             state = "deferred"
         else:
             state = "idle"
@@ -213,16 +175,6 @@ def build_summarize_eta(stage_stats: dict[str, dict]) -> dict:
     }
 
 
-def build_summarize_by_machine(summarize_split: dict[str, dict]) -> dict:
-    """summarize 머신별 완료 실적 분담 (macmini vs macbook) — 보드 레인의
-    오프로드 가시화용. rows_to_summarize_split 이 이미 만든 값을 응답 형태로
-    투영(done_1h/done_today 만, done_15m 은 내부 state 판정 전용이라 제외)."""
-    def m(key: str) -> dict:
-        s = summarize_split.get(key, {})
-        return {"done_1h": int(s.get("done_1h", 0)), "done_today": int(s.get("done_today", 0))}
-    return {"macmini": m("macmini"), "macbook": m("macbook")}
-
-
 def build_trend(
     inflow_buckets: dict[str, int],
     done_buckets: dict[str, int],
@@ -287,28 +239,23 @@ def build_totals(stage_stats: dict[str, dict]) -> dict:
 
 def compose_overview(
     stage_stats: dict[str, dict],
-    summarize_split: dict[str, dict],
     inflow_buckets: dict[str, int],
     done_buckets: dict[str, int],
     current_rows: list[dict],
     *,
-    deep_enabled: bool,
     now_kst: datetime,
 ) -> dict:
     """수집된 통계 → 응답 dict (계약 shape). 순수 함수 — DB 불요."""
     return {
-        "machines": build_machines(
-            stage_stats, summarize_split, current_rows, deep_enabled=deep_enabled
-        ),
+        "machines": build_machines(stage_stats, current_rows),
         "stages": build_stages(stage_stats),
         "summarize_eta": build_summarize_eta(stage_stats),
-        "summarize_by_machine": build_summarize_by_machine(summarize_split),
         "trend_24h": build_trend(inflow_buckets, done_buckets, now_kst),
         "totals": build_totals(stage_stats),
     }
 
 
-# ─── SQL 수집부 (총 5쿼리) ────────────────────────────────────────────────────
+# ─── SQL 수집부 (총 4쿼리) ────────────────────────────────────────────────────
 
 # 1) stage×status 집계 + 시간창 완료/유입 + 보류 (1방)
 _STAGE_STATS_SQL = """
@@ -333,23 +280,7 @@ _STAGE_STATS_SQL = """
     GROUP BY stage
 """
 
-# 2) summarize 풀 완료 실적 분리 (documents.ai_model_version 조인, 1방)
-#    스캔 하한 = 오늘 0시(KST)와 1h 전 중 더 이른 시각 (자정 직후 1h 창 보전).
-_SUMMARIZE_SPLIT_SQL = """
-    SELECT
-        COALESCE(d.ai_model_version = :macbook_alias, false)                 AS is_macbook,
-        COUNT(*) FILTER (WHERE q.completed_at > NOW() - INTERVAL '1 hour')   AS done_1h,
-        COUNT(*) FILTER (WHERE q.completed_at > :kst_midnight)               AS done_today,
-        COUNT(*) FILTER (WHERE q.completed_at > NOW() - INTERVAL '15 minutes') AS done_15m
-    FROM processing_queue q
-    JOIN documents d ON d.id = q.document_id
-    WHERE q.stage = 'summarize'
-      AND q.status = 'completed'
-      AND q.completed_at > LEAST(:kst_midnight, NOW() - INTERVAL '1 hour')
-    GROUP BY 1
-"""
-
-# 3/4) summarize 24h 추이 — KST 시간 버킷 (inflow/done 각 1방)
+# 2/3) summarize 24h 추이 — KST 시간 버킷 (inflow/done 각 1방)
 _TREND_INFLOW_SQL = """
     SELECT to_char(date_trunc('hour', created_at AT TIME ZONE 'Asia/Seoul'),
                    'YYYY-MM-DD HH24:00')                                     AS bucket,
@@ -371,7 +302,7 @@ _TREND_DONE_SQL = """
     GROUP BY 1
 """
 
-# 5) processing 행 + 표시용 제목 재료 (1방 — 머신별 2건 슬라이스는 판정부에서)
+# 4) processing 행 + 표시용 제목 재료 (1방 — 머신별 2건 슬라이스는 판정부에서)
 _CURRENT_SQL = """
     SELECT q.stage, q.document_id, d.title, d.original_filename, d.file_path
     FROM processing_queue q
@@ -383,20 +314,13 @@ _CURRENT_SQL = """
 
 
 async def build_overview(session: AsyncSession) -> dict:
-    """5쿼리 수집 → compose_overview 판정 → 응답 dict."""
+    """4쿼리 수집 → compose_overview 판정 → 응답 dict."""
     now_kst = datetime.now(KST)
     kst_midnight = now_kst.replace(hour=0, minute=0, second=0, microsecond=0)
-    deep_enabled = settings.ai is not None and settings.ai.deep is not None
 
     stage_rows = (
         await session.execute(text(_STAGE_STATS_SQL), {"kst_midnight": kst_midnight})
     ).all()
-    split_rows = (
-        await session.execute(
-            text(_SUMMARIZE_SPLIT_SQL),
-            {"kst_midnight": kst_midnight, "macbook_alias": _MACBOOK_MODEL_ALIAS},
-        )
-    ).all()
     inflow_rows = (await session.execute(text(_TREND_INFLOW_SQL))).all()
     done_rows = (await session.execute(text(_TREND_DONE_SQL))).all()
     current_result = (await session.execute(text(_CURRENT_SQL))).all()
@@ -414,11 +338,9 @@ async def build_overview(session: AsyncSession) -> dict:
 
     result = compose_overview(
         rows_to_stage_stats(stage_rows),
-        rows_to_summarize_split(split_rows),
         {row[0]: int(row[1]) for row in inflow_rows},
         {row[0]: int(row[1]) for row in done_rows},
         current_rows,
-        deep_enabled=deep_enabled,
         now_kst=now_kst,
     )
     # 큐 밖 관리 스크립트(백필 등) = background_jobs (migration 357). 테이블 부재 시 graceful([]).
@@ -426,13 +348,13 @@ async def build_overview(session: AsyncSession) -> dict:
     return result
 
 
-# kind -> 처리 머신 (보드 머신 카드 귀속용). 미상 kind = gpu(오케스트레이션 호스트).
+# kind -> 처리 머신 (보드 머신 카드 귀속용). 미상 kind = nas(오케스트레이션 호스트).
 _BG_JOB_MACHINE = {
     "global_digest": "macmini",
     "morning_briefing": "macmini",
     "section_summary": "macmini",
-    "hier_backfill": "gpu",
-    "hier_redecompose": "gpu",
+    "hier_backfill": "nas",
+    "hier_redecompose": "nas",
 }
 
 
@@ -466,7 +388,7 @@ async def _fetch_background_jobs(session: AsyncSession) -> list[dict]:
             "processed": int(r["processed"] or 0), "total": r["total"],
             "elapsed_sec": int(r["elapsed_sec"] or 0), "stale": bool(r["stale"]),
             "error": r["error"],
-            "machine": _BG_JOB_MACHINE.get(r["kind"], "gpu"),
+            "machine": _BG_JOB_MACHINE.get(r["kind"], "nas"),
         }
         for r in rows
     ]

app/services/search/rerank_service.py

@@ -17,6 +17,7 @@ snippet 생성:
 from __future__ import annotations
 
 import asyncio
+import os
 import re
 from typing import TYPE_CHECKING
 
@@ -33,8 +34,11 @@ logger = setup_logger("rerank")
 # 동시 rerank 호출 제한 (GPU saturation 방지)
 RERANK_SEMAPHORE = asyncio.Semaphore(2)
 
-# rerank input 크기 제한 (latency / VRAM hard cap)
-MAX_RERANK_INPUT = 200
+# rerank input 크기 제한 (latency / VRAM hard cap).
+# 2노드 이관(2026-07-02): env MAX_RERANK_INPUT 로 조정 가능 — 맥미니 llama.cpp 리랭크는
+# 후보 수에 선형(NAS발 실측 50=0.60s / 100=0.95s / 200=1.89s)이라 NAS 배포는 50 권장.
+# 기본 200 = 현행(GPU TEI) 무회귀.
+MAX_RERANK_INPUT = int(os.getenv("MAX_RERANK_INPUT", "200"))
 MAX_CHUNKS_PER_DOC = 2
 
 # Soft timeout (초)

app/services/search/retrieval_service.py

@@ -76,10 +76,15 @@ class AxisFilter:
     jurisdiction: str | None = None
     year_from: int | None = None
     year_to: int | None = None
+    domain_buckets: list[str] | None = None    # 377: domain_bucket = ANY (도메인 스코프)
+    exclude_buckets: list[str] | None = None    # 377: domain_bucket <> ALL (예: News 제외)
+    cloud_egress: bool = False    # 갭2: 클라우드 소비자 cloud-eligibility allowlist 강제(토큰 claim 유래)
 
     def active(self) -> bool:
         return bool(self.material_types or self.jurisdiction
-                    or self.year_from is not None or self.year_to is not None)
+                    or self.year_from is not None or self.year_to is not None
+                    or self.domain_buckets or self.exclude_buckets
+                    or self.cloud_egress)
 
 
 def _axis_sql(alias: str, af: "AxisFilter | None", params: dict) -> str:
@@ -104,6 +109,22 @@ def _axis_sql(alias: str, af: "AxisFilter | None", params: dict) -> str:
     if af.year_to is not None:
         cl.append(f"COALESCE({p}published_date, {p}created_at::date) <= make_date(:af_yt, 12, 31)")
         params["af_yt"] = af.year_to
+    if af.domain_buckets:
+        cl.append(f"{p}domain_bucket = ANY(:af_db)")
+        params["af_db"] = af.domain_buckets
+    if af.exclude_buckets:
+        cl.append(f"{p}domain_bucket <> ALL(:af_xdb)")
+        params["af_xdb"] = af.exclude_buckets
+    if af.cloud_egress:
+        # 갭2 클라우드 egress allowlist(default-deny). restricted 는 _license_sql 별도 차단.
+        cl.append(
+            f"({p}data_origin = 'external' OR ("
+            f"{p}data_origin = 'work' "
+            f"AND {p}domain_bucket IN ('Engineering','Safety','Law') "
+            f"AND ({p}source_channel IS NULL OR {p}source_channel::text NOT IN ('voice','chat','memo')) "
+            f"AND {p}category::text IS DISTINCT FROM 'memo' "
+            f"AND ({p}user_note IS NULL OR {p}user_note = '')))"
+        )
     return " AND " + " AND ".join(cl)
 
 
@@ -121,7 +142,21 @@ def _license_sql(alias: str) -> str:
     술어 정의 = license_filter.restricted_exclude_sql 공유(digest/briefing/study 풀이와 단일 source).
     """
     from services.search.license_filter import restricted_exclude_sql
-    return " AND " + restricted_exclude_sql(alias)
+    _p = (alias + ".") if alias else ""
+    # ASME clause-KB(379): clause docs (doc_kind='clause') = read/nav/backlink only, excluded from retrieval/digest legs.
+    return " AND " + restricted_exclude_sql(alias) + f" AND {_p}doc_kind = 'standard'"
+
+
+def cloud_eligible_doc_sql(alias: str = "") -> str:
+    """단일 문서가 cloud 소비자(예: Claude/MCP)에게 노출 가능한가 = search retrieval 과
+    동일한 egress allowlist(갭2) + license 제한(B-4) 결합 술어. fetch_document(cloud) 가
+    search 와 byte-동일 게이트를 공유하도록 단일 source([[feedback_structural_integrity_over_path_discipline]]).
+
+    cloud_egress·license leg 모두 bind 파라미터 없는 리터럴 술어라 호출측 추가 params 불요.
+    주의: _license_sql 은 소유자 단건 다운로드엔 미적용(a안)이지만, cloud 노출은 구매 전자책
+    verbatim 누출을 막아야 하므로 여기선 항상 적용 = search 와 동일(local 토큰은 이 게이트 미발동).
+    반환 ' AND (egress allowlist) AND (license)' (alias='' = 컬럼 직접 참조). default-deny."""
+    return _axis_sql(alias, AxisFilter(cloud_egress=True), {}) + _license_sql(alias)
 
 
 # 2단계 gate (R2-B1) — SQL string interpolation 직전 final allowlist.

app/services/study/concept_curriculum.py

@@ -0,0 +1,284 @@
+"""concept_curriculum — 이론공부 홈 재료 (오늘의 개념 · 진도 · 회독 SR).
+
+개념문서 = documents (user_tags = @library/{topic}/{과목}/... , 가스기사). is_read = 회독,
+md_content 의 ★ 개수 = 빈출 tier(★★★=3 / ★★=2 / else 1). 회독 SR = study_concept_progress
++ sr_schedule(문제 SR 공용 산술). 읽기 전용 집계 + mark_read(회독+SR 입고)만 write. LLM 0.
+
+문제풀이 표면 무접촉 — 여기서 읽는 study_question_progress 는 '문항 due 카운트'만(홈 표시용).
+"""
+
+from __future__ import annotations
+
+from datetime import datetime, timezone
+
+from sqlalchemy import func, or_, select, text
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from models.document_read import DocumentRead
+from models.study_concept_progress import StudyConceptProgress
+from models.study_question_progress import StudyQuestionProgress
+from models.study_topic import StudyTopic
+from services.study.concept_parser import parse_concept, resolve_related
+from services.study.sr_schedule import advance, first_due
+
+# 개념 행 조회 — 태그로 개념문서 필터 + 회독 진행 LEFT JOIN. md_content 는 전송 안 하고
+# ★ 유무만 서버측 boolean 으로(홈이 자주 호출돼도 페이로드 최소).
+# is_read = document_reads(회독 정본, is_read 컬럼 아님) EXISTS. library unread 와 동일 기준.
+_CONCEPT_ROWS_SQL = text(
+    """
+    SELECT d.id AS doc_id,
+           d.title AS title,
+           EXISTS (
+             SELECT 1 FROM document_reads r
+             WHERE r.document_id = d.id AND r.user_id = :uid
+           ) AS is_read,
+           (d.md_content LIKE '%★★★%') AS f3,
+           (d.md_content LIKE '%★★%')  AS f2,
+           split_part(replace(d.user_tags::text, '"', ''), '/', 3) AS subject,
+           p.review_stage AS review_stage,
+           p.due_at AS due_at,
+           p.last_read_at AS last_read_at
+    FROM documents d
+    LEFT JOIN study_concept_progress p
+      ON p.concept_doc_id = d.id AND p.user_id = :uid
+    WHERE d.user_tags::text LIKE :like
+      AND d.deleted_at IS NULL
+    """
+)
+
+
+async def _topic_name(session: AsyncSession, topic_id: int) -> str | None:
+    return (
+        await session.execute(select(StudyTopic.name).where(StudyTopic.id == topic_id))
+    ).scalar_one_or_none()
+
+
+async def _concept_rows(session: AsyncSession, user_id: int, topic_name: str):
+    like = f"%@library/{topic_name}/%"
+    return (
+        await session.execute(_CONCEPT_ROWS_SQL, {"uid": user_id, "like": like})
+    ).mappings().all()
+
+
+def _freq(row) -> int:
+    if row["f3"]:
+        return 3
+    if row["f2"]:
+        return 2
+    return 1
+
+
+def _is_due(row, now: datetime) -> bool:
+    return (
+        row["due_at"] is not None
+        and row["due_at"] <= now
+        and (row["review_stage"] or 0) < 4
+    )
+
+
+def _item(row) -> dict:
+    return {
+        "doc_id": row["doc_id"],
+        "title": row["title"],
+        "subject": row["subject"],
+        "freq": _freq(row),
+        "review_stage": row["review_stage"],
+        "due_at": row["due_at"],
+    }
+
+
+async def _question_due_count(session: AsyncSession, user_id: int, topic_id: int, now: datetime) -> int:
+    """문항 복습 due (기존 study_question_progress 엔진 재사용, 홈 표시용)."""
+    return (
+        await session.execute(
+            select(func.count())
+            .select_from(StudyQuestionProgress)
+            .where(
+                StudyQuestionProgress.user_id == user_id,
+                StudyQuestionProgress.study_topic_id == topic_id,
+                StudyQuestionProgress.due_at.is_not(None),
+                StudyQuestionProgress.due_at <= now,
+                or_(
+                    StudyQuestionProgress.review_stage.is_(None),
+                    StudyQuestionProgress.review_stage < 4,
+                ),
+            )
+        )
+    ).scalar_one()
+
+
+async def curriculum(session: AsyncSession, user_id: int, topic_id: int) -> dict:
+    """과목별 회독 진도 + 개념/문항 복습 due 요약 (진도 대시보드)."""
+    name = await _topic_name(session, topic_id)
+    rows = await _concept_rows(session, user_id, name) if name else []
+    now = datetime.now(timezone.utc)
+
+    subj: dict[str, dict] = {}
+    for r in rows:
+        s = subj.setdefault(r["subject"], {"subject": r["subject"], "total": 0, "read": 0})
+        s["total"] += 1
+        if r["is_read"]:
+            s["read"] += 1
+
+    total = len(rows)
+    read = sum(1 for r in rows if r["is_read"])
+    concept_due = sum(1 for r in rows if _is_due(r, now))
+    question_due = await _question_due_count(session, user_id, topic_id, now)
+
+    return {
+        "topic_id": topic_id,
+        "topic_name": name,
+        "subjects": sorted(subj.values(), key=lambda x: x["subject"]),
+        "total": total,
+        "read": read,
+        "concept_due": concept_due,
+        "question_due": question_due,
+    }
+
+
+async def today_concepts(
+    session: AsyncSession, user_id: int, topic_id: int, limit: int = 6
+) -> dict:
+    """오늘 공부할 개념 = 재복습(SR due) 먼저 → 미독(빈출 우선). 졸업/재복습대기 제외."""
+    name = await _topic_name(session, topic_id)
+    rows = await _concept_rows(session, user_id, name) if name else []
+    now = datetime.now(timezone.utc)
+
+    due = [r for r in rows if _is_due(r, now)]
+    due.sort(key=lambda r: r["due_at"])
+
+    # 미독 & 아직 SR 큐 진입 전(due_at NULL) → 빈출 높은 순
+    unread = [r for r in rows if not r["is_read"] and r["due_at"] is None]
+    unread.sort(key=lambda r: (-_freq(r), r["subject"], r["title"]))
+
+    picked = [{**_item(r), "reason": "재복습"} for r in due]
+    picked += [{**_item(r), "reason": "신규"} for r in unread]
+
+    return {
+        "concepts": picked[:limit],
+        "due_total": len(due),
+        "unread_total": len(unread),
+    }
+
+
+async def mark_read(
+    session: AsyncSession, user_id: int, topic_id: int, doc_id: int, now: datetime | None = None
+) -> dict:
+    """개념 회독 처리 = document_reads(+1) + 회독 SR 입고/전진.
+
+    회독 정본 = document_reads(append-only), documents.is_read 컬럼 아님(library unread 와 정합).
+    첫 회독 → first_due(stage 0, 내일). 이후 회독은 'due 도래(due_at<=now)' 때만 correct 로 전진
+    (이른 재열람/다중클릭 과전진 방지). stage 4 졸업 후엔 due_at NULL 이라 전진 없음.
+    """
+    now = now or datetime.now(timezone.utc)
+
+    # 회독 로그 append (+1) — 사용자 명시 회독. 자동 아님(엔드포인트 = 명시 POST).
+    session.add(DocumentRead(user_id=user_id, document_id=doc_id, read_at=now))
+
+    prog = (
+        await session.execute(
+            select(StudyConceptProgress).where(
+                StudyConceptProgress.user_id == user_id,
+                StudyConceptProgress.concept_doc_id == doc_id,
+            )
+        )
+    ).scalar_one_or_none()
+
+    if prog is None:
+        stage, due = first_due(now)
+        prog = StudyConceptProgress(
+            user_id=user_id,
+            study_topic_id=topic_id,
+            concept_doc_id=doc_id,
+            review_stage=stage,
+            due_at=due,
+            last_read_at=now,
+        )
+        session.add(prog)
+    else:
+        # due 도래 시에만 전진 — 미래 due(재열람 이른 클릭)는 stage 불변, last_read_at 만 갱신.
+        if prog.due_at is not None and prog.due_at <= now:
+            res = advance(prog.review_stage, "correct", now)
+            if res is not None:
+                prog.review_stage, prog.due_at = res
+        prog.last_read_at = now
+
+    await session.commit()
+    await session.refresh(prog)
+    return {"ok": True, "review_stage": prog.review_stage, "due_at": prog.due_at}
+
+
+_CONCEPT_ONE_SQL = text(
+    """
+    SELECT d.id AS doc_id, d.title AS title, d.md_content AS md_content,
+           split_part(replace(d.user_tags::text, '"', ''), '/', 3) AS subject,
+           (d.md_content LIKE '%★★★%') AS f3,
+           (d.md_content LIKE '%★★%')  AS f2,
+           EXISTS (
+             SELECT 1 FROM document_reads r
+             WHERE r.document_id = d.id AND r.user_id = :uid
+           ) AS is_read,
+           p.review_stage AS review_stage,
+           p.due_at AS due_at
+    FROM documents d
+    LEFT JOIN study_concept_progress p ON p.concept_doc_id = d.id AND p.user_id = :uid
+    WHERE d.id = :doc_id AND d.deleted_at IS NULL AND d.user_tags::text LIKE :like
+    """
+)
+
+
+async def concept_detail(
+    session: AsyncSession, user_id: int, topic_id: int, doc_id: int
+) -> dict | None:
+    """개념 리더 재료 — md 구조 파싱 + 관련개념 백링크 해소 + 회독/SR 상태 + 같은 과목 이전/다음."""
+    name = await _topic_name(session, topic_id)
+    if not name:
+        return None
+    like = f"%@library/{name}/%"
+    row = (
+        await session.execute(
+            _CONCEPT_ONE_SQL, {"uid": user_id, "doc_id": doc_id, "like": like}
+        )
+    ).mappings().first()
+    if row is None:
+        return None
+
+    parsed = parse_concept(row["md_content"] or "")
+
+    # 백링크 해소 + 이전/다음 = 같은 토픽 개념 title 인덱스(회독 rows 재사용)
+    idx = await _concept_rows(session, user_id, name)
+    title_index = [(r["doc_id"], r["title"], r["subject"]) for r in idx]
+    resolved = resolve_related(parsed["related"], title_index)
+
+    # 이전/다음 = 같은 과목, title 순
+    same = sorted(
+        [(r["doc_id"], r["title"]) for r in idx if r["subject"] == row["subject"]],
+        key=lambda x: (x[1] or "", x[0]),
+    )
+    ids = [d for d, _ in same]
+    prev_id = next_id = None
+    if doc_id in ids:
+        pos = ids.index(doc_id)
+        if pos > 0:
+            prev_id = ids[pos - 1]
+        if pos < len(ids) - 1:
+            next_id = ids[pos + 1]
+
+    freq = 3 if row["f3"] else (2 if row["f2"] else 1)
+
+    return {
+        "doc_id": row["doc_id"],
+        "db_title": row["title"],
+        "title": parsed["title"] or row["title"],
+        "subject": row["subject"],
+        "freq": freq,
+        "summary": parsed["summary"],
+        "body": parsed["body"],
+        "bincheol": parsed["bincheol"],
+        "related": resolved,
+        "is_read": row["is_read"],
+        "review_stage": row["review_stage"],
+        "due_at": row["due_at"],
+        "prev_id": prev_id,
+        "next_id": next_id,
+    }

app/services/study/concept_links.py

@@ -0,0 +1,139 @@
+"""concept_links — 이론↔문제 브리지 롤업 (Stage B).
+
+study_concept_links(개념 doc ↔ 기출문항, 임베딩 코사인) + study_question_progress(내 풀이상태)를
+조인해 (a) 개념별 관련 기출 + 내 정답률(related_questions), (b) 개념 약점 지도(weakness_map) 산출.
+읽기 전용 집계 · LLM 0. 링크 적재는 scripts/concept_links_backfill.sql(임베딩) 배치.
+정답률 = 링크된 문항 중 progress.last_outcome 기준(attempted=풀이이력 보유, correct=최근정답).
+"""
+
+from __future__ import annotations
+
+from sqlalchemy import text
+from sqlalchemy.ext.asyncio import AsyncSession
+
+_ACCURACY_WEAK_PCT = 60  # 정답률 < 60% = 약점(attempted>0 일 때만)
+
+_AGG_SQL = text(
+    """
+    SELECT count(*) AS linked,
+           count(pr.study_question_id) FILTER (WHERE pr.last_outcome IS NOT NULL) AS attempted,
+           count(*) FILTER (WHERE pr.last_outcome = 'correct') AS correct
+    FROM study_concept_links l
+    LEFT JOIN study_question_progress pr
+      ON pr.study_question_id = l.question_id AND pr.user_id = :uid
+    WHERE l.concept_doc_id = :doc_id AND l.link_source = 'embedding'
+    """
+)
+
+_QROWS_SQL = text(
+    """
+    SELECT q.id AS id, q.subject AS subject, q.exam_round AS exam_round,
+           q.exam_question_number AS qnum, l.score AS score,
+           pr.last_outcome AS last_outcome, pr.review_stage AS review_stage
+    FROM study_concept_links l
+    JOIN study_questions q ON q.id = l.question_id AND q.deleted_at IS NULL AND q.is_active
+    LEFT JOIN study_question_progress pr
+      ON pr.study_question_id = q.id AND pr.user_id = :uid
+    WHERE l.concept_doc_id = :doc_id AND l.link_source = 'embedding'
+    ORDER BY l.score DESC
+    LIMIT :limit
+    """
+)
+
+_WEAKNESS_SQL = text(
+    """
+    SELECT d.id AS doc_id, d.title AS title,
+           split_part(replace(d.user_tags::text, '"', ''), '/', 3) AS subject,
+           count(l.id) AS linked,
+           count(pr.study_question_id) FILTER (WHERE pr.last_outcome IS NOT NULL) AS attempted,
+           count(*) FILTER (WHERE pr.last_outcome = 'correct') AS correct
+    FROM documents d
+    JOIN study_concept_links l ON l.concept_doc_id = d.id AND l.link_source = 'embedding'
+    LEFT JOIN study_question_progress pr
+      ON pr.study_question_id = l.question_id AND pr.user_id = :uid
+    WHERE d.user_tags::text LIKE :like AND d.deleted_at IS NULL
+    GROUP BY d.id, d.title, subject
+    """
+)
+
+
+async def related_questions(
+    session: AsyncSession, user_id: int, doc_id: int, limit: int = 20
+) -> dict:
+    """개념 doc 의 관련 기출 + 내 정답률(전체 링크 기준 집계 + 상위 N 표시용)."""
+    agg = (
+        await session.execute(_AGG_SQL, {"uid": user_id, "doc_id": doc_id})
+    ).mappings().first()
+    rows = (
+        await session.execute(
+            _QROWS_SQL, {"uid": user_id, "doc_id": doc_id, "limit": limit}
+        )
+    ).mappings().all()
+
+    linked = (agg["linked"] if agg else 0) or 0
+    attempted = (agg["attempted"] if agg else 0) or 0
+    correct = (agg["correct"] if agg else 0) or 0
+    accuracy = round(100 * correct / attempted) if attempted else None
+
+    return {
+        "linked": linked,
+        "attempted": attempted,
+        "correct": correct,
+        "accuracy": accuracy,
+        "questions": [
+            {
+                "id": r["id"],
+                "subject": r["subject"],
+                "exam_round": r["exam_round"],
+                "qnum": r["qnum"],
+                "score": round(r["score"], 3) if r["score"] is not None else None,
+                "last_outcome": r["last_outcome"],
+                "review_stage": r["review_stage"],
+            }
+            for r in rows
+        ],
+    }
+
+
+async def weakness_map(
+    session: AsyncSession, user_id: int, topic_name: str, limit: int = 12
+) -> dict:
+    """개념 약점 지도 — 링크된 기출 정답률로 개념 채색. 약점(attempted>0·정답률<60%) 우선 정렬."""
+    like = f"%@library/{topic_name}/%"
+    rows = (
+        await session.execute(_WEAKNESS_SQL, {"uid": user_id, "like": like})
+    ).mappings().all()
+
+    concepts = []
+    for r in rows:
+        attempted = r["attempted"] or 0
+        correct = r["correct"] or 0
+        accuracy = round(100 * correct / attempted) if attempted else None
+        if accuracy is None:
+            state = "unattempted"
+        elif accuracy < _ACCURACY_WEAK_PCT:
+            state = "weak"
+        else:
+            state = "ok"
+        concepts.append(
+            {
+                "doc_id": r["doc_id"],
+                "title": r["title"],
+                "subject": r["subject"],
+                "linked": r["linked"] or 0,
+                "attempted": attempted,
+                "accuracy": accuracy,
+                "state": state,
+            }
+        )
+
+    # 약점 우선(정답률 오름차순) → 미평가는 뒤로. 홈 위젯용 상위 N.
+    weak = sorted(
+        [c for c in concepts if c["state"] == "weak"],
+        key=lambda c: (c["accuracy"], -c["attempted"], c["doc_id"]),
+    )
+    return {
+        "weak": weak[:limit],
+        "weak_total": len(weak),
+        "evaluated_total": sum(1 for c in concepts if c["state"] != "unattempted"),
+    }

app/services/study/concept_parser.py

@@ -0,0 +1,175 @@
+"""concept_parser — 개념노트 markdown 구조 파서 + 관련개념 백링크 해소 (이론 리더용).
+
+정찰 실측 불변식(273/273): 개념노트는 고정 골격을 100% 따름 —
+    # {H1 제목}                     (첫 줄, DB title 과 다른 표시용 제목)
+    > **한 줄 요약**: {요약}          (blockquote, 라벨 고정)
+    ## {본문 라벨}  ...              (BODY, 자유 라벨 H2 0~N, 트레일 ★ 가능)
+    ## 빈출 포인트                    (항상, 관련개념 직전)
+    ## 관련 개념                      (항상, 문서 최종 섹션)
+
+코드펜스(``` ASCII 도식) 내부의 ##/- 는 무시. 헤딩 트레일 ★ 는 스트립(라벨 정규화).
+'빈출 포인트'/'관련 개념' 앵커만 이름으로 잡고 나머지 BODY 는 순서·위치로 처리(라벨 화이트리스트 금지).
+순수 함수 · LLM 0.
+"""
+
+from __future__ import annotations
+
+import re
+
+_FENCE = re.compile(r"^\s*```")
+_H1 = re.compile(r"^#\s+(.+?)\s*$")
+_H2 = re.compile(r"^##\s+(.+?)\s*$")  # ### 는 매칭 안 됨(## 뒤 \s 요구)
+_SUMMARY = re.compile(r"^>\s*\*\*한 줄 요약\*\*:\s*(.+)$")
+_STAR_SUFFIX = re.compile(r"\s*★+\s*$")
+_TRAIL_STARS = re.compile(r"★+\s*$")
+_BINCHEOL_ITEM = re.compile(r"^\s*-\s+(★*)\s*(.+)$")
+_RELATED_ITEM = re.compile(r"^\s*-\s+(.+)$")
+_PAREN = re.compile(r"\s*\(.*$")  # 괄호부터 끝(clarifier 힌트 절단)
+_NUM_PREFIX = re.compile(r"^\d+_")
+_STRIP_SYM = re.compile(r"[\s_·,./()\-]")
+
+_ANCHOR_BINCHEOL = "빈출 포인트"
+_ANCHOR_RELATED = "관련 개념"
+
+
+def parse_concept(md: str) -> dict:
+    """개념노트 md → {title, summary, body[{label,stars,md}], bincheol[{tier,text}], related[{raw,phrase,hint}]}."""
+    lines = (md or "").split("\n")
+    title: str | None = None
+    summary: str | None = None
+    body: list[dict] = []
+    bincheol_lines: list[str] = []
+    related_lines: list[str] = []
+
+    in_fence = False
+    zone = "pre"  # pre | body | bincheol | related
+    body_cur: dict | None = None
+
+    def emit(line: str) -> None:
+        if body_cur is not None:
+            body_cur["_lines"].append(line)
+        elif zone == "bincheol":
+            bincheol_lines.append(line)
+        elif zone == "related":
+            related_lines.append(line)
+        # pre-zone 내용(요약 앞 잡음)은 버림
+
+    for ln in lines:
+        if _FENCE.match(ln):
+            in_fence = not in_fence
+            emit(ln)
+            continue
+        if in_fence:
+            emit(ln)
+            continue
+
+        if title is None:
+            m = _H1.match(ln)
+            if m:
+                title = m.group(1).strip()
+                continue
+        if summary is None:
+            m = _SUMMARY.match(ln)
+            if m:
+                summary = m.group(1).strip()
+                continue
+
+        m2 = _H2.match(ln)
+        if m2:
+            raw_label = m2.group(1).strip()
+            star_m = _TRAIL_STARS.search(raw_label)
+            stars = len(star_m.group(0).strip()) if star_m else 0
+            label = _STAR_SUFFIX.sub("", raw_label).strip()
+            if label == _ANCHOR_BINCHEOL:
+                zone = "bincheol"
+                body_cur = None
+                continue
+            if label == _ANCHOR_RELATED:
+                zone = "related"
+                body_cur = None
+                continue
+            body_cur = {"label": label, "stars": stars, "_lines": []}
+            body.append(body_cur)
+            zone = "body"
+            continue
+
+        emit(ln)
+
+    body_out = []
+    for s in body:
+        text = "\n".join(s["_lines"]).strip()
+        if text or s["label"]:
+            body_out.append({"label": s["label"], "stars": s["stars"], "md": text})
+
+    bincheol = []
+    for ln in bincheol_lines:
+        m = _BINCHEOL_ITEM.match(ln)
+        if m:
+            bincheol.append({"tier": len(m.group(1)), "text": m.group(2).strip()})
+
+    related = []
+    for ln in related_lines:
+        m = _RELATED_ITEM.match(ln)
+        if m:
+            raw = m.group(1).strip()
+            phrase = _PAREN.sub("", raw).strip()
+            hint = raw[len(phrase):].strip() if len(raw) > len(phrase) else ""
+            if phrase:
+                related.append({"raw": raw, "phrase": phrase, "hint": hint})
+
+    return {
+        "title": title,
+        "summary": summary,
+        "body": body_out,
+        "bincheol": bincheol,
+        "related": related,
+    }
+
+
+def _normalize(s: str) -> str:
+    """해소용 정규화: NN_ 접두 제거 → 소문자 → 공백/기호 제거. 영문은 lowercase 유지."""
+    s = _NUM_PREFIX.sub("", s or "")
+    s = s.lower()
+    s = _STRIP_SYM.sub("", s)
+    return s
+
+
+def resolve_related(related: list[dict], title_index: list[tuple]) -> list[dict]:
+    """관련개념 구절 → 개념 doc 해소. title_index = [(doc_id, title, subject), ...].
+
+    다단 fallback(정찰 ~79%): 정규화 exact → 양방향 substring(≥2자 가드) → 미해소=dangling(doc_id None).
+    """
+    norm_exact: dict[str, int] = {}
+    norm_list: list[tuple[str, int, str]] = []
+    for did, ttl, _subj in title_index:
+        n = _normalize(ttl)
+        if n:
+            norm_exact.setdefault(n, did)
+            norm_list.append((n, did, ttl))
+
+    out = []
+    for it in related:
+        pn = _normalize(it["phrase"])
+        did: int | None = None
+        rtitle: str | None = None
+        if pn and len(pn) >= 2:
+            if pn in norm_exact:
+                did = norm_exact[pn]
+            else:
+                # substring 폴백: title-norm ⊆ phrase-norm 방향만(짧은 phrase 가 더 큰 title 을
+                # 삼키는 오결선 방지, 예: '염산'→'염산나트륨' X) + 길이차 최소(가장 구체적) +
+                # doc_id tiebreak(순서 무관 결정성). 후보 없으면 dangling(doc_id None).
+                cands = [
+                    (abs(len(n) - len(pn)), cand, ttl)
+                    for n, cand, ttl in norm_list
+                    if len(n) >= 2 and n in pn
+                ]
+                if cands:
+                    cands.sort(key=lambda c: (c[0], c[1]))
+                    _, did, rtitle = cands[0]
+        if did is not None and rtitle is None:
+            rtitle = next((t for d, t, _ in title_index if d == did), None)
+        out.append(
+            {"phrase": it["phrase"], "hint": it["hint"], "doc_id": did, "title": rtitle}
+        )
+    return out

app/services/study/publish_enqueue.py

@@ -68,10 +68,10 @@ async def enqueue_question_publish(session: AsyncSession, q: Any) -> None:
         await enqueue_publish(session, kind=KIND_EXPLANATION, source_id=q.id, payload=expl)
 
 
-async def backfill_publish_questions(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> int:
+async def backfill_publish_questions(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> tuple[int, int]:
     """active(미삭제) 문항을 id>after_id 부터 bounded 로 outbox 적재.
 
-    반환 = enqueue 한 문항 수(0 이면 끝). 큰 셋은 마지막 id 로 페이지 반복. caller commit.
+    반환 = (enqueue 수, 마지막 처리 id). caller 는 수==limit 면 last_id 로 다음 페이지. caller commit.
     """
     rows = (
         await session.execute(
@@ -83,7 +83,7 @@ async def backfill_publish_questions(session: AsyncSession, *, after_id: int = 0
     ).scalars().all()
     for q in rows:
         await enqueue_question_publish(session, q)
-    return len(rows)
+    return len(rows), (rows[-1].id if rows else after_id)
 
 
 async def enqueue_topic_publish(session: AsyncSession, topic: Any) -> None:
@@ -91,10 +91,10 @@ async def enqueue_topic_publish(session: AsyncSession, topic: Any) -> None:
     await enqueue_publish(session, kind=KIND_TOPIC, source_id=topic.id, payload=project_topic(topic))
 
 
-async def backfill_publish_topics(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> int:
+async def backfill_publish_topics(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> tuple[int, int]:
     """active(미삭제) 주제를 id>after_id 부터 bounded 로 outbox 적재(S-1 초기 백필).
 
-    반환 = enqueue 한 주제 수(0 이면 끝). 큰 셋은 마지막 id 로 페이지 반복. caller commit.
+    반환 = (enqueue 수, 마지막 처리 id). caller 는 수==limit 면 last_id 로 다음 페이지. caller commit.
     멱등 = 발행 워커의 (payload_hash, deleted) 디둡이 no-op 재투영 흡수(중복 enqueue 무해).
     """
     rows = (
@@ -107,7 +107,7 @@ async def backfill_publish_topics(session: AsyncSession, *, after_id: int = 0, l
     ).scalars().all()
     for t in rows:
         await enqueue_topic_publish(session, t)
-    return len(rows)
+    return len(rows), (rows[-1].id if rows else after_id)
 
 
 async def enqueue_card_publish(session: AsyncSession, card: Any) -> None:
@@ -123,10 +123,10 @@ async def enqueue_card_publish(session: AsyncSession, card: Any) -> None:
         await enqueue_publish(session, kind=KIND_CARD, source_id=card.id, payload=project_card(card))
 
 
-async def backfill_publish_cards(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> int:
+async def backfill_publish_cards(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> tuple[int, int]:
     """검수완료(needs_review=False)·미삭제 카드를 id>after_id 부터 bounded 로 outbox 적재(S-2 초기 백필).
 
-    반환 = enqueue 한 카드 수(0 이면 끝). 멱등 = 워커 (payload_hash, deleted) 디둡. caller commit.
+    반환 = (enqueue 수, 마지막 처리 id). caller 는 수==limit 면 last_id 로 다음 페이지. 멱등 = 워커 디둡. caller commit.
     """
     rows = (
         await session.execute(
@@ -142,7 +142,7 @@ async def backfill_publish_cards(session: AsyncSession, *, after_id: int = 0, li
     ).scalars().all()
     for c in rows:
         await enqueue_card_publish(session, c)
-    return len(rows)
+    return len(rows), (rows[-1].id if rows else after_id)
 
 
 async def enqueue_card_progress_publish(session: AsyncSession, progress: Any) -> None:
@@ -155,11 +155,11 @@ async def enqueue_card_progress_publish(session: AsyncSession, progress: Any) ->
     )
 
 
-async def backfill_publish_card_progress(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> int:
+async def backfill_publish_card_progress(session: AsyncSession, *, after_id: int = 0, limit: int = 200) -> tuple[int, int]:
     """모든 card progress row 를 id>after_id 부터 bounded 로 outbox 적재(S-4 초기 백필).
 
     ★필터 없음 = ALL row(due_at NULL sentinel·terminal 포함) — due-only 백필은 sentinel 누락.
-    반환 = enqueue 한 row 수(0 이면 끝). 멱등 = 워커 디둡. caller commit.
+    반환 = (enqueue 수, 마지막 처리 id). caller 는 수==limit 면 last_id 로 다음 페이지. 멱등 = 워커 디둡. caller commit.
     """
     rows = (
         await session.execute(
@@ -171,4 +171,4 @@ async def backfill_publish_card_progress(session: AsyncSession, *, after_id: int
     ).scalars().all()
     for p in rows:
         await enqueue_card_progress_publish(session, p)
-    return len(rows)
+    return len(rows), (rows[-1].id if rows else after_id)

app/services/summarize_units.py

@@ -0,0 +1,224 @@
+"""summarize_units — 거대문서 요약 전용 분할(map-reduce 유닛) 순수함수 (presegment PR1).
+
+plan ds-presegment-mapreduce-2 (2026-06-29 설계 합의 · PR0 실측 봉인):
+  - CAP_TOKENS = 12,000 tok/unit — greedy-pack 상한 (PR0: giant 236건 실측 캘리브레이션)
+  - TRIGGER_TOKENS = 25,000 tok — 이하는 단일콜 유지, 초과 시 map-reduce
+  - 3-way over% 게이트 (단독 CAP 초과 섹션의 토큰 비중. 헤딩 개수는 무의미 — ASME 1,494개):
+      over% == 0        → 'auto'   (TIER1: 로컬 자동 분할, PR0 실측 78%)
+      0 < over% <= 40   → 'hybrid' (패킹분 로컬 + 초과 섹션만 클로드, 8%)
+      over% > 40        → 'whole'  (TIER2: 클로드 전체 분할, 14%)
+  - 토큰 추정 = PR0 실 Qwen 토크나이저 캘리브레이션: 한글 0.529 tok/char · 기타 0.217.
+    구 휴리스틱(0.625/0.25)은 ~15% 과대라 폐기.
+
+불변식:
+  - 순수함수 — DB/네트워크/파일 접촉 0. 분할 = 요약 전용 아티팩트(문서 아님·검색/임베딩 미편입).
+  - leaf 추출 = hier_decomp.builder 재사용, leaf_hard_max=∞ 로 window-split 억제
+    (헤딩 leaf 만 — PR0 측정환경과 동일). 인접 섹션만 greedy-pack(순서 보존·중간 폐기 0
+    — 구 deep_summary 의 head/mid/tail 가운데 폐기 버그를 커버리지로 대체).
+  - 배선(deep_summary 분기·HOLD·클로드 알람)은 PR2/PR3 — 본 모듈은 계획만 산출.
+
+호출: plan_summarize_units(md_text) -> UnitPlan
+"""
+from __future__ import annotations
+
+import sys
+from dataclasses import dataclass, field
+
+# 상대 import — 컨테이너(services.*)와 repo-root 테스트(app.services.*) 양쪽에서 동작.
+# (구 `from app.services...` 절대 import 는 컨테이너에 app 패키지가 없어 ModuleNotFoundError —
+#  PR1 은 소비자 0 이라 잠복했던 버그, PR2 배선 시점에 수정.)
+from .hier_decomp.builder import HierNode, build_hier_tree
+
+CAP_TOKENS = 12_000
+TRIGGER_TOKENS = 25_000
+HYBRID_MAX_OVER_PCT = 40.0
+
+# PR0 실 Qwen tokenizer 캘리브레이션 (tok/char)
+KO_TOK_PER_CHAR = 0.529
+OTHER_TOK_PER_CHAR = 0.217
+
+_HANGUL_RANGES = (
+    (0xAC00, 0xD7A3),  # 완성형 음절
+    (0x1100, 0x11FF),  # 자모
+    (0x3130, 0x318F),  # 호환 자모
+)
+
+
+def _is_hangul(ch: str) -> bool:
+    cp = ord(ch)
+    return any(lo <= cp <= hi for lo, hi in _HANGUL_RANGES)
+
+
+def estimate_tokens(text: str) -> int:
+    """PR0 캘리브레이션 기반 토큰 추정 (한글 0.529 · 기타 0.217 tok/char)."""
+    if not text:
+        return 0
+    ko = sum(1 for ch in text if _is_hangul(ch))
+    other = len(text) - ko
+    return round(ko * KO_TOK_PER_CHAR + other * OTHER_TOK_PER_CHAR)
+
+
+@dataclass
+class SummarizeUnit:
+    """map-reduce 1유닛 — 인접 leaf 섹션들의 greedy-pack (요약 전용, 문서 아님)."""
+    index: int
+    section_titles: list[str | None] = field(default_factory=list)
+    text: str = ""
+    est_tokens: int = 0
+    over_cap: bool = False  # 단독 섹션이 CAP 초과 (hybrid 시 클로드 대상)
+
+
+@dataclass
+class UnitPlan:
+    mode: str                    # 'single' | 'map_reduce'
+    tier: str | None             # map_reduce 시 'auto' | 'hybrid' | 'whole'
+    total_est_tokens: int = 0
+    over_pct: float = 0.0
+    units: list[SummarizeUnit] = field(default_factory=list)
+
+
+def extract_leaves(md_text: str) -> list[HierNode]:
+    """헤딩 leaf 만 추출 — leaf_hard_max=∞ 로 window-split 억제 (PR0 측정환경 동일)."""
+    nodes = build_hier_tree(
+        md_text,
+        leaf_target_max=sys.maxsize,
+        leaf_hard_max=sys.maxsize,
+    )
+    return [n for n in nodes if n.is_leaf]
+
+
+def greedy_pack(leaves: list[HierNode], cap: int = CAP_TOKENS) -> list[SummarizeUnit]:
+    """인접 leaf 를 순서 보존하며 est_tokens<=cap 으로 pack. 단독 초과 leaf = 전용 유닛(over_cap)."""
+    units: list[SummarizeUnit] = []
+    cur_titles: list[str | None] = []
+    cur_texts: list[str] = []
+    cur_tokens = 0
+
+    def _flush() -> None:
+        nonlocal cur_titles, cur_texts, cur_tokens
+        if cur_texts:
+            units.append(SummarizeUnit(
+                index=len(units),
+                section_titles=cur_titles,
+                text="\n\n".join(cur_texts),
+                est_tokens=cur_tokens,
+            ))
+            cur_titles, cur_texts, cur_tokens = [], [], 0
+
+    for leaf in leaves:
+        t = estimate_tokens(leaf.text)
+        if t > cap:
+            _flush()
+            units.append(SummarizeUnit(
+                index=len(units),
+                section_titles=[leaf.section_title],
+                text=leaf.text,
+                est_tokens=t,
+                over_cap=True,
+            ))
+            continue
+        if cur_tokens + t > cap:
+            _flush()
+        cur_titles.append(leaf.section_title)
+        cur_texts.append(leaf.text)
+        cur_tokens += t
+    _flush()
+    return units
+
+
+def over_pct(leaves: list[HierNode], cap: int = CAP_TOKENS) -> float:
+    """단독 CAP 초과 섹션들의 토큰 비중(%) — 3-way 게이트 입력."""
+    total = 0
+    over = 0
+    for leaf in leaves:
+        t = estimate_tokens(leaf.text)
+        total += t
+        if t > cap:
+            over += t
+    if total == 0:
+        return 0.0
+    return over * 100.0 / total
+
+
+def gate(over: float) -> str:
+    """over% → tier. 0=auto / (0,40]=hybrid / >40=whole. 클로드 결과 재검증에도 재사용."""
+    if over <= 0.0:
+        return "auto"
+    if over <= HYBRID_MAX_OVER_PCT:
+        return "hybrid"
+    return "whole"
+
+
+def plan_summarize_units(
+    md_text: str, *,
+    cap: int = CAP_TOKENS,
+    trigger: int = TRIGGER_TOKENS,
+) -> UnitPlan:
+    """문서 → 요약 실행 계획. trigger 이하=single(현행 단일콜), 초과=map_reduce(tier+units)."""
+    total = estimate_tokens(md_text)
+    if total <= trigger:
+        return UnitPlan(mode="single", tier=None, total_est_tokens=total)
+    leaves = extract_leaves(md_text)
+    pct = over_pct(leaves, cap)
+    return UnitPlan(
+        mode="map_reduce",
+        tier=gate(pct),
+        total_est_tokens=total,
+        over_pct=round(pct, 2),
+        units=greedy_pack(leaves, cap),
+    )
+
+
+# ─── PR2 — map/reduce 프롬프트 조립 순수함수 (deep_summary_worker 가 소비) ───
+
+def render_map_slice(unit: SummarizeUnit, total_units: int) -> str:
+    """map 콜의 {original_text_slices} 대체 — 유닛 위치·섹션 라벨 + 본문."""
+    titles = " · ".join(t for t in unit.section_titles if t) or "(무제 구간)"
+    return f"[유닛 {unit.index + 1}/{total_units} — 섹션: {titles}]\n{unit.text}"
+
+
+def _format_unit_summary(res: dict, total_units: int) -> str:
+    """map 결과 1건 → reduce 입력 블록. res 키 = index/titles/tldr/detail/inconsistencies."""
+    titles = " · ".join(t for t in (res.get("titles") or []) if t) or "(무제 구간)"
+    lines = [f"[유닛 {int(res.get('index', 0)) + 1}/{total_units} — 섹션: {titles}]"]
+    if res.get("tldr"):
+        lines.append(f"TLDR: {res['tldr']}")
+    if res.get("detail"):
+        lines.append(str(res["detail"]))
+    for inc in res.get("inconsistencies") or []:
+        if isinstance(inc, dict):
+            lines.append(f"불일치({inc.get('kind', '')}): {inc.get('desc', '')}")
+    return "\n".join(lines)
+
+
+def build_reduce_units_block(
+    results: list[dict],
+    budget_tokens: int,
+    *,
+    min_detail_chars: int = 200,
+) -> tuple[str, bool]:
+    """reduce 입력 블록 조립 — budget_tokens 이하 보장(캡 초과 0 검증 게이트의 reduce 측).
+
+    초과 시 detail 만 비례 절단(라벨·TLDR·불일치 보전, 원문 순서 유지). 반환 (block, truncated).
+    """
+    total_units = len(results)
+    work = [dict(r) for r in results]
+    truncated = False
+    for _ in range(4):
+        block = "\n\n".join(_format_unit_summary(r, total_units) for r in work)
+        est = estimate_tokens(block)
+        if est <= budget_tokens:
+            return block, truncated
+        ratio = budget_tokens / est
+        for r in work:
+            detail = str(r.get("detail") or "")
+            keep = max(min_detail_chars, int(len(detail) * ratio * 0.9))
+            if len(detail) > keep:
+                r["detail"] = detail[:keep] + "…(절단)"
+                truncated = True
+    # 최후 방어 — 비례 절단이 floor(min_detail_chars)에 막히면 문자 하드 컷(KO 최악 비율 가정)
+    block = "\n\n".join(_format_unit_summary(r, total_units) for r in work)
+    if estimate_tokens(block) > budget_tokens:
+        block = block[: max(1, int(budget_tokens / KO_TOK_PER_CHAR))]
+        truncated = True
+    return block, truncated

app/workers/classify_worker.py

@@ -608,7 +608,9 @@ async def process(
             except Exception as exc:
                 if legacy_cfg is not None and is_deferrable_error(exc):
                     raise StageDeferred(f"macbook_unavailable:{type(exc).__name__}") from exc
-                logger.warning(f"[summary-fallback] id={document_id}: {exc}")
+                # ai_summary=NULL 로 완료되면 digest/briefing 이 조용히 제외 → ERROR 로 가시화
+                # (best-effort 강등 자체는 유지, 운영 추적성만 보강).
+                logger.error(f"[summary-fallback] id={document_id} ai_summary 미생성: {exc}")
 
     finally:
         await client.close()

app/workers/csb_collector.py

@@ -140,7 +140,8 @@ async def _download_pdf(url: str, dest: Path) -> int:
     if len(resp.content) > _MAX_PDF_BYTES:
         raise FeedError(f"PDF 크기 초과 ({len(resp.content)} bytes): {url}")
     dest.parent.mkdir(parents=True, exist_ok=True)
-    dest.write_bytes(resp.content)
+    # 최대 50MB PDF write 는 동기 blocking — 이벤트루프 점유 회피 to_thread (R5 동형).
+    await asyncio.to_thread(dest.write_bytes, resp.content)
     return len(resp.content)
 
 
@@ -190,9 +191,11 @@ async def _ingest_pdf(session, page_slug: str, pdf_url: str) -> bool:
 
     dest = Path(settings.nas_mount_path) / rel_path
     size = await _download_pdf(pdf_url, dest)
+    # 50MB PDF read + sha256 는 동기 blocking(I/O+CPU) — 이벤트루프 점유 회피 to_thread (R5 동형).
+    file_hash = await asyncio.to_thread(lambda: hashlib.sha256(dest.read_bytes()).hexdigest())
     doc = Document(
         file_path=rel_path,
-        file_hash=hashlib.sha256(dest.read_bytes()).hexdigest(),
+        file_hash=file_hash,
         file_format="pdf",
         file_size=size,
         file_type="immutable",

app/workers/deep_summary_worker.py

@@ -10,7 +10,9 @@ EscalationEnvelope + subject_domain 을 읽어, PR-A policy 템플릿 `p3c_deep_
 
 from __future__ import annotations
 
+import asyncio
 import json
+import os
 import time
 from datetime import datetime, timezone
 
@@ -29,10 +31,25 @@ from models.queue import ProcessingQueue, StageDeferred
 from policy.prompt_render import render_26b, policy_version as compute_policy_version
 from services.document_telemetry import record_analyze_event
 from services.search.llm_gate import Priority, acquire_mlx_gate
+from services.summarize_units import (
+    CAP_TOKENS,
+    UnitPlan,
+    build_reduce_units_block,
+    estimate_tokens,
+    plan_summarize_units,
+    render_map_slice,
+)
 
 logger = setup_logger("deep_summary_worker")
 
 DEEP_SUMMARY_TASK = "p3c_deep_summary"
+# presegment PR2 (plan ds-presegment-mapreduce-2) — 거대문서 map-reduce
+REDUCE_TASK = "p3c_deep_summary_reduce"
+# HYBRID/TIER2(클로드 유인 분할 필요) HOLD 재확인 간격. PR3(알람·경계 주입) 전까지는
+# 이 간격으로 재계획만 반복한다 — attempts 미소모(StageDeferred)라 영구 failed 없음.
+HOLD_RETRY_MINUTES = int(os.getenv("DEEP_SUMMARY_HOLD_RETRY_MINUTES", "1440"))
+# reduce 프롬프트 오버헤드가 비정상적으로 커도 유닛 블록 예산은 이 밑으로 안 내려감(방어).
+REDUCE_BUDGET_FLOOR_TOKENS = 1_000
 
 # inconsistencies kind 허용 목록 (feedback_document_server_domain_scope.md — 구매/계약 제외)
 ALLOWED_INCONSISTENCY_KINDS = {
@@ -94,6 +111,25 @@ async def process(
 
     envelope = EscalationEnvelope.from_json(json.dumps(envelope_raw))
 
+    # ─── presegment PR2 게이트 (plan ds-presegment-mapreduce-2) ───
+    # TRIGGER(25K tok) 이하 = 아래 기존 단일콜 경로 그대로(무회귀). 초과 시 3-way:
+    #   auto(over%==0)   → 로컬 map-reduce (유닛별 26B → reduce)
+    #   hybrid/whole     → HOLD(awaiting_split) — 맥미니 미전송, 클로드 유인 분할은 PR3
+    # 게이트/유닛은 전체 extracted_text 기준 — 단일콜의 head/mid/tail "가운데 폐기"를
+    # 전 유닛 커버리지로 대체한다. build_hier_tree 가 거대 md 에서 초 단위 CPU 라
+    # 이벤트루프 점유 회피 위해 to_thread (presegment_worker._read_toc 와 동일 패턴).
+    unit_plan = await asyncio.to_thread(plan_summarize_units, doc.extracted_text or "")
+    if unit_plan.mode == "map_reduce":
+        # units 빈 auto 는 이론상 불가(비어있지 않은 텍스트 = leaf >= 1)지만, 빈 reduce
+        # 단일콜(환각 위험)로 흐르지 않게 방어적으로 HOLD 로 보낸다.
+        if unit_plan.tier != "auto" or not unit_plan.units:
+            await _hold_awaiting_split(session, queue_row, unit_plan, document_id)
+        await _process_map_reduce(
+            doc, queue_row, envelope, subject_domain, unit_plan, session,
+            defer_on_deep_unavailable=defer_on_deep_unavailable,
+        )
+        return
+
     # 원문 슬라이스 추출 (envelope.original_pointers.text_ranges 기반)
     slices = _build_text_slices(doc.extracted_text or "", envelope.original_pointers)
 
@@ -214,6 +250,267 @@ async def process(
     )
 
 
+async def _hold_awaiting_split(
+    session: AsyncSession, queue_row: ProcessingQueue, plan: UnitPlan, document_id: int
+) -> None:
+    """HYBRID/TIER2 — 클로드 유인 분할 대기(HOLD). 맥미니 미전송, StageDeferred 보류.
+
+    payload.presegment.awaiting_split 마킹을 먼저 commit — StageDeferred 핸들러
+    (queue_consumer)는 새 세션에서 행을 다시 읽어 deferred_until 만 병합하므로 유실 없음.
+    알람(ntfy)·클로드 경계 주입은 PR3 — 그 전까지는 HOLD_RETRY_MINUTES 간격 재계획만 반복.
+    무인 자동 cloud 호출 금지 룰 준수(클로드 경로는 항상 유인 게이트).
+    """
+    payload = dict(queue_row.payload or {})
+    preseg = dict(payload.get("presegment") or {})
+    preseg.update({
+        "awaiting_split": True,
+        "tier": plan.tier,
+        "over_pct": plan.over_pct,
+        "total_est_tokens": plan.total_est_tokens,
+        "units": len(plan.units),
+        # 클로드가 분할해야 할 초과 섹션 표본 (PR3 알람 본문용)
+        "oversized_sections": [
+            (u.section_titles[0] if u.section_titles else None)
+            for u in plan.units if u.over_cap
+        ][:20],
+    })
+    payload["presegment"] = preseg
+    queue_row.payload = payload  # 재할당 = JSONB 변경 감지
+    await session.commit()
+    logger.info(
+        f"[deep] id={document_id} awaiting_split tier={plan.tier} over_pct={plan.over_pct} "
+        f"total_est_tokens={plan.total_est_tokens} units={len(plan.units)} "
+        f"→ HOLD ({HOLD_RETRY_MINUTES}분 후 재확인, 클로드 분할=PR3 유인)"
+    )
+    raise StageDeferred(
+        f"awaiting_split:{plan.tier}", retry_after_minutes=HOLD_RETRY_MINUTES
+    )
+
+
+async def _call_26b(
+    client: AIClient, prompt: str, *, defer_on_deep_unavailable: bool, document_id: int
+):
+    """map/reduce 공용 26B 호출 — 단일콜 경로와 동일한 deep 슬롯 우선 + fair-share 폴백.
+
+    반환 (raw, used_cfg). 맥북(deep) 불가 시 consumer 경로는 맥미니 primary 로 즉시
+    처리(동일 모델 — 강등 아님), drain 경로는 StageDeferred 전파(맥북 레버 시멘틱).
+    """
+    deep_cfg = client.ai.deep
+    if deep_cfg is not None:
+        try:
+            return await call_deep_or_defer(client, prompt), deep_cfg
+        except StageDeferred:
+            if defer_on_deep_unavailable:
+                raise
+            logger.info(f"[deep] id={document_id} 맥북 불가 → 맥미니 primary 처리 (fair-share)")
+    async with acquire_mlx_gate(Priority.BACKGROUND):
+        return await client.call_primary(prompt), settings.ai.primary
+
+
+def _parse_deep_output(raw: str) -> tuple[DeepSummaryOutput | None, str | None]:
+    """raw → DeepSummaryOutput. 단일콜 경로와 동일한 3단 파서. 실패 시 (None, parse_error)."""
+    try:
+        parsed = _parse_outermost_json(raw) or parse_json_response(raw)
+        if not parsed:
+            parsed = _regex_extract_fields(raw)
+        return DeepSummaryOutput.model_validate(parsed or {}), None
+    except (ValidationError, ValueError, TypeError) as exc:
+        return None, f"parse:{type(exc).__name__}"
+
+
+async def _process_map_reduce(
+    doc: Document,
+    queue_row: ProcessingQueue,
+    envelope: EscalationEnvelope,
+    subject_domain: str,
+    plan: UnitPlan,
+    session: AsyncSession,
+    *,
+    defer_on_deep_unavailable: bool,
+) -> None:
+    """TIER1 자동 — 유닛별 map(26B) → reduce(26B) → 단일콜과 동일 필드 기록.
+
+    멱등 재개: 성공 유닛은 payload.presegment.map_results 에 즉시 commit —
+    502/defer/재시작 후 재클레임 시 완료 유닛은 건너뛴다. 유닛 인덱스는
+    plan_summarize_units 가 같은 extracted_text 에 결정적이라 attempt 간 안정.
+    파싱 실패 유닛이 남으면 raise → queue_consumer 의 기존 attempts/백오프 재사용
+    (실패 유닛만 재호출되므로 재시도 비용 = 잔여 유닛뿐).
+    """
+    document_id = doc.id
+    units = plan.units
+    n = len(units)
+    payload = dict(queue_row.payload or {})
+    preseg = dict(payload.get("presegment") or {})
+    preseg.pop("awaiting_split", None)  # 재계획으로 auto 가 된 경우 HOLD 마킹 해제
+    map_results: dict = dict(preseg.get("map_results") or {})
+
+    logger.info(
+        f"[deep] id={document_id} map_reduce 시작 units={n} over_pct={plan.over_pct} "
+        f"total_est_tokens={plan.total_est_tokens} resume={len(map_results)}/{n}"
+    )
+
+    rendered = render_26b(DEEP_SUMMARY_TASK, subject_domain)
+    envelope_injection = envelope.to_system_injection()
+
+    client = AIClient()
+    start = time.perf_counter()
+    used_cfg = client.ai.deep or settings.ai.primary
+    failed_units: list[int] = []
+    try:
+        # ── map: 유닛별 26B (콜 사이마다 gate 를 놓아 짧은 인터랙티브 요청이 끼어든다) ──
+        for unit in units:
+            key = str(unit.index)
+            if key in map_results:
+                continue
+            prompt = (
+                rendered
+                .replace("{escalation_envelope_json}", envelope_injection)
+                .replace("{original_text_slices}", render_map_slice(unit, n))
+            )
+            # 검증 게이트 "모든 LLM 콜 캡 초과 0" 을 로그로 단정 가능하게 남긴다.
+            logger.info(
+                f"[deep] id={document_id} map {unit.index + 1}/{n} "
+                f"unit_tokens={unit.est_tokens} prompt_est_tokens={estimate_tokens(prompt)} "
+                f"cap={CAP_TOKENS}"
+            )
+            raw, used_cfg = await _call_26b(
+                client, prompt,
+                defer_on_deep_unavailable=defer_on_deep_unavailable,
+                document_id=document_id,
+            )
+            out, perr = _parse_deep_output(raw)
+            if out is None or not (out.detail or out.tldr):
+                # 실패 유닛은 persist 하지 않음 — 재시도가 이 유닛만 다시 호출한다.
+                failed_units.append(unit.index)
+                logger.warning(
+                    f"[deep] id={document_id} map {unit.index + 1}/{n} 결과 비었음/파싱 실패"
+                    f"({perr}) — 유닛 재시도 대상"
+                )
+                continue
+            # ★매 유닛 새 dict 로 재구성 (in-place 변경 금지) — 직전 commit 의 committed
+            # 스냅샷이 같은 중첩 객체를 참조하면 old==new 로 보여 SQLAlchemy 가 UPDATE 를
+            # 스킵한다(60254 라이브에서 unit 0 만 persist 된 aliasing 버그의 fix).
+            map_results = {
+                **map_results,
+                key: {
+                    "index": unit.index,
+                    "titles": [t for t in unit.section_titles if t][:8],
+                    "tldr": out.tldr,
+                    "detail": out.detail,
+                    "inconsistencies": _filter_inconsistencies(out.inconsistencies or []),
+                },
+            }
+            preseg = {
+                **preseg,
+                "tier": plan.tier,
+                "over_pct": plan.over_pct,
+                "total_est_tokens": plan.total_est_tokens,
+                "units": n,
+                "map_results": map_results,
+            }
+            payload = {**payload, "presegment": preseg}
+            queue_row.payload = payload  # 재할당 = JSONB 변경 감지
+            await session.commit()  # 유닛 단위 멱등 재개 지점
+
+        if failed_units:
+            raise ValueError(
+                f"map 유닛 {len(failed_units)}/{n}건 결과 없음 — 재시도 대상: {failed_units[:10]}"
+            )
+
+        # ── reduce: 요약들의 요약 1콜 (유닛 블록도 캡 이하로 절단 보장) ──
+        reduce_rendered = render_26b(REDUCE_TASK, subject_domain)
+        base_prompt = (
+            reduce_rendered
+            .replace("{escalation_envelope_json}", envelope_injection)
+            .replace("{unit_count}", str(n))
+        )
+        budget = max(
+            REDUCE_BUDGET_FLOOR_TOKENS, CAP_TOKENS - estimate_tokens(base_prompt)
+        )
+        ordered = [map_results[str(u.index)] for u in units]
+        block, reduce_truncated = build_reduce_units_block(ordered, budget)
+        reduce_prompt = base_prompt.replace("{unit_summaries}", block)
+        logger.info(
+            f"[deep] id={document_id} reduce units={n} "
+            f"prompt_est_tokens={estimate_tokens(reduce_prompt)} cap={CAP_TOKENS} "
+            f"truncated={reduce_truncated}"
+        )
+        raw, used_cfg = await _call_26b(
+            client, reduce_prompt,
+            defer_on_deep_unavailable=defer_on_deep_unavailable,
+            document_id=document_id,
+        )
+    except StageDeferred:
+        logger.info(
+            f"[deep] id={document_id} map_reduce 보류 — 완료 유닛 {len(map_results)}/{n} 보존"
+        )
+        raise
+    except Exception as exc:
+        # 단일콜 경로와 동일 — 호출 실패는 전파해 queue_consumer 가 재시도/dead-letter 처리.
+        logger.warning(f"[deep] id={document_id} map_reduce 실패: {exc}")
+        raise
+    finally:
+        await client.close()
+
+    latency_ms = int((time.perf_counter() - start) * 1000)
+    deep_out, parse_error = _parse_deep_output(raw)
+    if deep_out is None:
+        # 단일콜 경로와 동일 시멘틱 — doc 미기록(legacy 결과 보존), 이벤트로 가시화.
+        deep_out = DeepSummaryOutput()
+        logger.warning(f"[deep] id={document_id} reduce 파싱 실패 ({parse_error}) — doc 미기록")
+
+    if not parse_error:
+        doc.ai_detail_summary = (deep_out.detail or "").strip() or None
+        # 불일치 = reduce 출력 + map 유닛 합본 dedup — reduce 가 떨궈도 유닛 발견분 보전.
+        merged = _filter_inconsistencies(deep_out.inconsistencies or [])
+        seen = {(i["kind"], i["desc"]) for i in merged}
+        for res in ordered:
+            for inc in res.get("inconsistencies") or []:
+                k = (inc.get("kind"), inc.get("desc"))
+                if k not in seen:
+                    seen.add(k)
+                    merged.append(inc)
+        doc.ai_inconsistencies = merged
+        doc.ai_analysis_tier = "deep"
+        doc.ai_processed_at = datetime.now(timezone.utc)
+
+    try:
+        pv = compute_policy_version(REDUCE_TASK)
+    except Exception:
+        pv = None
+
+    await record_analyze_event(
+        doc_id=document_id,
+        user_id=None,
+        mode="summary_deep",
+        text_limit=used_cfg.context_char_limit or 260000,
+        truncated=reduce_truncated,
+        layers_returned=["detail_summary", "inconsistencies"] if not parse_error else [],
+        cached=False,
+        latency_ms=latency_ms,
+        model_name=used_cfg.model,
+        prompt_version=(f"{REDUCE_TASK}@{pv}" if pv else REDUCE_TASK),
+        error_code=parse_error,
+        source="document_server",
+        subject_domain=subject_domain,
+        risk_flags=list(envelope.risk_flags),
+        high_impact_task=None,
+        escalation_reasons=list(envelope.escalation_reasons),
+        confidence=deep_out.confidence,
+        policy_version=pv,
+        shadow_would_route_to="primary",
+        tier="primary",
+        escalated_to_26b=True,
+        suppressed_reason=None,
+    )
+
+    logger.info(
+        f"[deep] id={document_id} map_reduce 완료 units={n} "
+        f"detail_len={len(doc.ai_detail_summary or '')} inc={len(doc.ai_inconsistencies or [])} "
+        f"latency_ms={latency_ms} parse_error={parse_error}"
+    )
+
+
 def _build_text_slices(text: str, pointers: dict) -> str:
     """original_pointers.text_ranges 의 [{start, end}] 를 실제 본문 슬라이스로 합친다.
 

app/workers/extract_worker.py

@@ -110,6 +110,11 @@ def _get_pdf_page_count(
 
 async def _call_ocr(file_path: Path, is_image: bool, max_pages: int = 200) -> str | None:
     """OCR 서비스 호출 — 타임아웃 페이지 수 비례"""
+    if not settings.ocr_enabled:
+        # 2노드 이관(2026-07-02): GPU Surya 폐기 — 명시 비활성. None 반환 = 기존 soft-fail
+        # 의미론(호출자가 ocr_attempted/skip_reason 메타 기록). 스캔 문서는 비전 배치 경로 별도.
+        logger.warning("[ocr] OCR_ENABLED=false — skip (스캔·이미지 추출은 비전 배치 경로)")
+        return None
     container_path = f"/documents/{file_path.relative_to(Path(settings.nas_mount_path))}"
     timeout = 60 if is_image else min(600, max(120, max_pages * 3))
     try:

app/workers/file_watcher.py

@@ -251,104 +251,118 @@ async def watch_inbox():
     for extra_path in settings.additional_watch_targets:
         targets.append((extra_path, "library"))
 
-    async with async_session() as session:
-        # ─── Web/ 트랙 (devonagent) — DEVONthink Smart Rule 이 떨군 .html 만 진입 ───
-        if web_root.exists():
-            # rglob NFS 디렉토리 walk(blocking stat 다발)를 off-thread 로 수집 (R5).
-            for file_path in await asyncio.to_thread(lambda: list(web_root.rglob("*.html"))):
-                if not file_path.is_file() or should_skip(file_path):
-                    continue
-                rel_path = str(file_path.relative_to(nas_root))
-                added, _ = await _ingest_web_file(session, file_path, rel_path)
+    # 파일별 독립 세션+commit 으로 격리 — 한 파일 실패(예: rglob↔stat 사이 삭제로 FileNotFoundError,
+    # flush 오류)가 watch_inbox 전체를 raise·롤백해 그 사이클 등록분을 모두 잃거나, 결정적 poison
+    # 파일이 매 사이클 같은 지점에서 중단시키는 것을 차단 (news_collector/csb_collector 와 동형).
+    # ─── Web/ 트랙 (devonagent) — DEVONthink Smart Rule 이 떨군 .html 만 진입 ───
+    if web_root.exists():
+        # rglob NFS 디렉토리 walk(blocking stat 다발)를 off-thread 로 수집 (R5).
+        for file_path in await asyncio.to_thread(lambda: list(web_root.rglob("*.html"))):
+            if not file_path.is_file() or should_skip(file_path):
+                continue
+            rel_path = str(file_path.relative_to(nas_root))
+            try:
+                async with async_session() as session:
+                    added, _ = await _ingest_web_file(session, file_path, rel_path)
+                    await session.commit()
                 new_count += added
-
-        # ─── PKM 트랙 (기존 drive_sync) ─────────────────────────────────────────
-        for sub, expected_category in targets:
-            scan_root = pkm_root / sub
-            if not scan_root.exists():
+            except Exception as e:
+                logger.warning("[Web] 파일 처리 실패 skip path=%s: %s", rel_path, e)
                 continue
 
-            # 안전 자료실 A-2/B-4 — 타깃 폴더 기반 (material, jurisdiction, license)
-            target_mt, target_jur, target_license = _TARGET_AXIS.get(
-                Path(sub).name, (None, None, None)
+    # ─── PKM 트랙 (기존 drive_sync) ─────────────────────────────────────────
+    for sub, expected_category in targets:
+        scan_root = pkm_root / sub
+        if not scan_root.exists():
+            continue
+
+        # 안전 자료실 A-2/B-4 — 타깃 폴더 기반 (material, jurisdiction, license)
+        target_mt, target_jur, target_license = _TARGET_AXIS.get(
+            Path(sub).name, (None, None, None)
+        )
+
+        # NFS 디렉토리 walk(blocking) off-thread 수집 (R5).
+        for file_path in await asyncio.to_thread(lambda: list(scan_root.rglob("*"))):
+            if not file_path.is_file() or should_skip(file_path):
+                continue
+
+            category, needs_conversion, next_stage = _route_media(
+                file_path, expected_category
             )
 
-            # NFS 디렉토리 walk(blocking) off-thread 수집 (R5).
-            for file_path in await asyncio.to_thread(lambda: list(scan_root.rglob("*"))):
-                if not file_path.is_file() or should_skip(file_path):
-                    continue
+            # audio/video 폴더에 엉뚱한 확장자가 들어왔거나 Inbox 에
+            # audio/video 가 잘못 떨어진 경우 — 이 라운드에서 아예 skip
+            if category is None and next_stage is None:
+                continue
 
-                category, needs_conversion, next_stage = _route_media(
-                    file_path, expected_category
-                )
-
-                # audio/video 폴더에 엉뚱한 확장자가 들어왔거나 Inbox 에
-                # audio/video 가 잘못 떨어진 경우 — 이 라운드에서 아예 skip
-                if category is None and next_stage is None:
-                    continue
-
-                rel_path = str(file_path.relative_to(nas_root))
+            rel_path = str(file_path.relative_to(nas_root))
+            try:
                 # GB 파일 SHA-256 은 이벤트 루프를 점유 → 같은 루프의 모든 1분 주기 consumer
                 # + FastAPI 요청이 수십초~분 동시 정지. to_thread 오프로드. 스캔 루프가 이미
                 # 순차라 file_hash 는 한 번에 하나만 실행(직렬화) — 병렬 해싱 X = NFS 2.5GbE
-                # 대역폭·버퍼 메모리 blowup 방지 (R5).
+                # 대역폭·버퍼 메모리 blowup 방지 (R5). 세션 밖에서 계산(커넥션 미점유).
                 fhash = await asyncio.to_thread(file_hash, file_path)
 
-                result = await session.execute(
-                    select(Document).where(Document.file_path == rel_path)
-                )
-                existing = result.scalar_one_or_none()
-
-                if existing is None:
-                    ext = file_path.suffix.lstrip(".").lower() or "unknown"
-                    doc = Document(
-                        file_path=rel_path,
-                        file_hash=fhash,
-                        file_format=ext,
-                        file_size=file_path.stat().st_size,
-                        file_type="immutable",
-                        title=file_path.stem,
-                        source_channel="drive_sync",
-                        category=category,
-                        needs_conversion=needs_conversion,
-                        # 안전 자료실 A-2/B-4 — watch 타깃 매핑 (KGS=law/KR 등, 비대상=NULL)
-                        material_type=target_mt,
-                        jurisdiction=target_jur,
+                async with async_session() as session:
+                    result = await session.execute(
+                        select(Document).where(Document.file_path == rel_path)
                     )
-                    # B-4 — 타깃 폴더 license 주입(restricted 포함, 비대상=미주입). classify 는
-                    # material_type IS NULL 일 때만 제안 + extract_meta 미기록이라 주입 보존.
-                    if target_license:
-                        doc.extract_meta = {"license": dict(target_license)}
-                    session.add(doc)
-                    await session.flush()
+                    existing = result.scalar_one_or_none()
 
-                    if next_stage:
-                        await enqueue_stage(session, doc.id, next_stage)
-                    new_count += 1
+                    if existing is None:
+                        ext = file_path.suffix.lstrip(".").lower() or "unknown"
+                        doc = Document(
+                            file_path=rel_path,
+                            file_hash=fhash,
+                            file_format=ext,
+                            file_size=file_path.stat().st_size,
+                            file_type="immutable",
+                            title=file_path.stem,
+                            source_channel="drive_sync",
+                            category=category,
+                            needs_conversion=needs_conversion,
+                            # 안전 자료실 A-2/B-4 — watch 타깃 매핑 (KGS=law/KR 등, 비대상=NULL)
+                            material_type=target_mt,
+                            jurisdiction=target_jur,
+                        )
+                        # B-4 — 타깃 폴더 license 주입(restricted 포함, 비대상=미주입). classify 는
+                        # material_type IS NULL 일 때만 제안 + extract_meta 미기록이라 주입 보존.
+                        if target_license:
+                            doc.extract_meta = {"license": dict(target_license)}
+                        session.add(doc)
+                        await session.flush()
 
-                elif existing.file_hash != fhash:
-                    existing.file_hash = fhash
-                    existing.file_size = file_path.stat().st_size
-                    # 기존 문서에 category/quarantine flag 가 비어있으면 보정
-                    if existing.category is None and category is not None:
-                        existing.category = category
-                    if needs_conversion and not getattr(existing, "needs_conversion", False):
-                        existing.needs_conversion = True
-                    # B-4 — 축/license 보정(B-4 이전 적재분이 재변경 시): material 미설정 시 주입,
-                    # license 부재 시에만 merge 주입(clobber 회피 — 기존 extract_meta 키 보존).
-                    if existing.material_type is None and target_mt is not None:
-                        existing.material_type = target_mt
-                        existing.jurisdiction = target_jur
-                    if target_license and not (existing.extract_meta or {}).get("license"):
-                        meta = dict(existing.extract_meta or {})
-                        meta["license"] = dict(target_license)
-                        existing.extract_meta = meta
+                        if next_stage:
+                            await enqueue_stage(session, doc.id, next_stage)
+                        await session.commit()
+                        new_count += 1
 
-                    if next_stage:
-                        await enqueue_stage(session, existing.id, next_stage)
-                    changed_count += 1
+                    elif existing.file_hash != fhash:
+                        existing.file_hash = fhash
+                        existing.file_size = file_path.stat().st_size
+                        # 기존 문서에 category/quarantine flag 가 비어있으면 보정
+                        if existing.category is None and category is not None:
+                            existing.category = category
+                        if needs_conversion and not getattr(existing, "needs_conversion", False):
+                            existing.needs_conversion = True
+                        # B-4 — 축/license 보정(B-4 이전 적재분이 재변경 시): material 미설정 시 주입,
+                        # license 부재 시에만 merge 주입(clobber 회피 — 기존 extract_meta 키 보존).
+                        if existing.material_type is None and target_mt is not None:
+                            existing.material_type = target_mt
+                            existing.jurisdiction = target_jur
+                        if target_license and not (existing.extract_meta or {}).get("license"):
+                            meta = dict(existing.extract_meta or {})
+                            meta["license"] = dict(target_license)
+                            existing.extract_meta = meta
 
-        await session.commit()
+                        if next_stage:
+                            await enqueue_stage(session, existing.id, next_stage)
+                        await session.commit()
+                        changed_count += 1
+                    # else: 무변경 → 쓰기 없음 (세션 자동 닫힘, commit 불요)
+            except Exception as e:
+                logger.warning("[PKM] 파일 처리 실패 skip path=%s: %s", rel_path, e)
+                continue
 
     if new_count or changed_count:
         logger.info(f"[Inbox+§3] 새 파일 {new_count}건, 변경 파일 {changed_count}건 등록")

app/workers/marker_worker.py

@@ -300,6 +300,11 @@ async def _process_single(
             f"[marker] transient error id={document_id} kind={type(exc).__name__}: {exc}"
         )
         raise
+    except json.JSONDecodeError as exc:
+        # 200 응답의 truncated/malformed body — 연결 흔들림 등 transient. _fail(non-retryable)
+        # 로 박지 말고 raise → queue retry (max_attempts 까지). 진짜 손상이면 재시도 후 failed.
+        logger.warning(f"[marker] malformed json body (200) id={document_id}: {exc}")
+        raise
     except Exception as exc:
         logger.exception(f"[marker] unexpected error id={document_id}: {exc}")
         await _fail(session, document_id, str(exc)[:1000])

app/workers/presegment_worker.py

@@ -497,12 +497,18 @@ async def process(document_id: int, session: AsyncSession) -> None:
         logger.warning(f"[presegment] id={document_id} file not found ({raw}) → extract")
         return
 
+    import asyncio
+
     import fitz  # PyMuPDF — extract_worker/marker_worker 와 동일 의존
 
+    def _read_toc(path: str):
+        # fitz open/get_toc 는 동기 blocking — live 스테이지라 이벤트루프(같은 루프의 1분 consumer +
+        # FastAPI 요청) 점유 회피 위해 to_thread 오프로드(거대/손상 PDF 파싱 수백 ms~초).
+        with fitz.open(path) as pdf:
+            return pdf.page_count, (pdf.get_toc(simple=True) or [])
+
     try:
-        with fitz.open(str(source)) as pdf:
-            page_count = pdf.page_count
-            toc = pdf.get_toc(simple=True) or []
+        page_count, toc = await asyncio.to_thread(_read_toc, str(source))
     except Exception as exc:
         # PDF 손상 등 — 분할 불가. 단일문서로 통과(extract 가 PyMuPDF/OCR 로 재시도하며 가시화).
         logger.warning(

app/workers/stt_worker.py

@@ -42,6 +42,14 @@ async def process(document_id: int, session: AsyncSession) -> None:
         logger.warning(f"[stt] id={document_id} file_path 없음 — skip")
         return
 
+    if not settings.stt_enabled:
+        # 2노드 이관(2026-07-02): GPU stt-service 폐기 — 명시 비활성. silent 금지:
+        # 경고 로그 + extract_meta 터미널 기록 (재시도 안 함, 상태 가시).
+        doc.extract_meta = {**(doc.extract_meta or {}), "stt_skip_reason": "disabled", "stt_terminal": True}
+        await session.commit()
+        logger.warning(f"[stt] id={document_id} STT_ENABLED=false — 터미널 skip (전사 없음)")
+        return
+
     # NAS 마운트 경로로 절대화 (services/stt 컨테이너도 동일 경로에 bind mount)
     container_path = str(Path(settings.nas_mount_path) / doc.file_path)
 

app/workers/study_publish_worker.py

@@ -28,6 +28,8 @@ logger = setup_logger("study_publish_worker")
 BATCH_SIZE = 500
 # pg_advisory_xact_lock 전역 단일 라이터 키(발행 워커 전용 임의 상수, 타 advisory 락과 비충돌).
 ADVISORY_LOCK_KEY = 838201
+# 행별 격리 재시도 상한 — 초과 시 failed_at 스탬프(terminal)로 select 에서 제외.
+MAX_OUTBOX_ATTEMPTS = 5
 
 
 async def consume_publish_outbox() -> None:
@@ -46,11 +48,15 @@ async def consume_publish_outbox() -> None:
             max_rev = int(
                 (await session.execute(select(func.coalesce(func.max(Published.rev), 0)))).scalar() or 0
             )
-            # 3) 미처리 outbox 를 커밋순(id)으로.
+            # 3) 미처리 outbox 를 커밋순(id)으로. failed_at(terminal) 은 제외 — poison 행이
+            #    head-of-line 을 영구 점유하지 않게 함.
             rows = (
                 await session.execute(
                     select(PublishOutbox)
-                    .where(PublishOutbox.processed_at.is_(None))
+                    .where(
+                        PublishOutbox.processed_at.is_(None),
+                        PublishOutbox.failed_at.is_(None),
+                    )
                     .order_by(PublishOutbox.id.asc())
                     .limit(BATCH_SIZE)
                 )
@@ -60,59 +66,86 @@ async def consume_publish_outbox() -> None:
 
             now = datetime.now(timezone.utc)
             published_count = 0
+            failed_count = 0
             for ob in rows:
-                existing = (
-                    await session.execute(
-                        select(Published).where(
-                            Published.kind == ob.kind,
-                            Published.source_id == ob.source_id,
-                        )
-                    )
-                ).scalar_one_or_none()
+                try:
+                    # 행 단위 savepoint 격리 — 한 행의 예외가 배치 전체(앞 행 processed_at 포함)를
+                    # 롤백해 poison 행이 다음 사이클에 다시 최저 id 로 선택되는 무한 재선택을 차단.
+                    async with session.begin_nested():
+                        existing = (
+                            await session.execute(
+                                select(Published).where(
+                                    Published.kind == ob.kind,
+                                    Published.source_id == ob.source_id,
+                                )
+                            )
+                        ).scalar_one_or_none()
 
-                # (payload_hash, deleted) 디둡 — no-op 재투영은 rev 안 올림.
-                if (
-                    existing is not None
-                    and existing.payload_hash == ob.payload_hash
-                    and existing.deleted == ob.deleted
-                ):
-                    ob.processed_at = now
+                        # (payload_hash, deleted) 디둡 — no-op 재투영은 rev 안 올림.
+                        is_noop = (
+                            existing is not None
+                            and existing.payload_hash == ob.payload_hash
+                            and existing.deleted == ob.deleted
+                        )
+                        if is_noop:
+                            ob.processed_at = now
+                        else:
+                            new_rev = max_rev + 1
+                            if existing is None:
+                                session.add(
+                                    Published(
+                                        kind=ob.kind,
+                                        source_id=ob.source_id,
+                                        pub_id=uuid.uuid4().hex,
+                                        payload=ob.payload,
+                                        payload_hash=ob.payload_hash,
+                                        schema_version=ob.schema_version,
+                                        rev=new_rev,
+                                        deleted=ob.deleted,
+                                        created_at=now,
+                                        updated_at=now,
+                                    )
+                                )
+                            else:
+                                existing.payload = ob.payload
+                                existing.payload_hash = ob.payload_hash
+                                existing.schema_version = ob.schema_version
+                                existing.deleted = ob.deleted
+                                existing.rev = new_rev
+                                existing.updated_at = now
+                            ob.processed_at = now
+                            # 배치 내 동일 (kind, source_id) 후속 행이 직전 반영을 보도록 flush(최신 승).
+                            await session.flush()
+                except Exception as row_err:
+                    # savepoint 롤백 = 이 행의 쓰기(processed_at 포함) 취소. attempts/failed_at 만
+                    # 바깥 트랜잭션에 누적돼 최종 commit 으로 영속(영구 재선택 방지).
+                    ob.attempts = (ob.attempts or 0) + 1
+                    if ob.attempts >= MAX_OUTBOX_ATTEMPTS:
+                        ob.failed_at = now
+                        failed_count += 1
+                        logger.error(
+                            "publish_outbox_row_terminal id=%s kind=%s source_id=%s attempts=%s: %s",
+                            ob.id, ob.kind, ob.source_id, ob.attempts, row_err,
+                        )
+                    else:
+                        logger.warning(
+                            "publish_outbox_row_retry id=%s kind=%s source_id=%s attempts=%s: %s",
+                            ob.id, ob.kind, ob.source_id, ob.attempts, row_err,
+                        )
                     continue
-
-                max_rev += 1
-                if existing is None:
-                    session.add(
-                        Published(
-                            kind=ob.kind,
-                            source_id=ob.source_id,
-                            pub_id=uuid.uuid4().hex,
-                            payload=ob.payload,
-                            payload_hash=ob.payload_hash,
-                            schema_version=ob.schema_version,
-                            rev=max_rev,
-                            deleted=ob.deleted,
-                            created_at=now,
-                            updated_at=now,
-                        )
-                    )
                 else:
-                    existing.payload = ob.payload
-                    existing.payload_hash = ob.payload_hash
-                    existing.schema_version = ob.schema_version
-                    existing.deleted = ob.deleted
-                    existing.rev = max_rev
-                    existing.updated_at = now
-
-                ob.processed_at = now
-                # 배치 내 동일 (kind, source_id) 후속 행이 직전 반영을 보도록 flush(최신 승).
-                await session.flush()
-                published_count += 1
+                    # savepoint 커밋 성공 시에만 rev 카운터 전진(실패 행은 rev 미소모 → 드물게 gap,
+                    # 단일 라이터·커밋순 부여라 viewer since-rev 증분 동기 정합엔 무해).
+                    if not is_noop:
+                        max_rev = new_rev
+                        published_count += 1
 
             await session.commit()
             logger.info(
-                "publish_outbox_drained scanned=%s published=%s max_rev=%s",
+                "publish_outbox_drained scanned=%s published=%s failed=%s max_rev=%s",
                 len(rows),
                 published_count,
+                failed_count,
                 max_rev,
             )
         except Exception as e:

config.yaml

@@ -60,6 +60,9 @@ ai:
     rerank:
       endpoint: "http://reranker:80/rerank"
       model: "bge-reranker-v2-m3"
+      # 2노드 이관: "tei"(GPU TEI /rerank, 기본) | "llamacpp"(맥미니 llama.cpp,
+      # 예: endpoint http://100.76.254.116:8807/v1/rerank). 미지원 값 = 기동 시 ValueError.
+      protocol: "tei"
 
     # Phase 3.5a answerability classifier. 2026-05-14 GPU LLM 제거 후 Mac mini 26B 로 swap.
     # classifier_service 가 hasattr 체크로 optional 이므로 이 섹션 제거 시 classifier gate 는 자동 skip (score-only).

docker-compose.yml

@@ -3,7 +3,13 @@ services:
     image: pgvector/pgvector:pg16
     volumes:
       - pgdata:/var/lib/postgresql/data
-      - ./migrations:/docker-entrypoint-initdb.d
+      # ★ 2026-06-29 fresh-DB/DR 부팅 fix: initdb.d 마운트 제거(기존 `./migrations:/docker-entrypoint-initdb.d`).
+      # 빈 볼륨 첫 기동 시 postgres 엔트리포인트가 migrations/*.sql(001~) 을 psql autocommit 으로 실행해
+      # 스키마는 만들되 schema_migrations 스탬프는 안 남김(runner 만 생성) → fastapi init_db 가 documents
+      # 존재로 'fresh' 를 오판해 baseline(_load_baseline_if_fresh) 로드를 건너뛰고, 빈 schema_migrations
+      # 로 001 부터 재replay → `CREATE TABLE users`(IF NOT EXISTS 없음) 충돌 → 부팅 크래시(DR/신규환경).
+      # fresh-boot 은 init_db 의 baseline 적재 + migration runner 단일 경로로 일원화(설계 의도). 기존 prod
+      # 볼륨은 비어있지 않아 init scripts 가 애초에 미발동 → 무영향.
     environment:
       POSTGRES_DB: pkm
       POSTGRES_USER: pkm

docs/architecture.md

@@ -1094,7 +1094,7 @@ services:
     image: pgvector/pgvector:pg16
     volumes:
       - pgdata:/var/lib/postgresql/data
-      - ./migrations:/docker-entrypoint-initdb.d
+      # initdb.d 마운트 제거(2026-06-29): fresh-boot 은 fastapi init_db+baseline 단일 경로.
     environment:
       POSTGRES_DB: pkm
       POSTGRES_USER: pkm

docs/storage_layout.md

@@ -71,7 +71,7 @@ GPU 서버의 NFS mount (`/proc/mounts` 실측):
 
 | 컨테이너 | 마운트 | 모드 | 비고 |
 |---|---|---|---|
-| postgres | `pgdata:/var/lib/postgresql/data` + `./migrations:/docker-entrypoint-initdb.d` | rw | DB 본체 named volume |
+| postgres | `pgdata:/var/lib/postgresql/data` | rw | DB 본체 named volume (initdb.d 마운트는 2026-06-29 제거 — 아래 관찰) |
 | kordoc-service | `${NAS}/Document_Server:/documents` | **ro** | PDF/HWP parse |
 | ocr-service | `${NAS}/Document_Server:/documents` + `ocr_models:/root/.cache` | **ro** + rw | |
 | marker-service | `${NAS}/Document_Server:/documents` + `marker_models:/models` | **ro** + rw | PDF→markdown |
@@ -84,7 +84,7 @@ GPU 서버의 NFS mount (`/proc/mounts` 실측):
 **관찰**:
 - worker 컨테이너 (kordoc/ocr/marker/stt) 는 모두 NAS **read-only** 마운트 → 원본 안전.
 - fastapi 만 NAS **rw** → 업로드/preview/extracted_images 쓰기 단일 책임.
-- `./migrations` 이 postgres 의 `docker-entrypoint-initdb.d` 와 fastapi 의 `/app/migrations` 양쪽에 마운트. 단 실제 migration runner 는 fastapi `init_db()` 만 사용 (postgres init scripts 는 첫 생성 시만 실행 → 효과 X, 안전).
+- `./migrations` 은 fastapi 의 `/app/migrations` 에만 마운트. migration runner 는 fastapi `init_db()` 단일 경로. (~2026-06-29: postgres `docker-entrypoint-initdb.d` 마운트 제거. 기존엔 "첫 생성 시만 실행 → 효과 X" 로 봤으나, 빈 볼륨 첫 기동 시 postgres 가 migrations/*.sql 을 실제 실행해 스키마는 만들되 schema_migrations 스탬프를 안 남겨 → init_db 의 baseline fresh 판정을 깨고 부팅 크래시 유발. fresh-DB/DR 부팅을 init_db+baseline 단일 경로로 일원화.)
 
 ## 정책 정리
 

frontend/src/lib/components/ProcessingFlowBoard.svelte

@@ -1,6 +1,7 @@
 <script lang="ts">
-  // 처리 머신 보드 v3 — 통합안 (plan ds-board-merged: C2 머신레인 + C3 번다운/정직ETA).
-  //   · 머신 3레인(GPU/맥미니/맥북) = "누가 일하나" + 요약 오프로드(맥북 합류) 가시화
+  // 처리 머신 보드 v4 — 2026-07-02 컷오버 후 2노드 (나스+맥미니).
+  //   · 머신 2레인(나스/맥미니) = "누가 일하나" — 나스=DS 본체 Docker(추출/마크다운/
+  //     청크·임베딩 등), 맥미니=단일 생성 LLM 허브(분류/요약/심층분석 + bge-m3/리랭크)
   //   · 지배 백로그 번다운 패널 = "언제 끝나나" + 유입 차감한 정직 ETA(summarize_eta)
   //   · 신선도 '갱신 N초 전' + stale 경고 / 실패 드로어·상세 패널은 v2 자산 재사용.
   // 데이터 = GET /api/queue/overview (60s 폴링 store) + GET /api/queue/failed (드로어).
@@ -193,7 +194,7 @@
   const machineByKey = $derived(
     new Map<FlowMachine, MachineOverview>(overview.machines.map((m) => [m.key as FlowMachine, m])),
   );
-  const LANE_ORDER: FlowMachine[] = ['gpu', 'macmini', 'macbook'];
+  const LANE_ORDER: FlowMachine[] = ['nas', 'macmini'];
   const lanes = $derived(
     LANE_ORDER.map((key) => ({
       key,
@@ -203,13 +204,6 @@
     })),
   );
 
-  // 요약 오프로드 분담 — 맥미니 vs 맥북 (A-1 summarize_by_machine)
-  const split = $derived(overview.summarize_by_machine);
-  const splitTotal1h = $derived(Math.max(1, split.macmini.done_1h + split.macbook.done_1h));
-  const macbookSharePct = $derived(Math.round((split.macbook.done_1h / splitTotal1h) * 100));
-  // 맥북이 요약을 실제로 가져가는 중인가 (합류 표식 게이트)
-  const offloadActive = $derived(split.macbook.done_1h > 0);
-
   // ─── 백그라운드 작업 (큐 밖 스크립트 backfill) — processing_queue 사각지대 노출 ───
   const bgJobs = $derived(overview.background_jobs ?? []);
   const runningBg = $derived(bgJobs.filter((j) => j.state === 'running'));
@@ -266,7 +260,7 @@
         : `갱신 ${Math.round(ageSec / 60)}분 전`,
   );
 
-  // ─── 24h 번다운 (C3) — 요약 유입 vs 소화 + 맥북 합류 변곡점 마커 ───
+  // ─── 24h 번다운 (C3) — 요약 유입 vs 소화 ───
   const burn = $derived.by(() => {
     const t = overview.trend_24h;
     if (!t || t.length === 0) return null;
@@ -279,20 +273,12 @@
       t.map((b, i) => `${(i * step).toFixed(1)},${y(sel(b))}`).join(' ');
     const doneLine = line((b) => b.done);
     const area = `0,${h} ${doneLine} ${w.toFixed(1)},${h}`;
-    // 합류 변곡점 = done 최대 버킷 (맥북 야간 drain 합류 추정)
-    let mi = 0;
-    t.forEach((b, i) => {
-      if (b.done > t[mi].done) mi = i;
-    });
     return {
       w,
       h,
       area,
       doneLine,
       inflowLine: line((b) => b.inflow),
-      markX: (mi * step).toFixed(1),
-      markHour: t[mi].hour,
-      markDone: t[mi].done,
       peak: max,
     };
   });
@@ -332,7 +318,7 @@
     </span>
   </div>
 
-  <!-- 머신 레인 (누가 일하나 + 요약 오프로드) -->
+  <!-- 머신 레인 (누가 일하나) -->
   <div class="grid gap-2 mb-3">
     {#each lanes as lane (lane.key)}
       <div class="bg-surface border border-default rounded-card px-3.5 py-2.5">
@@ -342,11 +328,8 @@
           <span class="text-[10px] text-faint font-mono">{lane.meta.model}</span>
           <span class="text-[11px] text-dim tabular-nums ml-1">{formatRate(lane.card?.done_1h ?? 0)}/h</span>
           {#each bgForMachine(lane.key) as j (j.id)}<span class="text-[10px] font-semibold text-success tabular-nums ml-1">생성 중: {j.label ?? j.kind}{#if j.total} {j.processed}/{j.total}{/if}</span>{/each}
-          {#if lane.key === 'macbook' && (lane.card?.deferred_pending ?? 0) > 0}
-            <span class="text-[10px] font-semibold text-warning tabular-nums">보류 {lane.card?.deferred_pending}</span>
-          {/if}
-          {#if lane.card?.state === 'deferred'}
-            <span class="text-[9px] text-warning">잠듦 — 요약은 맥미니로 복귀</span>
+          {#if (lane.card?.deferred_pending ?? 0) > 0}
+            <span class="text-[10px] font-semibold text-warning tabular-nums" title="LLM 백오프 — 자동 재개 대기">보류 {lane.card?.deferred_pending}</span>
           {/if}
         </div>
         <div class="flex items-stretch gap-1.5 flex-wrap">
@@ -368,26 +351,8 @@
               </div>
               <div class="text-sm font-extrabold tabular-nums leading-tight text-text">{n.pending.toLocaleString()}<span class="text-[9px] text-faint font-normal ml-0.5">대기</span></div>
               <div class="text-[9px] text-dim tabular-nums whitespace-nowrap">{formatRate(n.done1h)}/h · 오늘 {n.doneToday.toLocaleString()}</div>
-              {#if n.def.key === 'summarize'}
-                <div class="mt-1 h-1 w-full rounded-full overflow-hidden flex" title="맥미니 {split.macmini.done_1h}/h · 맥북 {split.macbook.done_1h}/h">
-                  <span class="block h-full mtag-macmini-bar" style="width:{100 - macbookSharePct}%"></span>
-                  <span class="block h-full mtag-macbook-bar" style="width:{macbookSharePct}%"></span>
-                </div>
-                <div class="text-[9px] text-faint tabular-nums whitespace-nowrap mt-0.5">맥미니 {split.macmini.done_1h} · 맥북 {split.macbook.done_1h}/h</div>
-              {/if}
             </button>
           {/each}
-          {#if lane.key === 'macbook' && offloadActive}
-            <button
-              class="text-left rounded-lg border border-dashed border-warning/50 px-2.5 py-1.5 cursor-pointer hover:bg-surface-hover min-w-[96px]"
-              onclick={() => toggleNode('summarize')}
-              title="맥북이 요약을 맥미니에서 가져와 처리 중"
-            >
-              <div class="flex items-center gap-1 text-[11px] font-semibold text-text whitespace-nowrap">요약 합류 <span class="text-[8px] font-bold text-warning">OFFLOAD</span></div>
-              <div class="text-sm font-extrabold tabular-nums leading-tight text-text">{split.macbook.done_1h}<span class="text-[9px] text-faint font-normal ml-0.5">/h</span></div>
-              <div class="text-[9px] text-dim tabular-nums whitespace-nowrap">요약의 {macbookSharePct}% 담당</div>
-            </button>
-          {/if}
         </div>
       </div>
     {/each}
@@ -399,15 +364,11 @@
       <div class="flex items-center gap-2 mb-2">
         <span class="text-[11px] font-bold text-text">요약 백로그 24시간</span>
         <span class="text-[9px] text-faint">유입(회색) vs 소화(녹색)</span>
-        {#if offloadActive}<span class="text-[9px] text-warning ml-auto">맥북 합류 {burn.markHour} — 소화 급증</span>{/if}
       </div>
       <svg viewBox="0 0 {burn.w} {burn.h}" class="block w-full" style="height:64px" preserveAspectRatio="none" role="img" aria-label="요약 백로그 24시간 번다운">
         <polygon points={burn.area} fill="currentColor" class="text-success" opacity="0.12" />
         <polyline points={burn.inflowLine} fill="none" stroke="currentColor" stroke-width="1.2" class="text-faint" />
         <polyline points={burn.doneLine} fill="none" stroke="currentColor" stroke-width="1.6" class="text-success" />
-        {#if offloadActive}
-          <line x1={burn.markX} y1="0" x2={burn.markX} y2={burn.h} stroke="currentColor" stroke-width="1" stroke-dasharray="2 2" class="text-warning" opacity="0.7" />
-        {/if}
       </svg>
       <div class="flex flex-wrap gap-x-4 gap-y-1 mt-2 pt-2 border-t border-default text-[10px] text-dim tabular-nums">
         {#each mainNodes.filter((n) => n.pending > 0 && n.def.key !== 'summarize') as n (n.def.key)}
@@ -558,13 +519,9 @@
 </div>
 
 <style>
-  /* 머신 색 — 디자인 토큰 외 3색 (gpu 청/macmini 보라/macbook 황) — 이 컴포넌트 한정 */
-  .mtag-gpu { background: #e7eef6; color: #3b6ea5; }
+  /* 머신 색 — 디자인 토큰 외 2색 (nas 청/macmini 보라) — 이 컴포넌트 한정 */
+  .mtag-nas { background: #e7eef6; color: #3b6ea5; }
   .mtag-macmini { background: #efe9f7; color: #8a5fbf; }
-  .mtag-macbook { background: #f7eedd; color: #b07a10; }
-  /* 요약 오프로드 분담 막대 채움 (맥미니 보라 / 맥북 황) */
-  .mtag-macmini-bar { background: #8a5fbf; }
-  .mtag-macbook-bar { background: #b07a10; }
   .node-sel { outline: 2px solid #3b6ea5; outline-offset: 1px; }
   .detail-frame { border-color: #3b6ea5; }
   .detail-head { background: #e7eef6; }

frontend/src/lib/components/QueueDrawer.svelte

@@ -1,6 +1,6 @@
 <script lang="ts">
   // 처리 현황 드로어 (안6 라이트) — 전 페이지 상태 스트립 클릭 시 우측에서 열림.
-  // 머신 미니카드 3 + ETA 한 줄 + 실패 합계 + 홈 링크 축약본. 상세는 홈 보드가 담당.
+  // 머신 미니카드 2(나스/맥미니) + ETA 한 줄 + 실패 합계 + 홈 링크 축약본. 상세는 홈 보드가 담당.
   // 데이터 = queueOverview store 공유 (60s 폴링, 실패 시 null → 안내문으로 degrade).
   // 열림 상태는 uiState 단일 drawer slot('queue') — 사이드바 드로어와 동시 오픈 차단.
   import { X } from 'lucide-svelte';
@@ -51,7 +51,7 @@
 
       <div class="p-4 space-y-3">
         {#if data}
-          <!-- 머신 미니카드 3 -->
+          <!-- 머신 미니카드 (나스/맥미니) -->
           {#each data.machines as m (m.key)}
             <div class="bg-surface border border-default rounded-lg px-3.5 py-2.5">
               <div class="flex items-center justify-between gap-2">

frontend/src/lib/components/RelatedDocs.svelte

@@ -0,0 +1,45 @@
+<script>
+  // 관련 문서 (유사도) — 문서 레벨 임베딩 KNN. 자기완결: docId 받아 /related 조회.
+  import { onMount } from 'svelte';
+  import { api } from '$lib/api';
+
+  let { documentId } = $props();
+  let items = $state([]);
+  let loaded = $state(false);
+
+  const KIND = { law: '법령', guide: '지침', paper: '논문', standard: '표준', incident: '사례' };
+
+  onMount(async () => {
+    try {
+      const r = await api(`/documents/${documentId}/related?limit=6`);
+      items = r?.related ?? [];
+    } catch (e) { /* silent */ }
+    finally { loaded = true; }
+  });
+</script>
+
+{#if items.length}
+  <div class="rel">
+    <div class="lab">관련 문서</div>
+    {#each items as it (it.id)}
+      <a class="ri" href={`/documents/${it.id}`}>
+        <span class="rt">{it.title}</span>
+        <span class="rm">
+          {#if it.material_type && KIND[it.material_type]}<span class="kind">{KIND[it.material_type]}</span>{/if}
+          <span class="rs">{Math.round((it.sim ?? 0) * 100)}</span>
+        </span>
+      </a>
+    {/each}
+  </div>
+{/if}
+
+<style>
+  .rel { background: var(--surface); border: 1px solid var(--border); border-radius: 14px; padding: 13px; }
+  .lab { font-size: 10.5px; font-weight: 700; color: var(--text-dim); letter-spacing: .4px; margin-bottom: 8px; }
+  .ri { display: flex; align-items: baseline; gap: 8px; padding: 5px 6px; border-radius: 7px; text-decoration: none; }
+  .ri:hover { background: var(--surface-hover, #ecf0e8); }
+  .rt { flex: 1; font-size: 12px; line-height: 1.4; color: var(--text); overflow: hidden; display: -webkit-box; -webkit-line-clamp: 2; -webkit-box-orient: vertical; }
+  .rm { flex-shrink: 0; display: flex; align-items: center; gap: 5px; }
+  .kind { font-size: 9px; font-weight: 700; color: var(--accent-hover, #3d7256); background: #e3efe2; border: 1px solid #cfe3cd; border-radius: 4px; padding: 0 4px; }
+  .rs { font-size: 10.5px; font-family: ui-monospace, Menlo, monospace; color: var(--faint, #9aa090); }
+</style>

frontend/src/lib/components/Sidebar.svelte

@@ -2,7 +2,7 @@
   import { page } from '$app/stores';
   import { goto } from '$app/navigation';
   import { api } from '$lib/api';
-  import { ChevronRight, ChevronDown, FolderOpen, FolderTree, Inbox, Clock, Mail, Scale, StickyNote, GraduationCap, CalendarCheck, MessageCircle, Hash } from 'lucide-svelte';
+  import { ChevronRight, ChevronDown, FolderOpen, FolderTree, Inbox, Clock, Mail, Scale, StickyNote, GraduationCap, CalendarCheck, MessageCircle, Hash, HardHat } from 'lucide-svelte';
 
   let tree = $state([]);
   let loading = $state(true);
@@ -195,6 +195,13 @@
     >
       <FolderTree size={14} /> 자료실
     </a>
+    <a
+      href="/safety"
+      class="w-full flex items-center gap-2 px-3 py-1.5 rounded-md text-sm transition-colors
+        {$page.url.pathname.startsWith('/safety') ? 'bg-accent/15 text-accent' : 'text-dim hover:bg-surface hover:text-text'}"
+    >
+      <HardHat size={14} /> 안전 자료실
+    </a>
     <a
       href="/clause"
       class="w-full flex items-center gap-2 px-3 py-1.5 rounded-md text-sm transition-colors

frontend/src/lib/components/SlimRail.svelte

@@ -0,0 +1,33 @@
+<script>
+  // 시안 B — 글로벌 네비 슬림 아이콘 레일 (분류 사이드바 접힘 상태). 앱 토큰 사용.
+  import { page } from '$app/stores';
+  import { Home, FolderTree, Newspaper, StickyNote, Hash, GraduationCap, MessageCircle, Inbox, CalendarCheck } from 'lucide-svelte';
+
+  const items = [
+    { href: '/', icon: Home, label: '홈', exact: true },
+    { href: '/library', icon: FolderTree, label: '문서' },
+    { href: '/news', icon: Newspaper, label: '뉴스' },
+    { href: '/memos', icon: StickyNote, label: '메모' },
+    { href: '/clause', icon: Hash, label: '절' },
+    { href: '/events', icon: CalendarCheck, label: '일정' },
+    { href: '/study', icon: GraduationCap, label: '공부' },
+    { href: '/chat', icon: MessageCircle, label: '이드' },
+    { href: '/inbox', icon: Inbox, label: '편지함' },
+  ];
+  let path = $derived($page.url.pathname);
+  const active = (it) => (it.exact ? path === it.href : path.startsWith(it.href));
+</script>
+
+<nav class="flex flex-col items-center gap-1 py-2 h-full overflow-y-auto bg-sidebar">
+  {#each items as it (it.href)}
+    {@const Icon = it.icon}
+    <a
+      href={it.href}
+      title={it.label}
+      class="flex flex-col items-center justify-center gap-0.5 w-12 h-[46px] rounded-lg text-dim hover:bg-surface-hover hover:text-accent transition-colors {active(it) ? 'bg-surface-active text-accent font-semibold' : ''}"
+    >
+      <Icon size={17} strokeWidth={1.75} />
+      <span class="text-[8.5px] leading-none tracking-tight">{it.label}</span>
+    </a>
+  {/each}
+</nav>

frontend/src/lib/types/queue.ts

@@ -5,7 +5,7 @@
  * 필드 변경 시 양쪽 동시 수정 필수.
  */
 
-export type MachineKey = 'gpu' | 'macmini' | 'macbook';
+export type MachineKey = 'nas' | 'macmini';
 
 /** 머신 상태 — active(가동) / deferred(보류) / idle(대기) */
 export type MachineState = 'active' | 'deferred' | 'idle';
@@ -29,7 +29,7 @@ export interface MachineOverview {
   /** 최근 1시간 완료 건수 (처리율 N/h 표기) */
   done_1h: number;
   done_today: number;
-  /** 보류 건수 — 맥북 sleep 등으로 자동 재개 대기 중 */
+  /** 보류 건수 — LLM 허브 백오프 등으로 자동 재개 대기 중 */
   deferred_pending: number;
   current: MachineCurrentItem[];
 }
@@ -50,12 +50,6 @@ export interface TrendPoint {
   done: number;
 }
 
-/** summarize 머신별 완료 실적 분담 (오프로드 가시화 — ds-board-merged A-1) */
-export interface SummarizeByMachine {
-  macmini: { done_1h: number; done_today: number };
-  macbook: { done_1h: number; done_today: number };
-}
-
 export interface QueueTotals {
   pending: number;
   processing: number;
@@ -93,7 +87,6 @@ export interface BackgroundJob {
 export interface QueueOverview {
   machines: MachineOverview[];
   summarize_eta: SummarizeEta;
-  summarize_by_machine: SummarizeByMachine;
   trend_24h: TrendPoint[];
   stages: QueueStageRow[];
   totals: QueueTotals;

frontend/src/lib/utils/queueDisplay.ts

@@ -62,7 +62,7 @@ export function formatAgeSec(sec: number): string {
  * ★ 모델/엔진 교체 시 이 블록 1곳만 수정 (예: 맥미니 모델 스왑).
  */
 
-export type FlowMachine = 'gpu' | 'macmini' | 'macbook';
+export type FlowMachine = 'nas' | 'macmini';
 
 export interface FlowNodeDef {
   key: string;
@@ -79,26 +79,25 @@ export interface FlowNodeDef {
 
 /** 메인 흐름 (문서 진행 순서). 뉴스 등 소스별 스킵 경로는 그림에 안 그림 — 단순화 한계. */
 export const FLOW_NODES: FlowNodeDef[] = [
-  { key: 'extract', label: '추출', stages: ['extract'], machine: 'gpu', engine: 'Surya OCR', sub: 'ocr-service' },
-  { key: 'markdown', label: '마크다운', stages: ['markdown'], machine: 'gpu', engine: 'Marker', sub: 'marker-service' },
+  { key: 'extract', label: '추출', stages: ['extract'], machine: 'nas', engine: 'kordoc', sub: 'kordoc' },
+  { key: 'markdown', label: '마크다운', stages: ['markdown'], machine: 'nas', engine: 'Marker', sub: 'marker-service' },
   { key: 'classify', label: '분류', stages: ['classify'], machine: 'macmini', engine: 'Qwen3.6-27B', sub: 'classify + triage' },
   { key: 'summarize', label: '요약', stages: ['summarize'], machine: 'macmini', engine: 'Qwen3.6-27B', sub: 'summarize' },
-  { key: 'chunkembed', label: '청크 · 임베딩', stages: ['chunk', 'embed'], machine: 'gpu', engine: 'TEI bge-m3', sub: 'text-embeddings-inference' },
-  { key: 'deep', label: '심층분석', stages: ['deep_summary'], machine: 'macbook', engine: 'Qwen3.6-27B', sub: 'deep_summary' },
+  { key: 'chunkembed', label: '청크 · 임베딩', stages: ['chunk', 'embed'], machine: 'nas', engine: 'bge-m3 (맥미니 콜)', sub: 'embed worker' },
+  { key: 'deep', label: '심층분석', stages: ['deep_summary'], machine: 'macmini', engine: 'Qwen3.6-27B', sub: 'deep_summary' },
 ];
 
 /** 보조 노드 — 메인 흐름 밖 (활동 있을 때만 보조 라인에 표시) */
 export const AUX_NODES: FlowNodeDef[] = [
-  { key: 'fulltext', label: '전문 수집', stages: ['fulltext'], machine: 'gpu', engine: 'Playwright', sub: 'playwright-fetcher' },
-  { key: 'stt', label: '전사', stages: ['stt'], machine: 'gpu', engine: 'Whisper', sub: 'stt-service' },
-  { key: 'util', label: '미리보기 · 썸네일', stages: ['preview', 'thumbnail'], machine: 'gpu', engine: '유틸', sub: 'ffmpeg' },
+  { key: 'fulltext', label: '전문 수집', stages: ['fulltext'], machine: 'nas', engine: 'Playwright', sub: 'playwright-fetcher' },
+  { key: 'stt', label: '전사', stages: ['stt'], machine: 'nas', engine: 'Whisper', sub: 'stt-service' },
+  { key: 'util', label: '미리보기 · 썸네일', stages: ['preview', 'thumbnail'], machine: 'nas', engine: '유틸', sub: 'ffmpeg' },
 ];
 
-/** 머신 스트립 메타 — 모델 표기 단일 지점 */
+/** 머신 스트립 메타 — 모델 표기 단일 지점 (2026-07-02 컷오버: 나스+맥미니 2노드) */
 export const MACHINE_META: Record<FlowMachine, { label: string; model: string }> = {
-  gpu: { label: 'GPU 서버', model: '특화 엔진' },
-  macmini: { label: '맥미니', model: 'Qwen3.6-27B-6bit · 24/7' },
-  macbook: { label: '맥북 M5 Max', model: 'Qwen3.6-27B · 야간 drain' },
+  nas: { label: '나스', model: 'DS 본체 Docker · 특화 엔진' },
+  macmini: { label: '맥미니', model: 'Qwen3.6-27B-6bit · bge-m3 · 24/7' },
 };
 
 /** 흐름 보드 단계 라벨 (드로어/상세 행 표기) */

frontend/src/routes/+layout.svelte

@@ -11,6 +11,7 @@
   import { queueOverview } from '$lib/stores/queueOverview';
   import { MACHINE_STATE_LABEL, machineChipClass } from '$lib/utils/queueDisplay';
   import Sidebar from '$lib/components/Sidebar.svelte';
+  import SlimRail from '$lib/components/SlimRail.svelte';
   import SystemStatusDot from '$lib/components/SystemStatusDot.svelte';
   import QueueDrawer from '$lib/components/QueueDrawer.svelte';
   import QuickMemoButton from '$lib/components/QuickMemoButton.svelte';
@@ -21,7 +22,7 @@
   const PUBLIC_PATHS = ['/login', '/setup', '/__styleguide'];
   const NO_CHROME_PATHS = ['/login', '/setup', '/__styleguide'];
   // /news = 풀스크린 브리핑 → 데스크탑 상시 사이드바 없음
-  const NO_SIDEBAR_PATHS = ['/news'];
+  const NO_SIDEBAR_PATHS = ['/news', '/book'];  // /book = 책 몰입(글로벌 분류 트리 숨김, 상단 네비 유지)
 
   // toast 의미 토큰 매핑 (A-8 B3)
   const TOAST_CLASS = {
@@ -71,7 +72,7 @@
   // 처리 현황 스트립 (안6 라이트) — 60s 폴링 store 공유. fetch 실패/401 시
   // store 가 null → 스트립 자체를 숨김 (silent 비차단, 로그인 페이지 동일).
   let queue = $derived($queueOverview);
-  let queueMacbook = $derived(queue?.machines?.find((m) => m.key === 'macbook') ?? null);
+  let queueMacmini = $derived(queue?.machines?.find((m) => m.key === 'macmini') ?? null);
   function toggleQueueDrawer() {
     if (ui.isDrawerOpen('queue')) ui.closeDrawer();
     else ui.openDrawer('queue');
@@ -188,8 +189,8 @@
           </span>
           <span class="tabular-nums shrink-0">대기 <strong class="text-text">{queue.totals.pending.toLocaleString()}</strong></span>
           <span class="tabular-nums shrink-0 {queue.totals.failed > 0 ? 'text-error font-semibold' : ''}">실패 <strong class={queue.totals.failed > 0 ? '' : 'text-text'}>{queue.totals.failed.toLocaleString()}</strong></span>
-          {#if queueMacbook}
-            <span class="text-[10px] font-bold rounded-full px-2 py-0.5 shrink-0 {machineChipClass(queueMacbook.state)}">맥북 {MACHINE_STATE_LABEL[queueMacbook.state]}</span>
+          {#if queueMacmini}
+            <span class="text-[10px] font-bold rounded-full px-2 py-0.5 shrink-0 {machineChipClass(queueMacmini.state)}">맥미니 {MACHINE_STATE_LABEL[queueMacmini.state]}</span>
           {/if}
           <span class="ml-auto flex items-center gap-0.5 text-faint shrink-0">자세히 <ChevronDown size={11} /></span>
         </button>
@@ -198,8 +199,8 @@
       <!-- 메인: 데스크탑 상시 사이드바 + 콘텐츠 -->
       <div class="flex-1 min-h-0 flex">
         {#if showSidebar}
-          <aside class="hidden lg:block shrink-0 overflow-hidden transition-[width] duration-200 ease-out {sidebarCollapsed ? 'w-0 border-r-0' : 'w-sidebar border-r border-default'}">
-            <Sidebar />
+          <aside class="hidden lg:block shrink-0 overflow-hidden transition-[width] duration-200 ease-out {sidebarCollapsed ? 'w-14 border-r border-default' : 'w-sidebar border-r border-default'}">
+            {#if sidebarCollapsed}<SlimRail />{:else}<Sidebar />{/if}
           </aside>
         {/if}
         <main class="flex-1 min-w-0 overflow-auto">

frontend/src/routes/book/[id]/+page.svelte

@@ -0,0 +1,281 @@
+<script>
+  // ASME/법령 절-KB — 코드북·공부 리더 (r2). parent 표준/법령을 한 권의 책처럼.
+  // 좌 인덱스(Part/章→절/조) · 중 본문(MarkdownDoc=공식·표·이미지) · breadcrumb·이전다음·양방향 백링크.
+  import { onMount, tick } from 'svelte';
+  import { page } from '$app/stores';
+  import { goto } from '$app/navigation';
+  import { api } from '$lib/api';
+  import MarkdownDoc from '$lib/components/MarkdownDoc.svelte';
+
+  let parentId = $state(null);
+  let parentTitle = $state('');
+  let clauses = $state([]);
+  let selectedId = $state(null);
+  let clauseDoc = $state(null);
+  let links = $state(null);
+  let expanded = $state({});
+  let loading = $state(false);
+  let q = $state('');
+
+  // 공부도구 (노트/형광펜/암기카드) — clause_study
+  let studyItems = $state([]);
+  let studyOpen = $state(false);
+  let noteDraft = $state('');
+  const KLABEL = { note: '노트', highlight: '형광펜', card: '암기카드' };
+  async function loadStudy(id) {
+    try { const r = await api(`/documents/${id}/study`); studyItems = r?.items ?? []; }
+    catch { studyItems = []; }
+  }
+  async function addStudy(kind, payload) {
+    if (!selectedId) return;
+    try { await api(`/documents/${selectedId}/study`, { method: 'POST', body: JSON.stringify({ kind, payload }) }); await loadStudy(selectedId); }
+    catch (e) { console.warn(e); }
+  }
+  function selText() { return (typeof window !== 'undefined' && window.getSelection ? window.getSelection().toString() : '').trim(); }
+  function addNote() { const t = noteDraft.trim(); if (!t) return; addStudy('note', { text: t }); noteDraft = ''; }
+  function addHighlight() { const s = selText(); if (!s) { studyOpen = true; alert('본문에서 형광펜 칠할 부분을 먼저 드래그하세요'); return; } addStudy('highlight', { text: s }); studyOpen = true; }
+  function addCard() {
+    const s = selText();
+    const code = links?.clause_code ?? selMeta?.clause_code ?? '';
+    addStudy('card', { cue: `${code} ${strip(clauseDoc?.title, code)}`.trim(), fact: s || (clauseDoc?.md_content ?? clauseDoc?.extracted_text ?? '').replace(/[#*>]/g, '').slice(0, 280).trim() });
+    studyOpen = true;
+  }
+  async function delStudy(id) {
+    try { await api(`/documents/${selectedId}/study/${id}`, { method: 'DELETE' }); await loadStudy(selectedId); } catch {}
+  }
+
+  let parts = $derived.by(() => {
+    const out = [], idx = {};
+    for (const c of clauses) {
+      const p = c.clause_part || '·';
+      if (!(p in idx)) { idx[p] = out.length; out.push({ part: p, items: [] }); }
+      out[idx[p]].items.push(c);
+    }
+    return out;
+  });
+  let visibleParts = $derived.by(() => {
+    const term = q.trim().toLowerCase();
+    if (!term) return parts;
+    return parts.map(g => ({ part: g.part, items: g.items.filter(c =>
+      (c.clause_code || '').toLowerCase().includes(term) || (c.title || '').toLowerCase().includes(term)) }))
+      .filter(g => g.items.length);
+  });
+  let selMeta = $derived(clauses.find((c) => c.id === selectedId) || null);
+  const strip = (t, c) => (t || '').replace(c || '', '').replace(/^[(\s)]+|[(\s)]+$/g, '').trim();
+
+  async function loadBook() {
+    const r = await api(`/documents/${parentId}/clauses`);
+    parentTitle = r?.parent_title ?? '';
+    clauses = r?.clauses ?? [];
+    const e = {};
+    for (const c of clauses) e[c.clause_part || '·'] = true;
+    expanded = e;
+  }
+  async function loadClause(id) {
+    if (!id) return;
+    loading = true;
+    selectedId = id;
+    try {
+      const [d, l] = await Promise.all([api(`/documents/${id}`), api(`/documents/${id}/backlinks`)]);
+      clauseDoc = d; links = l;
+      loadStudy(id);
+      const sel = clauses.find((c) => c.id === id);
+      if (sel) expanded = { ...expanded, [sel.clause_part || '·']: true };
+      goto(`/book/${parentId}?c=${id}`, { replaceState: true, keepFocus: true, noScroll: true });
+      await tick(); window.scrollTo({ top: 0 });
+    } finally { loading = false; }
+  }
+  onMount(async () => {
+    parentId = Number($page.params.id);
+    await loadBook();
+    const c = Number($page.url.searchParams.get('c'));
+    await loadClause(c && clauses.find((x) => x.id === c) ? c : clauses[0]?.id);
+  });
+</script>
+
+<div class="book">
+  <!-- top bar -->
+  <div class="bar">
+    <span class="brand">절-KB</span>
+    <span class="crumbs">{parentTitle} {#if selMeta}<b class="sep">›</b> {selMeta.clause_part} <b class="sep">›</b> <b>{links?.clause_code ?? selMeta.clause_code}</b>{/if}</span>
+    <div class="search"><input placeholder="절·조 번호 또는 키워드" bind:value={q} /></div>
+    <div class="tools"><span class="tool on">읽기</span><span class="tool">형광펜</span><span class="tool">노트</span><span class="tool">암기카드</span></div>
+  </div>
+
+  <div class="main">
+    <!-- left index -->
+    <aside class="idx">
+      <a class="btitle" href={`/documents/${parentId}`}>{parentTitle || '표준'}</a>
+      <div class="bmeta">절 {clauses.length} · 한 권의 책처럼 탐색</div>
+      {#each visibleParts as g (g.part)}
+        <div class="parttab" role="button" tabindex="0" onclick={() => (expanded = { ...expanded, [g.part]: !expanded[g.part] })}>
+          <span class="bar2"></span><span class="pname">{g.part}</span><span class="ct">{g.items.length}</span>
+        </div>
+        {#if expanded[g.part] || q.trim()}
+          {#each g.items as c (c.id)}
+            <div class="ci" class:on={c.id === selectedId} role="button" tabindex="0" onclick={() => loadClause(c.id)}>
+              <span class="no">{c.clause_code}</span><span class="tt">{strip(c.title, c.clause_code)}</span>
+            </div>
+          {/each}
+        {/if}
+      {/each}
+    </aside>
+
+    <!-- reader -->
+    <section class="read">
+      <div class="col">
+        {#if clauseDoc}
+          <div class="studybar">
+            <button class="sbtn" title="선택 형광펜" onclick={addHighlight}>▰</button>
+            <button class="sbtn" class:on={studyOpen} title="노트/공부" onclick={() => (studyOpen = !studyOpen)}>✎</button>
+            <button class="sbtn" title="암기카드 추가" onclick={addCard}>＋</button>
+            {#if studyItems.length}<span class="scount">{studyItems.length}</span>{/if}
+          </div>
+          <div class="kicker"><span class="pth">{selMeta?.clause_part}</span></div>
+          <div class="h-no">{links?.clause_code ?? selMeta?.clause_code}</div>
+          <h1 class="h-title">{strip(clauseDoc.title, links?.clause_code ?? '')}</h1>
+
+          <div class="flow">
+            <button class="fl" disabled={!links?.prev} onclick={() => loadClause(links?.prev?.id)}>← {links?.prev?.clause_code ?? ''}</button>
+            <button class="fl next" disabled={!links?.next} onclick={() => loadClause(links?.next?.id)}>{links?.next?.clause_code ?? ''} →</button>
+          </div>
+
+          {#key clauseDoc.id}
+            <div class="docbody">
+              <MarkdownDoc documentId={clauseDoc.id} mdContent={clauseDoc.md_content ?? clauseDoc.extracted_text} mdStatus={null} class="prose prose-base max-w-none" />
+            </div>
+          {/key}
+
+          {#if links && (links.forward.length || links.back.length)}
+            <section class="conn">
+              {#if links.forward.length}
+                <div><h4>이 절이 참조 <span>{links.forward.length}</span></h4>
+                  <div class="chiprow">{#each links.forward as f}
+                    {#if f.doc_id}<button class="ref" onclick={() => loadClause(f.doc_id)}>{f.code}</button>
+                    {:else}<span class="ref dg" title="외부/미분해">{f.code}</span>{/if}
+                  {/each}</div></div>
+              {/if}
+              {#if links.back.length}
+                <div><h4>이 절을 참조 <span>{links.back.length}</span></h4>
+                  <div class="chiprow">{#each links.back as b}<button class="ref" onclick={() => loadClause(b.doc_id)}>{b.code}</button>{/each}</div></div>
+              {/if}
+            </section>
+          {/if}
+
+          {#if studyOpen}
+            <section class="study">
+              <div class="slab">공부 — 노트 · 형광펜 · 암기카드{#if studyItems.length} <span>{studyItems.length}</span>{/if}</div>
+              <div class="noteadd">
+                <textarea bind:value={noteDraft} placeholder="이 절에 노트…" rows="2"></textarea>
+                <button class="nbtn" onclick={addNote}>노트 저장</button>
+              </div>
+              {#if studyItems.length}
+                <ul class="slist">
+                  {#each studyItems as it (it.id)}
+                    <li class="sitem">
+                      <span class="skind k-{it.kind}">{KLABEL[it.kind] ?? it.kind}</span>
+                      <span class="stext">{it.payload?.text ?? it.payload?.cue ?? ''}</span>
+                      <button class="sdel" title="삭제" onclick={() => delStudy(it.id)}>×</button>
+                    </li>
+                  {/each}
+                </ul>
+              {:else}
+                <p class="shint">본문을 드래그한 뒤 형광펜(▰)/암기카드(＋), 또는 위에 노트를 적으세요.</p>
+              {/if}
+            </section>
+          {/if}
+
+          <div class="pager">
+            <button class="pg" disabled={!links?.prev} onclick={() => loadClause(links?.prev?.id)}>
+              <div class="d">← 이전</div><div class="t"><span class="pno">{links?.prev?.clause_code ?? '—'}</span> {strip(links?.prev?.title, links?.prev?.clause_code)}</div></button>
+            <button class="pg next" disabled={!links?.next} onclick={() => loadClause(links?.next?.id)}>
+              <div class="d">다음 →</div><div class="t"><span class="pno">{links?.next?.clause_code ?? '—'}</span> {strip(links?.next?.title, links?.next?.clause_code)}</div></button>
+          </div>
+        {:else}
+          <p class="empty">{loading ? '불러오는 중…' : '왼쪽에서 절을 선택하세요'}</p>
+        {/if}
+      </div>
+    </section>
+  </div>
+</div>
+
+<style>
+  :global(body) { background: var(--bg); }
+  .book { --paper:#fbfcf9; --serif:"Iowan Old Style","Palatino Linotype","Noto Serif KR",Georgia,serif;
+    display:flex; flex-direction:column; min-height:100vh; }
+  .bar { display:flex; align-items:center; gap:14px; height:50px; padding:0 18px; background:var(--paper); border-bottom:1px solid var(--border); }
+  .brand { font-weight:700; font-size:13.5px; color:var(--text); }
+  .crumbs { color:var(--text-dim); font-size:12.5px; overflow:hidden; text-overflow:ellipsis; white-space:nowrap; max-width:46%; }
+  .crumbs b { color:var(--text); font-weight:600; } .crumbs .sep { color:#c8d6c0; margin:0 5px; }
+  .search { margin-left:auto; }
+  .search input { width:280px; background:var(--surface); border:1px solid var(--border); border-radius:9px; padding:7px 12px; font-size:13px; color:var(--text); outline:none; }
+  .search input:focus { border-color:var(--accent); }
+  .tools { display:flex; gap:2px; }
+  .tool { font-size:12px; color:var(--text-dim); padding:6px 10px; border-radius:8px; border:1px solid transparent; cursor:pointer; }
+  .tool:hover { background:var(--surface); } .tool.on { background:#ecf0e8; border-color:var(--border); color:var(--accent-hover); font-weight:600; }
+
+  .main { display:flex; align-items:flex-start; flex:1; }
+  .idx { width:264px; flex-shrink:0; align-self:stretch; border-right:1px solid var(--border);
+    background:linear-gradient(180deg,#f6f8f3,#f1f4ec); padding:16px 10px 30px 16px; position:sticky; top:0; max-height:100vh; overflow:auto; }
+  .btitle { display:block; font-family:var(--serif); font-size:15.5px; font-weight:600; color:var(--text); text-decoration:none; line-height:1.32; }
+  .btitle:hover { text-decoration:underline; }
+  .bmeta { font-size:11px; color:#9aa090; margin:3px 0 14px; }
+  .parttab { display:flex; align-items:center; gap:8px; margin:11px 0 4px; padding:3px 4px; border-radius:6px; cursor:pointer;
+    font-size:11px; font-weight:700; letter-spacing:.5px; color:var(--text-dim); text-transform:uppercase; }
+  .parttab:hover { background:#fff; } .parttab .bar2 { width:3px; height:12px; border-radius:2px; background:var(--domain-engineering); }
+  .parttab .pname { flex:1; overflow:hidden; text-overflow:ellipsis; white-space:nowrap; } .parttab .ct { color:#9aa090; font-weight:600; letter-spacing:0; }
+  .ci { display:flex; gap:9px; align-items:baseline; padding:4px 9px; border-radius:7px; cursor:pointer; line-height:1.4; }
+  .ci .no { font-family:ui-monospace,Menlo,monospace; font-size:11px; color:var(--accent); font-weight:600; min-width:52px; white-space:nowrap; }
+  .ci .tt { font-size:12.5px; color:var(--text-dim); overflow:hidden; text-overflow:ellipsis; }
+  .ci:hover { background:#fff; }
+  .ci.on { background:#fff; box-shadow:inset 3px 0 0 var(--accent), 0 1px 2px rgba(35,41,31,.05); }
+  .ci.on .no { color:var(--accent-hover); font-weight:700; } .ci.on .tt { color:var(--text); font-weight:600; }
+
+  .read { flex:1; min-width:0; padding:34px 40px 80px; }
+  .col { max-width:680px; margin:0 auto; position:relative; }
+  .studybar { position:absolute; right:-30px; top:4px; display:flex; flex-direction:column; gap:6px; }
+  .sbtn { width:34px; height:34px; border-radius:9px; border:1px solid var(--border); background:var(--paper); color:var(--text-dim); font-size:13px; cursor:pointer; }
+  .sbtn:hover { background:var(--surface); color:var(--accent-hover); }
+  .kicker { margin-bottom:5px; } .kicker .pth { font-size:11.5px; color:#9aa090; font-weight:600; letter-spacing:.3px; }
+  .h-no { font-family:ui-monospace,Menlo,monospace; font-size:13px; color:var(--accent); font-weight:700; letter-spacing:.5px; }
+  .h-title { font-family:var(--serif); font-size:26px; line-height:1.24; font-weight:600; margin:2px 0 14px; letter-spacing:-.2px; color:var(--text); }
+  .flow { display:flex; justify-content:space-between; gap:8px; margin-bottom:18px; }
+  .flow .fl { font-size:11.5px; color:var(--text-dim); background:var(--surface); border:1px solid var(--border); border-radius:8px; padding:5px 11px; cursor:pointer; }
+  .flow .fl:hover:not(:disabled) { background:#ecf0e8; } .flow .fl:disabled { opacity:.35; cursor:default; }
+  .docbody { font-size:15.5px; }
+  .docbody :global(.prose) { color:#2a3024; line-height:1.78; }
+  .docbody :global(.prose h1), .docbody :global(.prose h2), .docbody :global(.prose h3) { font-family:var(--serif); }
+  .docbody :global(a) { color:var(--accent-hover); }
+  .conn { margin-top:34px; padding-top:18px; border-top:1px solid var(--border); display:grid; grid-template-columns:1fr 1fr; gap:22px; }
+  .conn h4 { font-size:11px; font-weight:700; color:var(--text-dim); letter-spacing:.4px; margin:0 0 9px; } .conn h4 span { color:#9aa090; font-weight:500; }
+  .chiprow { display:flex; flex-wrap:wrap; gap:5px; }
+  .ref { font-family:ui-monospace,Menlo,monospace; font-size:11.5px; font-weight:600; color:var(--accent-hover); background:#eef4ec; border:1px solid #d9e6d8; border-radius:6px; padding:2px 8px; cursor:pointer; }
+  .ref:hover { background:#e2efe0; } .ref.dg { color:#9aa090; background:var(--surface); border-color:var(--border); cursor:default; }
+  .pager { display:flex; gap:10px; margin-top:30px; }
+  .pg { flex:1; text-align:left; border:1px solid var(--border); border-radius:11px; padding:11px 14px; background:var(--paper); cursor:pointer; }
+  .pg.next { text-align:right; } .pg:hover:not(:disabled) { border-color:#cfd7c6; background:#fff; } .pg:disabled { opacity:.4; cursor:default; }
+  .pg .d { font-size:10.5px; color:#9aa090; } .pg .t { font-size:12.5px; color:var(--text-dim); font-weight:600; margin-top:1px; overflow:hidden; text-overflow:ellipsis; white-space:nowrap; }
+  .pg .pno { font-family:ui-monospace,Menlo,monospace; color:var(--accent); }
+  .empty { color:#9aa090; text-align:center; padding:80px 0; }
+  .sbtn.on { background:#ecf0e8; color:var(--accent-hover,#3d7256); border-color:var(--border); }
+  .scount { font-size:9px; font-weight:700; color:#fff; background:var(--accent,#4f8a6b); border-radius:8px; padding:1px 5px; text-align:center; }
+  .study { margin-top:24px; padding:14px; border:1px solid var(--border); border-radius:12px; background:var(--surface); }
+  .slab { font-size:11px; font-weight:700; color:var(--text-dim); letter-spacing:.3px; margin-bottom:9px; }
+  .slab span { color:var(--accent-hover,#3d7256); }
+  .noteadd { display:flex; gap:8px; align-items:flex-end; margin-bottom:10px; }
+  .noteadd textarea { flex:1; resize:vertical; border:1px solid var(--border); border-radius:8px; padding:7px 9px; font-size:12.5px; font-family:inherit; color:var(--text); background:var(--paper,#fbfcf9); outline:none; }
+  .noteadd textarea:focus { border-color:var(--accent); }
+  .nbtn { flex-shrink:0; font-size:12px; color:#fff; background:var(--accent,#4f8a6b); border:0; border-radius:8px; padding:8px 12px; cursor:pointer; }
+  .nbtn:hover { background:var(--accent-hover,#3d7256); }
+  .slist { list-style:none; margin:0; padding:0; display:flex; flex-direction:column; gap:5px; }
+  .sitem { display:flex; align-items:baseline; gap:8px; padding:6px 8px; border-radius:8px; background:var(--paper,#fbfcf9); border:1px solid var(--border); }
+  .skind { flex-shrink:0; font-size:9.5px; font-weight:700; border-radius:4px; padding:1px 6px; }
+  .k-note { color:#3d7256; background:#e3efe2; border:1px solid #cfe3cd; }
+  .k-highlight { color:#8a6306; background:#faf3e2; border:1px solid #ecdca3; }
+  .k-card { color:#1d4ed8; background:#eef4fc; border:1px solid #d7e4f7; }
+  .stext { flex:1; font-size:12px; line-height:1.5; color:var(--text); white-space:pre-wrap; word-break:break-word; }
+  .sdel { flex-shrink:0; background:none; border:0; color:var(--faint,#9aa090); cursor:pointer; font-size:14px; }
+  .sdel:hover { color:var(--error,#c0392b); }
+  .shint { font-size:11.5px; color:var(--faint,#9aa090); margin:0; }
+  @media(max-width:820px){ .idx{display:none} .read{padding:24px 18px} .conn{grid-template-columns:1fr} .studybar{position:static;flex-direction:row} .crumbs{max-width:30%} .search input{width:150px} }
+</style>

frontend/src/routes/documents/[id]/+page.svelte

@@ -16,6 +16,7 @@
   import Skeleton from '$lib/components/ui/Skeleton.svelte';
   import HandwriteCanvas from '$lib/components/HandwriteCanvas.svelte';
   import MarkdownDoc from '$lib/components/MarkdownDoc.svelte';
+  import RelatedDocs from '$lib/components/RelatedDocs.svelte';
   import { renderDocMarkdown } from '$lib/utils/docMarkdown';
   import MarkdownStatusBadge from '$lib/components/MarkdownStatusBadge.svelte';
   import NoteEditor from '$lib/components/editors/NoteEditor.svelte';
@@ -321,6 +322,7 @@
 <!-- ════ 우 슬림 레일 (시안 카드 스타일) ════ -->
 {#snippet rail()}
   <div style="display:flex;flex-direction:column;gap:11px;font-size:14px;">
+    <RelatedDocs documentId={doc.id} />
     {#if doc.ai_tldr || doc.ai_summary}
       <div style="background:#f4f7f1;border:1px solid #dde3d6;border-radius:14px;padding:13px;">
         <div style="font-size:10.5px;font-weight:700;color:#697061;letter-spacing:.4px;margin-bottom:7px;">TL;DR</div>

frontend/src/routes/safety/+layout.svelte

@@ -0,0 +1,34 @@
+<script>
+  // 안전 자료실 (safety-library-1 Phase 3) — 재해/법령·지침/서적·표준·매뉴얼 3탭.
+  import { page } from '$app/stores';
+
+  const TABS = [
+    { href: '/safety/incidents', label: '재해사례' },
+    { href: '/safety/laws', label: '법령·지침' },
+    { href: '/safety/materials', label: '서적·표준·매뉴얼' },
+  ];
+</script>
+
+<div class="max-w-5xl mx-auto px-4 py-5 flex flex-col gap-4">
+  <header>
+    <h1 class="text-lg font-bold text-text">안전 자료실</h1>
+    <p class="text-xs text-dim mt-0.5">재해사례·법령·지침·표준 — 자료유형(material_type) 축 기반</p>
+  </header>
+
+  <nav class="flex gap-1 border-b border-default" aria-label="안전 자료실 탭">
+    {#each TABS as tab}
+      <a
+        href={tab.href}
+        aria-current={$page.url.pathname === tab.href ? 'page' : undefined}
+        class="px-3 py-2 text-sm font-medium border-b-2 -mb-px transition-colors
+          {$page.url.pathname === tab.href
+            ? 'border-accent text-accent'
+            : 'border-transparent text-dim hover:text-text'}"
+      >
+        {tab.label}
+      </a>
+    {/each}
+  </nav>
+
+  <slot />
+</div>

frontend/src/routes/safety/+page.svelte

@@ -0,0 +1,9 @@
+<script>
+  // /safety 진입 = 재해 탭 redirect (plan: +page=재해 탭 redirect)
+  import { onMount } from 'svelte';
+  import { goto } from '$app/navigation';
+
+  onMount(() => {
+    goto('/safety/incidents', { replaceState: true });
+  });
+</script>

frontend/src/routes/safety/SafetyDocList.svelte

@@ -0,0 +1,75 @@
+<script>
+  // 안전 자료실 공용 목록 — material_type + jurisdiction 필터로 GET /documents/ 조회.
+  // C-1 계약: material_type 지정 = 기본 exclude(news·law_monitor·note) 해제 (documents.py list_documents).
+  import { api } from '$lib/api';
+  import { addToast } from '$lib/stores/toast';
+  import DocumentCard from '$lib/components/DocumentCard.svelte';
+
+  let { materialType, jurisdiction = '' } = $props();
+
+  const PAGE_SIZE = 20;
+  let docs = $state([]);
+  let total = $state(0);
+  let nextPage = $state(1);
+  let loading = $state(false);
+
+  async function load(reset = false) {
+    loading = true;
+    const pageToLoad = reset ? 1 : nextPage;
+    try {
+      const params = new URLSearchParams();
+      params.set('material_type', materialType);
+      if (jurisdiction) params.set('jurisdiction', jurisdiction);
+      params.set('page', String(pageToLoad));
+      params.set('page_size', String(PAGE_SIZE));
+      const result = await api(`/documents/?${params}`);
+      docs = reset ? result.items : [...docs, ...result.items];
+      total = result.total;
+      nextPage = pageToLoad + 1;
+    } catch {
+      addToast('error', '안전 자료 로딩 실패');
+    } finally {
+      loading = false;
+    }
+  }
+
+  $effect(() => {
+    // 필터 변경 시 1페이지부터 재조회 (materialType/jurisdiction 읽기 = 반응 트리거)
+    void materialType;
+    void jurisdiction;
+    docs = [];
+    load(true);
+  });
+
+  let hasMore = $derived(docs.length < total);
+</script>
+
+<div class="flex flex-col gap-2">
+  {#if !loading || docs.length > 0}
+    <p class="text-xs text-dim tabular-nums">총 {total.toLocaleString()}건</p>
+  {/if}
+
+  {#if docs.length > 0}
+    <div class="flex flex-col gap-2">
+      {#each docs as doc (doc.id)}
+        <DocumentCard {doc} />
+      {/each}
+    </div>
+  {:else if !loading}
+    <div class="py-12 text-center text-sm text-dim">
+      해당 조건의 자료가 없습니다.
+    </div>
+  {/if}
+
+  {#if loading}
+    <div class="py-6 text-center text-sm text-dim">불러오는 중…</div>
+  {:else if hasMore}
+    <button
+      type="button"
+      onclick={() => load(false)}
+      class="self-center px-4 py-1.5 rounded-md text-sm text-dim border border-default hover:bg-surface hover:text-text transition-colors"
+    >
+      더 보기 ({docs.length}/{total.toLocaleString()})
+    </button>
+  {/if}
+</div>

frontend/src/routes/safety/incidents/+page.svelte

@@ -0,0 +1,29 @@
+<script>
+  // 재해사례 탭 — material_type=incident (KOSHA 사고사망·재해사례·CSB 등).
+  // 케이스 그룹핑(boardno 본문+첨부 1카드)은 API 확장 필요라 후속(DS freeze 하 백엔드 무변경).
+  import SafetyDocList from '../SafetyDocList.svelte';
+
+  const JURISDICTIONS = [
+    { value: '', label: '전체' },
+    { value: 'KR', label: 'KR' },
+    { value: 'US', label: 'US' },
+  ];
+  let jurisdiction = $state('');
+</script>
+
+<div class="flex flex-col gap-3">
+  <div class="flex items-center gap-1.5" role="group" aria-label="관할 필터">
+    {#each JURISDICTIONS as j}
+      <button
+        type="button"
+        onclick={() => (jurisdiction = j.value)}
+        class="px-2.5 py-1 rounded-full text-xs font-medium transition-colors
+          {jurisdiction === j.value ? 'bg-accent/15 text-accent' : 'text-dim hover:bg-surface hover:text-text'}"
+      >
+        {j.label}
+      </button>
+    {/each}
+  </div>
+
+  <SafetyDocList materialType="incident" {jurisdiction} />
+</div>

frontend/src/routes/safety/laws/+page.svelte

@@ -0,0 +1,48 @@
+<script>
+  // 법령·지침 탭 — 법령(law, 버전체인 current 만 코퍼스 노출) / 지침(guide, KOSHA GUIDE 등).
+  // 법령 기본 관할 = KR (plan: country 누락 = KR 정규화). version_status 뱃지는 API 확장 후속.
+  import SafetyDocList from '../SafetyDocList.svelte';
+
+  const KINDS = [
+    { value: 'law', label: '법령' },
+    { value: 'guide', label: '지침' },
+  ];
+  const JURISDICTIONS = [
+    { value: 'KR', label: 'KR' },
+    { value: 'US', label: 'US' },
+    { value: '', label: '전체' },
+  ];
+  let kind = $state('law');
+  let jurisdiction = $state('KR');
+</script>
+
+<div class="flex flex-col gap-3">
+  <div class="flex items-center justify-between flex-wrap gap-2">
+    <div class="flex items-center gap-1" role="group" aria-label="자료유형">
+      {#each KINDS as k}
+        <button
+          type="button"
+          onclick={() => (kind = k.value)}
+          class="px-3 py-1 rounded-md text-sm font-medium transition-colors
+            {kind === k.value ? 'bg-accent/15 text-accent' : 'text-dim hover:bg-surface hover:text-text'}"
+        >
+          {k.label}
+        </button>
+      {/each}
+    </div>
+    <div class="flex items-center gap-1.5" role="group" aria-label="관할 필터">
+      {#each JURISDICTIONS as j}
+        <button
+          type="button"
+          onclick={() => (jurisdiction = j.value)}
+          class="px-2.5 py-1 rounded-full text-xs font-medium transition-colors
+            {jurisdiction === j.value ? 'bg-accent/15 text-accent' : 'text-dim hover:bg-surface hover:text-text'}"
+        >
+          {j.label}
+        </button>
+      {/each}
+    </div>
+  </div>
+
+  <SafetyDocList materialType={kind} {jurisdiction} />
+</div>

frontend/src/routes/safety/materials/+page.svelte

@@ -0,0 +1,29 @@
+<script>
+  // 서적·표준·매뉴얼 탭 — 필터 프리셋(전용 뷰는 50건+ 게이트 뒤, plan Phase 3).
+  import SafetyDocList from '../SafetyDocList.svelte';
+
+  const KINDS = [
+    { value: 'standard', label: '표준 (NB 등)' },
+    { value: 'book', label: '서적' },
+    { value: 'manual', label: '매뉴얼' },
+    { value: 'paper', label: '논문' },
+  ];
+  let kind = $state('standard');
+</script>
+
+<div class="flex flex-col gap-3">
+  <div class="flex items-center gap-1" role="group" aria-label="자료유형">
+    {#each KINDS as k}
+      <button
+        type="button"
+        onclick={() => (kind = k.value)}
+        class="px-3 py-1 rounded-md text-sm font-medium transition-colors
+          {kind === k.value ? 'bg-accent/15 text-accent' : 'text-dim hover:bg-surface hover:text-text'}"
+      >
+        {k.label}
+      </button>
+    {/each}
+  </div>
+
+  <SafetyDocList materialType={kind} />
+</div>

frontend/src/routes/study/+page.svelte

@@ -1,13 +1,58 @@
 <script>
-  // /study — 학습 hub.
-  // 주제로 보기(퀴즈·복습·통계) / 자료 학습 / 필사 세션 / 암기카드 검수.
+  // /study — 학습 hub + 데일리 랜딩('오늘의 공부' 대시보드).
+  // 상단 = 이론 홈(진도·오늘의 개념·복습 due, 재노출 트리거). 하단 = 기존 모드 진입.
   import { onMount } from 'svelte';
   import { api } from '$lib/api';
-  import { BookOpen, PenLine, GraduationCap, FolderKanban, Layers, Repeat, Flag, Inbox, Activity } from 'lucide-svelte';
+  import { addToast } from '$lib/stores/toast';
+  import { BookOpen, PenLine, GraduationCap, FolderKanban, Layers, Repeat, Flag, Inbox, Activity, CalendarCheck, Target } from 'lucide-svelte';
 
   let cardReviewCount = $state(0);
   let questionFlagCount = $state(0);
+
+  // 오늘의 공부 (이론 홈)
+  let curriculum = $state(null);
+  let todayConcepts = $state([]);
+  let weakConcepts = $state([]);        // 약점 개념(관련 기출 정답률 낮음)
+  let dashLoading = $state(true);
+
+  let readPct = $derived(
+    curriculum && curriculum.total ? Math.round((curriculum.read / curriculum.total) * 100) : 0
+  );
+
+  async function loadDashboard() {
+    dashLoading = true;
+    try {
+      const [cur, today] = await Promise.all([
+        api('/study/curriculum'),
+        api('/study/today-concepts?limit=6'),
+      ]);
+      curriculum = cur;
+      todayConcepts = today?.concepts ?? [];
+    } catch {
+      // 코어 대시보드 실패해도 허브 나머지는 동작 (조용히)
+    } finally {
+      dashLoading = false;
+    }
+    // 약점 개념 = 비차단(신규 엔드포인트 실패해도 코어 대시보드 블랙아웃 방지)
+    try {
+      const weak = await api('/study/concepts/weakness-map?limit=5');
+      weakConcepts = weak?.weak ?? [];
+    } catch {}
+  }
+
+  async function markRead(doc) {
+    try {
+      await api(`/study/concepts/${doc.doc_id}/read`, { method: 'POST' });
+      todayConcepts = todayConcepts.filter((c) => c.doc_id !== doc.doc_id);
+      addToast('success', `회독: ${doc.title}`);
+      loadDashboard(); // 진도 갱신
+    } catch {
+      addToast('error', '회독 처리 실패');
+    }
+  }
+
   onMount(async () => {
+    loadDashboard();
     try {
       const r = await api('/study-cards/needs-review/count');
       cardReviewCount = r?.count ?? 0;
@@ -27,6 +72,80 @@
     <p class="text-sm text-dim mt-1">주제별 퀴즈·복습(SRS)·통계 / 학습 자료 회독 / 손글씨 필사 세션.</p>
   </header>
 
+  <!-- 오늘의 공부 (이론 홈 대시보드 = 데일리 트리거) -->
+  <section class="mb-5 rounded-lg border border-default bg-surface p-4 md:p-5">
+    <div class="flex items-center gap-2 mb-3">
+      <CalendarCheck size={18} class="text-accent" />
+      <h2 class="text-base font-semibold text-text">오늘의 공부</h2>
+      {#if curriculum}
+        <span class="ml-auto text-xs text-dim">이론 회독 <span class="text-text font-medium">{curriculum.read}</span> / {curriculum.total} ({readPct}%)</span>
+      {/if}
+    </div>
+
+    {#if dashLoading}
+      <p class="text-xs text-dim">불러오는 중…</p>
+    {:else}
+      {#if curriculum}
+        <div class="h-2 rounded-full bg-bg overflow-hidden mb-3">
+          <div class="h-full bg-accent" style="width: {readPct}%"></div>
+        </div>
+        <div class="flex flex-wrap gap-x-4 gap-y-1 mb-4 text-xs text-dim">
+          {#each curriculum.subjects as s}
+            <span>{s.subject} <span class="text-text">{s.read}/{s.total}</span></span>
+          {/each}
+        </div>
+
+        <div class="flex flex-wrap gap-2 mb-4">
+          <a
+            href="/study/topics/{curriculum.topic_id}/review-queue"
+            class="flex items-center gap-1.5 rounded border border-default px-3 py-1.5 text-xs text-dim hover:border-accent hover:text-text transition-colors"
+          >
+            <Repeat size={13} /> 문항 복습 <span class="font-semibold text-text">{curriculum.question_due}</span>
+          </a>
+          <span class="flex items-center gap-1.5 rounded border border-default px-3 py-1.5 text-xs text-dim">
+            <BookOpen size={13} /> 개념 재복습 <span class="font-semibold text-text">{curriculum.concept_due}</span>
+          </span>
+        </div>
+      {/if}
+
+      <div class="text-xs text-dim mb-2">오늘의 개념</div>
+      {#if todayConcepts.length === 0}
+        <p class="text-xs text-dim">오늘 볼 개념이 없습니다. 잘 하고 있어요.</p>
+      {:else}
+        <ul class="space-y-1.5">
+          {#each todayConcepts as c (c.doc_id)}
+            <li class="flex items-center gap-2 rounded border border-default px-3 py-2">
+              <span class="text-accent shrink-0 text-xs" title="빈출">{#each Array(c.freq) as _}★{/each}</span>
+              <a href="/study/read/{c.doc_id}" class="text-sm text-text hover:text-accent truncate flex-1">{c.title}</a>
+              <span class="shrink-0 text-[10px] rounded-full px-2 py-0.5 {c.reason === '재복습' ? 'bg-accent/15 text-accent' : 'bg-surface border border-default text-dim'}">{c.reason}</span>
+              <button
+                type="button"
+                onclick={() => markRead(c)}
+                class="shrink-0 text-xs rounded border border-default px-2 py-1 text-dim hover:border-accent hover:text-accent transition-colors"
+              >읽음</button>
+            </li>
+          {/each}
+        </ul>
+      {/if}
+
+      {#if weakConcepts.length > 0}
+        <div class="mt-4 pt-3 border-t border-default">
+          <div class="text-xs text-dim mb-2 flex items-center gap-1.5">
+            <Target size={13} class="text-error" /> 약점 개념 <span class="text-faint">(관련 기출 정답률 낮음)</span>
+          </div>
+          <div class="flex flex-wrap gap-2">
+            {#each weakConcepts as w (w.doc_id)}
+              <a href="/study/read/{w.doc_id}"
+                class="text-xs rounded-full border border-error/40 bg-error/10 text-error px-3 py-1 hover:bg-error/20 transition-colors">
+                {w.title.replace(/^\d+_/, '')} <span class="font-semibold">{w.accuracy}%</span>
+              </a>
+            {/each}
+          </div>
+        </div>
+      {/if}
+    {/if}
+  </section>
+
   <a
     href="/study/topics"
     class="block mb-3 p-5 rounded-lg border border-default bg-surface hover:border-accent hover:bg-accent/5 transition-colors"
@@ -126,7 +245,8 @@
   <div class="mt-6 p-4 rounded-lg border border-dashed border-default/60 text-xs text-dim">
     <div class="font-medium text-dim mb-1">예정</div>
     <ul class="list-disc list-inside space-y-0.5">
-      <li>애플워치 빠른복습 + 공부 알람(push)</li>
+      <li>개념 학습 리더 (가리고 떠올리기 · 빈출★ · 관련개념 백링크)</li>
+      <li>이론↔문제 연결 (개념별 정답률 · 약점 개념 지도)</li>
     </ul>
   </div>
 </div>

frontend/src/routes/study/read/[docId]/+page.svelte

@@ -0,0 +1,254 @@
+<script>
+  /**
+   * /study/read/[docId] — 개념 학습 리더.
+   * 개념노트(가스기사 documents)를 구조(요약/본문/빈출★/관련개념)로 렌더 +
+   * '떠올리기' 능동 회상 토글 + 회독 SR(POST read) + 관련개념 백링크 + 이전/다음.
+   * 본문 렌더 = MarkdownDoc(KaTeX + docimg 내장). 서버 파싱 = /api/study/concepts/{id}.
+   */
+  import { page } from '$app/stores';
+  import { api } from '$lib/api';
+  import { addToast } from '$lib/stores/toast';
+  import { renderMathMarkdownInline } from '$lib/utils/mathMarkdown';
+  import MarkdownDoc from '$lib/components/MarkdownDoc.svelte';
+  import Button from '$lib/components/ui/Button.svelte';
+  import EmptyState from '$lib/components/ui/EmptyState.svelte';
+  import Skeleton from '$lib/components/ui/Skeleton.svelte';
+  import { BookOpen, ArrowLeft, Eye, EyeOff, Check, ChevronLeft, ChevronRight, FileQuestion } from 'lucide-svelte';
+
+  let docId = $derived($page.params.docId);
+
+  let concept = $state(null);
+  let relatedQ = $state(null);          // 관련 기출(이론↔문제, 비차단)
+  let loading = $state(true);
+  let notFound = $state(false);
+  let mode = $state('read');            // 'read' | 'recall'(떠올리기)
+  let revealed = $state({});            // {sectionIndex: true}
+  let marking = $state(false);
+
+  const STAGE_LABEL = { 0: '복습 시작', 1: '복습 1단계', 2: '복습 2단계', 3: '복습 3단계', 4: '학습 완료' };
+  const OUTCOME_MARK = { correct: '○', wrong: '✕', unsure: '?' };
+  const OUTCOME_CLASS = { correct: 'text-success', wrong: 'text-error', unsure: 'text-warning' };
+  const outcomeMark = (o) => OUTCOME_MARK[o] ?? '–';
+  const outcomeClass = (o) => OUTCOME_CLASS[o] ?? 'text-faint';
+
+  async function load() {
+    const reqId = docId; // in-flight 가드: 백링크 연타 시 stale 응답 무시
+    loading = true;
+    notFound = false;
+    concept = null;
+    relatedQ = null;
+    revealed = {};
+    mode = 'read';
+    try {
+      const data = await api(`/study/concepts/${reqId}`);
+      if (reqId !== docId) return; // 그새 다른 개념으로 이동 → 폐기
+      concept = data;
+    } catch (e) {
+      if (reqId !== docId) return;
+      if (e?.status === 404) notFound = true;
+      else addToast('error', '개념을 불러오지 못했습니다');
+      return; // 본문 실패 → 관련기출 스킵
+    } finally {
+      if (reqId === docId) loading = false;
+    }
+    // 관련 기출(비차단 — 실패해도 본문 표시엔 영향 없음)
+    try {
+      const rq = await api(`/study/concepts/${reqId}/questions?limit=6`);
+      if (reqId === docId) relatedQ = rq;
+    } catch {}
+  }
+
+  // $effect 가 마운트 1회 + docId 변경(백링크/이전·다음) 재로드를 모두 커버 (onMount 불필요)
+  $effect(() => {
+    void docId;
+    load();
+  });
+
+  function toggleMode() {
+    mode = mode === 'read' ? 'recall' : 'read';
+    revealed = {};
+  }
+  function reveal(i) {
+    revealed = { ...revealed, [i]: true };
+  }
+  function shown(i) {
+    return mode === 'read' || revealed[i];
+  }
+
+  async function markRead() {
+    marking = true;
+    try {
+      const r = await api(`/study/concepts/${docId}/read`, { method: 'POST' });
+      if (concept) {
+        concept.is_read = true;
+        concept.review_stage = r?.review_stage ?? concept.review_stage;
+        concept.due_at = r?.due_at ?? concept.due_at;
+      }
+      addToast('success', '회독 완료 — 다음 복습에 다시 나옵니다');
+    } catch {
+      addToast('error', '회독 처리 실패');
+    } finally {
+      marking = false;
+    }
+  }
+</script>
+
+<svelte:head><title>{concept?.title ?? '개념'} — 공부</title></svelte:head>
+
+<div class="p-4 md:p-6 max-w-3xl mx-auto">
+  <!-- 상단 네비 -->
+  <div class="flex items-center gap-2 text-xs md:text-sm mb-4 min-w-0">
+    <a href="/study" class="text-dim hover:text-text flex items-center gap-1 shrink-0">
+      <ArrowLeft size={14} /> 공부
+    </a>
+    {#if concept?.subject}
+      <span class="text-faint shrink-0">/</span>
+      <span class="text-dim truncate">{concept.subject}</span>
+    {/if}
+  </div>
+
+  {#if loading}
+    <Skeleton h="h-10" rounded="card" />
+    <div class="mt-3 space-y-2">
+      {#each Array(4) as _}<Skeleton h="h-24" rounded="card" />{/each}
+    </div>
+  {:else if notFound}
+    <EmptyState icon={BookOpen} title="개념을 찾을 수 없습니다" description="삭제되었거나 잘못된 주소입니다." />
+  {:else if concept}
+    <!-- 제목 + 빈출 tier -->
+    <header class="mb-3">
+      <div class="flex items-start gap-2">
+        <h1 class="text-xl md:text-2xl font-semibold text-text flex-1">{concept.title}</h1>
+        <span class="text-accent text-sm shrink-0 mt-1" title="빈출도">
+          {#each Array(concept.freq) as _}★{/each}
+        </span>
+      </div>
+      {#if concept.is_read || (concept.review_stage !== null && concept.review_stage !== undefined)}
+        <div class="mt-1 text-xs text-dim">
+          {#if concept.review_stage !== null && concept.review_stage !== undefined}
+            {STAGE_LABEL[concept.review_stage] ?? '복습 중'}
+          {:else}회독함{/if}
+        </div>
+      {/if}
+    </header>
+
+    <!-- 한 줄 요약 (고정 표시) -->
+    {#if concept.summary}
+      <div class="mb-4 rounded-lg border-l-4 border-accent bg-accent/10 px-4 py-3 markdown-body text-sm text-text">
+        {@html renderMathMarkdownInline(concept.summary)}
+      </div>
+    {/if}
+
+    <!-- 모드 토글 -->
+    <div class="flex items-center gap-2 mb-4">
+      <Button variant={mode === 'recall' ? 'primary' : 'secondary'} size="sm" icon={mode === 'recall' ? EyeOff : Eye} onclick={toggleMode}>
+        {mode === 'recall' ? '떠올리기 모드' : '읽기 모드'}
+      </Button>
+      {#if mode === 'recall'}
+        <span class="text-xs text-dim">각 섹션을 떠올린 뒤 확인하세요</span>
+      {/if}
+    </div>
+
+    <!-- 본문 섹션 -->
+    {#if concept.body.length > 0}
+      <div class="space-y-3 mb-5">
+        {#each concept.body as sec, i (i)}
+          <section class="rounded-lg border border-default bg-surface overflow-hidden">
+            <div class="flex items-center gap-2 px-4 py-2.5 border-b border-default bg-surface-hover">
+              <h2 class="text-sm font-semibold text-text flex-1">{sec.label}</h2>
+              {#if sec.stars > 0}
+                <span class="text-accent text-xs shrink-0">{#each Array(sec.stars) as _}★{/each}</span>
+              {/if}
+            </div>
+            {#if shown(i)}
+              <div class="px-4 py-3">
+                <MarkdownDoc documentId={concept.doc_id} mdContent={sec.md} mdStatus={null}
+                  class="markdown-body max-w-none text-text" />
+              </div>
+            {:else}
+              <button type="button" onclick={() => reveal(i)}
+                class="w-full px-4 py-6 text-center text-sm text-dim hover:text-accent hover:bg-accent/5 transition-colors">
+                <Eye size={16} class="inline mr-1" /> 떠올린 뒤 확인
+              </button>
+            {/if}
+          </section>
+        {/each}
+      </div>
+    {/if}
+
+    <!-- 빈출 포인트 -->
+    {#if concept.bincheol.length > 0}
+      <section class="mb-5 rounded-lg border border-default bg-surface p-4">
+        <h2 class="text-sm font-semibold text-text mb-2 flex items-center gap-1.5">
+          <span class="text-accent">★</span> 빈출 포인트
+        </h2>
+        <ul class="space-y-1.5">
+          {#each concept.bincheol as item}
+            <li class="flex gap-2 text-sm text-text">
+              <span class="text-accent shrink-0 text-xs mt-0.5">{#each Array(item.tier || 1) as _}★{/each}</span>
+              <span class="markdown-body flex-1">{@html renderMathMarkdownInline(item.text)}</span>
+            </li>
+          {/each}
+        </ul>
+      </section>
+    {/if}
+
+    <!-- 관련 개념 (백링크) -->
+    {#if concept.related.length > 0}
+      <section class="mb-5">
+        <h2 class="text-xs text-dim mb-2">관련 개념</h2>
+        <div class="flex flex-wrap gap-2">
+          {#each concept.related as rel}
+            {#if rel.doc_id}
+              <a href="/study/read/{rel.doc_id}"
+                class="text-xs rounded-full border border-accent/40 bg-accent/10 text-accent px-3 py-1 hover:bg-accent/20 transition-colors">
+                {rel.phrase}
+              </a>
+            {:else}
+              <span class="text-xs rounded-full border border-default bg-surface text-faint px-3 py-1" title="아직 없는 개념">
+                {rel.phrase}
+              </span>
+            {/if}
+          {/each}
+        </div>
+      </section>
+    {/if}
+
+    <!-- 관련 기출 (이론↔문제 브리지) -->
+    {#if relatedQ && relatedQ.linked > 0}
+      <section class="mb-5 rounded-lg border border-default bg-surface p-4">
+        <h2 class="text-sm font-semibold text-text mb-2 flex items-center gap-1.5">
+          <FileQuestion size={15} class="text-accent" /> 관련 기출
+          <span class="ml-1 text-xs font-normal text-dim">
+            {relatedQ.linked}문항{#if relatedQ.accuracy !== null} · 정답률 <span class="{relatedQ.accuracy < 60 ? 'text-error' : 'text-text'} font-medium">{relatedQ.accuracy}%</span>{:else} · 아직 안 풂{/if}
+          </span>
+        </h2>
+        <ul class="space-y-0.5">
+          {#each relatedQ.questions as q (q.id)}
+            <li>
+              <a href="/study/topics/4/questions/{q.id}"
+                class="flex items-center gap-2 text-xs py-1 text-dim hover:text-accent transition-colors">
+                <span class="{outcomeClass(q.last_outcome)} shrink-0 w-4 text-center font-bold">{outcomeMark(q.last_outcome)}</span>
+                <span class="truncate">{q.subject ?? '기출'}{#if q.exam_round} · {q.exam_round}{/if}</span>
+              </a>
+            </li>
+          {/each}
+        </ul>
+      </section>
+    {/if}
+
+    <!-- 액션바 -->
+    <div class="flex items-center gap-2 border-t border-default pt-4 mt-2">
+      {#if concept.prev_id}
+        <Button variant="ghost" size="sm" icon={ChevronLeft} href="/study/read/{concept.prev_id}">이전</Button>
+      {/if}
+      <div class="flex-1"></div>
+      <Button variant="primary" size="sm" icon={Check} onclick={markRead} loading={marking}>
+        {concept.is_read ? '다시 회독' : '회독 완료'}
+      </Button>
+      {#if concept.next_id}
+        <Button variant="secondary" size="sm" icon={ChevronRight} href="/study/read/{concept.next_id}">다음 개념</Button>
+      {/if}
+    </div>
+  {/if}
+</div>

migrations/378_publish_outbox_attempts_failed.sql

@@ -0,0 +1,9 @@
+-- 378_publish_outbox_attempts_failed.sql
+-- (번호: 멀티세션 중 prod 가 377_domain_bucket 을 선점 → 378 로 리넘버.)
+-- publish_outbox poison row head-of-line block 차단. 발행 워커가 행별 savepoint 격리 후
+-- 예외 시 attempts++ 하고 MAX 초과 시 failed_at 스탬프(terminal) → 그 행을 select 에서 제외해
+-- 후속 발행이 막히지 않게 함. 기존 미처리 행은 attempts=0 / failed_at=NULL 로 정상 재처리.
+-- (단일 ALTER = 1 statement = asyncpg prepared 호환.)
+ALTER TABLE publish_outbox
+    ADD COLUMN IF NOT EXISTS attempts SMALLINT NOT NULL DEFAULT 0,
+    ADD COLUMN IF NOT EXISTS failed_at TIMESTAMPTZ;

migrations/379_asme_clause_kb.sql

@@ -0,0 +1,37 @@
+-- 379_asme_clause_kb.sql
+-- ASME 절-지식베이스: 절 = 개별 documents 행(parent_id) + 절↔절 백링크 + 태깅 (additive, idempotent)
+-- 검색 무접촉: 절 doc 은 embedding NULL(벡터 제외) + doc_kind='clause'(retrieval doc-leg 필터로 제외).
+
+ALTER TABLE documents
+  ADD COLUMN IF NOT EXISTS parent_id    bigint REFERENCES documents(id),
+  ADD COLUMN IF NOT EXISTS doc_kind     text NOT NULL DEFAULT 'standard',
+  ADD COLUMN IF NOT EXISTS clause_code  text,
+  ADD COLUMN IF NOT EXISTS clause_part  text,
+  ADD COLUMN IF NOT EXISTS clause_order int;
+
+CREATE INDEX IF NOT EXISTS idx_documents_parent_id   ON documents(parent_id) WHERE parent_id IS NOT NULL;
+CREATE INDEX IF NOT EXISTS idx_documents_doc_kind    ON documents(doc_kind);
+CREATE INDEX IF NOT EXISTS idx_documents_clause_code ON documents(clause_code) WHERE clause_code IS NOT NULL;
+
+-- 절↔절 백링크 (dangling 허용: dst_doc_id nullable)
+CREATE TABLE IF NOT EXISTS clause_links (
+  id          bigserial PRIMARY KEY,
+  src_doc_id  bigint NOT NULL REFERENCES documents(id) ON DELETE CASCADE,
+  dst_code    text   NOT NULL,
+  dst_doc_id  bigint REFERENCES documents(id) ON DELETE SET NULL,
+  anchor      text,
+  ctx         text,
+  char_off    int
+);
+CREATE INDEX IF NOT EXISTS idx_clause_links_src     ON clause_links(src_doc_id);
+CREATE INDEX IF NOT EXISTS idx_clause_links_dst     ON clause_links(dst_doc_id) WHERE dst_doc_id IS NOT NULL;
+CREATE INDEX IF NOT EXISTS idx_clause_links_dstcode ON clause_links(dst_code);
+
+-- 태깅 (Part 자동 + 주제)
+CREATE TABLE IF NOT EXISTS document_tags (
+  doc_id    bigint NOT NULL REFERENCES documents(id) ON DELETE CASCADE,
+  tag       text   NOT NULL,
+  tag_kind  text   NOT NULL DEFAULT 'topic',
+  PRIMARY KEY (doc_id, tag)
+);
+CREATE INDEX IF NOT EXISTS idx_document_tags_tag ON document_tags(tag);

migrations/380_clause_study.sql

@@ -0,0 +1,9 @@
+-- 380_clause_study.sql — 절-문서 공부도구(노트/형광펜/암기카드) 저장. FK 없음(documents 락 회피).
+CREATE TABLE IF NOT EXISTS clause_study (
+  id         bigserial PRIMARY KEY,
+  doc_id     bigint NOT NULL,
+  kind       text   NOT NULL,            -- 'note' | 'highlight' | 'card'
+  payload    jsonb  NOT NULL DEFAULT '{}',
+  created_at timestamptz NOT NULL DEFAULT now()
+);
+CREATE INDEX IF NOT EXISTS idx_clause_study_doc ON clause_study(doc_id, kind);

migrations/381_study_concept_progress.sql

@@ -0,0 +1,16 @@
+-- 381_study_concept_progress.sql — 이론 개념(문서) 간격반복(SR) 진행. 이론공부 홈 트리거.
+-- concept_doc_id 는 documents.id 를 가리키나 FK 미설정(hot 테이블 락 회피, clause_study 380 선례).
+-- SR 산술은 study_question_progress 와 동일(sr_schedule 공용): stage 0→1→2→3(1·3·7·14일)→4 졸업.
+CREATE TABLE IF NOT EXISTS study_concept_progress (
+  id             bigserial PRIMARY KEY,
+  user_id        bigint NOT NULL REFERENCES users(id) ON DELETE CASCADE,
+  study_topic_id bigint NOT NULL REFERENCES study_topics(id) ON DELETE CASCADE,
+  concept_doc_id bigint NOT NULL,
+  review_stage   smallint,
+  due_at         timestamptz,
+  last_read_at   timestamptz,
+  created_at     timestamptz NOT NULL DEFAULT now(),
+  updated_at     timestamptz NOT NULL DEFAULT now(),
+  CONSTRAINT uq_concept_progress_user_doc UNIQUE (user_id, concept_doc_id)
+);
+CREATE INDEX IF NOT EXISTS idx_concept_progress_due ON study_concept_progress(user_id, due_at) WHERE due_at IS NOT NULL;

migrations/382_study_concept_links.sql

@@ -0,0 +1,15 @@
+-- 382_study_concept_links.sql — 개념문서 ↔ 기출문항 링크 (이론↔문제 브리지, Stage B).
+-- concept_doc_id=documents.id, question_id=study_questions.id — FK 없음(hot 테이블 락 회피, 선례).
+-- link_source: 'embedding'(bge-m3 코사인 top-k, 주력) | 'ref'(해설 .md 참조, 후속 enrichment).
+-- score=코사인 유사도(0~1). UNIQUE(doc,question,source) — source별 공존 허용(재튜닝=source 전삭제 후 재삽입).
+CREATE TABLE IF NOT EXISTS study_concept_links (
+  id             bigserial PRIMARY KEY,
+  concept_doc_id bigint NOT NULL,
+  question_id    bigint NOT NULL,
+  link_source    text   NOT NULL,
+  score          double precision,
+  created_at     timestamptz NOT NULL DEFAULT now(),
+  CONSTRAINT uq_concept_link UNIQUE (concept_doc_id, question_id, link_source)
+);
+CREATE INDEX IF NOT EXISTS idx_concept_links_doc ON study_concept_links(concept_doc_id);
+CREATE INDEX IF NOT EXISTS idx_concept_links_q ON study_concept_links(question_id);

scripts/asme_backlinks_persist.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+"""ASME clause-KB backlinks: resolve clause-id mentions in each clause doc -> clause_links.
+dst resolved to the clause doc of the same parent (top-level code); sub-code mention -> anchor;
+unresolved (cross-standard / material spec not split) -> dangling (dst_doc_id NULL).
+Idempotent per parent. Usage: python3 asme_backlinks_persist.py <parent_id> [--commit]
+"""
+import asyncio, os, re, sys
+
+MENTION_RE = re.compile(r'(?<![A-Za-z0-9])([A-Z]{1,4}-\d+(?:\.\d+)*[A-Za-z]?)(?![A-Za-z0-9])')
+def top(code): return re.match(r'^[A-Z]{1,4}-\d+', code).group(0)
+
+async def main():
+    parent = int(sys.argv[1]); commit = '--commit' in sys.argv
+    import asyncpg
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    docs = await conn.fetch("SELECT id, clause_code, md_content FROM documents "
+                            "WHERE parent_id=$1 AND doc_kind='clause' ORDER BY clause_order", parent)
+    code2id = {d['clause_code']: d['id'] for d in docs}
+    edges = []          # (src_id, dst_code, dst_doc_id, anchor, ctx, char_off)
+    resolved = dangling = 0
+    for d in docs:
+        body = d['md_content']; src_top = d['clause_code']
+        seen = set()
+        for m in MENTION_RE.finditer(body):
+            code = m.group(1); t = top(code)
+            if t == src_top: continue                 # self-reference
+            if (d['id'], code) in seen: continue      # dedup per (src,dst_code)
+            seen.add((d['id'], code))
+            dst_id = code2id.get(t)                    # resolve to same-parent clause doc
+            anchor = code.lower().replace('.', '-') if code != t else None
+            off = m.start()
+            ctx = re.sub(r'\s+', ' ', body[max(0, off-50):off+50]).strip()
+            edges.append((d['id'], code, dst_id, anchor, ctx, off))
+            if dst_id: resolved += 1
+            else: dangling += 1
+    print(f"parent={parent} clause_docs={len(docs)} edges={len(edges)} resolved={resolved} dangling={dangling}")
+    # top referenced clauses
+    from collections import Counter
+    tgt = Counter(top(e[1]) for e in edges if e[2])
+    print("most-referenced:", tgt.most_common(8))
+    if not commit:
+        print("DRY-RUN. pass --commit to persist."); await conn.close(); return
+    async with conn.transaction():
+        ids = [d['id'] for d in docs]
+        await conn.execute("DELETE FROM clause_links WHERE src_doc_id = ANY($1::bigint[])", ids)
+        await conn.executemany(
+            "INSERT INTO clause_links(src_doc_id,dst_code,dst_doc_id,anchor,ctx,char_off) "
+            "VALUES ($1,$2,$3,$4,$5,$6)", edges)
+        n = await conn.fetchval("SELECT count(*) FROM clause_links WHERE src_doc_id = ANY($1::bigint[])", ids)
+        print(f"COMMITTED: {n} clause_links for parent {parent}")
+    await conn.close()
+
+asyncio.run(main())

scripts/asme_clause_persist.py

@@ -0,0 +1,118 @@
+#!/usr/bin/env python3
+"""ASME clause-KB persist (v2: over-CAP pagination). Split a parent standard into per-clause
+documents (A-granularity); over-CAP clause bodies are paginated into readable page-docs.
+Idempotent per parent. doc_kind='clause', embedding NULL (search-excluded), parent_id=<parent>.
+Usage: python3 asme_clause_persist.py <parent_id> [--commit]
+"""
+import asyncio, os, re, sys, hashlib, statistics
+
+CAP = 12000; PAGE_TOK = 11000
+EN, KO = 0.217, 0.529
+LINE_RE = re.compile(r'^([ \t#>*]{0,8})([A-Z]{2,4}-\d+(?:\.\d+)*[A-Za-z]?)(.*)$')
+MENTION_RE = re.compile(r'(?<![A-Za-z0-9])([A-Z]{1,4}-\d+(?:\.\d+)*[A-Za-z]?)(?![A-Za-z0-9])')
+EXACT_TOP = re.compile(r'^[A-Z]{2,4}-\d+$')
+TITLE_AFTER = re.compile(r'^[\s.]*[A-Z(]')
+REF_LEAD = re.compile(r'^[\s.]*(and|or|to|of|in|on|the|as|is|are|shall|through|per|see|with|'
+                      r'for|by|that|which|such|또는|및|등|의|은|는|에|을|를|과|와)\b', re.I)
+
+def tok(s):
+    ko = sum(1 for c in s if '가' <= c <= '힣'); return int((len(s)-ko)*EN + ko*KO)
+
+def clean_title(rest):
+    t = re.sub(r'<sup>ð</sup>\s*\**\d*\**\s*<sup>Þ</sup>', '', rest)
+    t = re.sub(r'ð\**\d*\**Þ', '', t)
+    t = t.replace('**', '').replace('#', '')
+    return re.sub(r'\s+', ' ', t).strip(' *:—-')
+
+def is_header(markup, rest):
+    if '#' in markup or '*' in markup: return True
+    rs = rest.strip()
+    if rs == '': return True
+    if REF_LEAD.match(rest): return False
+    if rs[0] in ',;.)': return False
+    if '가' <= rs[0] <= '힣': return False
+    if rs[0].islower(): return False
+    return bool(TITLE_AFTER.match(rs))
+
+def paginate(body):
+    """split an over-CAP body into <=MAX_PAGES line-aligned pages of ~PAGE_TOK tokens."""
+    pages, cur, ct = [], [], 0
+    for ln in body.split('\n'):
+        lt = tok(ln) + 1
+        if ct + lt > PAGE_TOK and cur:
+            pages.append('\n'.join(cur)); cur, ct = [ln], lt
+        else:
+            cur.append(ln); ct += lt
+    if cur: pages.append('\n'.join(cur))
+    return pages
+
+def build_clauses(text):
+    lines = text.split('\n'); off = []; a = 0
+    for ln in lines: off.append(a); a += len(ln) + 1
+    bounds = []; seen = set()
+    for i, ln in enumerate(lines):
+        m = LINE_RE.match(ln)
+        if not m: continue
+        markup, code, rest = m.group(1), m.group(2), m.group(3)
+        if not EXACT_TOP.match(code): continue
+        if not is_header(markup, rest): continue
+        if code in seen: continue
+        seen.add(code); bounds.append((off[i], code, clean_title(rest)))
+    raw = []
+    for idx, (start, code, title) in enumerate(bounds):
+        end = bounds[idx+1][0] if idx+1 < len(bounds) else len(text)
+        body = text[start:end]
+        part = re.match(r'^[A-Z]{2,4}', code).group(0)
+        links = sorted(set(re.match(r'^[A-Z]{1,4}-\d+', mm).group(0)
+                           for mm in MENTION_RE.findall(body)) - {code})
+        raw.append(dict(code=code, part=part, title=(code + (' ' + title if title else '')),
+                        body=body, tok=tok(body), links=links))
+    # expand over-CAP into pages; assign running clause_order
+    final, order = [], 0
+    for c in raw:
+        if c['tok'] <= CAP:
+            final.append({**c, 'order': order}); order += 1; continue
+        pages = paginate(c['body'])
+        for pi, pb in enumerate(pages):
+            code = c['code'] if pi == 0 else f"{c['code']}·p{pi+1}"
+            title = c['title'] if pi == 0 else f"{c['title']} (페이지 {pi+1}/{len(pages)})"
+            final.append(dict(code=code, part=c['part'], order=order, title=title,
+                              body=pb, tok=tok(pb), links=c['links'] if pi == 0 else []))
+            order += 1
+    return final
+
+async def main():
+    parent = int(sys.argv[1]); commit = '--commit' in sys.argv
+    import asyncpg
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    row = await conn.fetchrow("SELECT md_content, ai_domain, data_origin FROM documents WHERE id=$1", parent)
+    if not row: print(f"parent {parent} not found"); return
+    clauses = build_clauses(row['md_content'])
+    toks = [c['tok'] for c in clauses]
+    over = [c for c in clauses if c['tok'] > CAP]
+    print(f"parent={parent} clause_docs={len(clauses)} median_tok={int(statistics.median(toks))} "
+          f"max_tok={max(toks)} over_cap_remaining={len(over)}")
+    if over: print("still over-CAP:", [f"{c['code']}:{c['tok']}t" for c in over])
+    if not commit:
+        print("DRY-RUN. pass --commit to persist."); await conn.close(); return
+    async with conn.transaction():
+        deld = await conn.execute("DELETE FROM documents WHERE parent_id=$1 AND doc_kind='clause'", parent)
+        print("deleted prior:", deld)
+        for c in clauses:
+            fh = hashlib.sha256(f"{parent}:{c['code']}:{c['body']}".encode()).hexdigest()
+            cid = await conn.fetchval("""
+                INSERT INTO documents
+                  (file_format, file_hash, title, md_content, parent_id, doc_kind,
+                   clause_code, clause_part, clause_order, ai_domain, data_origin,
+                   md_status, review_status, conversion_status, preview_status)
+                VALUES ('md',$1,$2,$3,$4,'clause',$5,$6,$7,$8,$9,'success','approved','none','none')
+                RETURNING id
+            """, fh, c['title'], c['body'], parent, c['code'], c['part'], c['order'],
+                 row['ai_domain'], row['data_origin'] or 'external')
+            await conn.execute("INSERT INTO document_tags(doc_id,tag,tag_kind) VALUES ($1,$2,'part') "
+                               "ON CONFLICT DO NOTHING", cid, c['part'])
+        n = await conn.fetchval("SELECT count(*) FROM documents WHERE parent_id=$1 AND doc_kind='clause'", parent)
+        print(f"COMMITTED: {n} clause docs for parent {parent}")
+    await conn.close()
+
+asyncio.run(main())

scripts/backfill_publish_card_progress.py

@@ -31,8 +31,8 @@ from core.database import async_session
 from models.study_memo_card_progress import StudyMemoCardProgress
 from services.study.publish_enqueue import backfill_publish_card_progress
 
-# 개인 학습툴 progress row 대비 넉넉. 도달 시 가드 경보.
-PAGE = 100000
+# 페이지 배치 크기 — after_id 루프로 전량 처리(bounded tx).
+PAGE = 5000
 
 
 async def run(dry_run: bool) -> None:
@@ -50,13 +50,17 @@ async def run(dry_run: bool) -> None:
         print("[dry-run] 적재 안 함. 실제 실행은 --dry-run 제거.")
         return
 
-    async with async_session() as session:
-        n = await backfill_publish_card_progress(session, after_id=0, limit=PAGE)
-        await session.commit()
+    total = 0
+    after = 0
+    while True:
+        async with async_session() as session:
+            n, after = await backfill_publish_card_progress(session, after_id=after, limit=PAGE)
+            await session.commit()
+        total += n
+        if n < PAGE:
+            break
 
-    print(f"\n[ok] outbox 적재 {n}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
-    if n >= PAGE:
-        print(f"[warn] PAGE({PAGE}) 도달 — progress 가 더 있을 수 있음. after_id 페이징 추가 필요.")
+    print(f"\n[ok] outbox 적재 {total}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
 
 
 def main() -> None:

scripts/backfill_publish_cards.py

@@ -31,8 +31,8 @@ from core.database import async_session
 from models.study_memo_card import StudyMemoCard
 from services.study.publish_enqueue import backfill_publish_cards
 
-# 개인 학습툴 카드 수 대비 넉넉(단일 outbox 적재 tx, 워커는 BATCH_SIZE 로 drain). 도달 시 가드 경보.
-PAGE = 100000
+# 페이지 배치 크기 — after_id 루프로 전량 처리(bounded tx). 워커는 BATCH_SIZE 로 drain.
+PAGE = 5000
 
 
 async def run(dry_run: bool) -> None:
@@ -55,13 +55,17 @@ async def run(dry_run: bool) -> None:
         print("[dry-run] 적재 안 함. 실제 실행은 --dry-run 제거.")
         return
 
-    async with async_session() as session:
-        n = await backfill_publish_cards(session, after_id=0, limit=PAGE)
-        await session.commit()
+    total = 0
+    after = 0
+    while True:
+        async with async_session() as session:
+            n, after = await backfill_publish_cards(session, after_id=after, limit=PAGE)
+            await session.commit()
+        total += n
+        if n < PAGE:
+            break
 
-    print(f"\n[ok] outbox 적재 {n}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
-    if n >= PAGE:
-        print(f"[warn] PAGE({PAGE}) 도달 — 카드가 더 있을 수 있음. after_id 페이징 추가 필요.")
+    print(f"\n[ok] outbox 적재 {total}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
 
 
 def main() -> None:

scripts/backfill_publish_topics.py

@@ -37,7 +37,7 @@ from core.database import async_session
 from models.study_topic import StudyTopic
 from services.study.publish_enqueue import backfill_publish_topics
 
-# 개인 학습툴 주제 수 대비 넉넉. 도달 시 overflow 가드가 경보.
+# 페이지 배치 크기 — after_id 루프로 전량 처리(bounded tx).
 PAGE = 5000
 
 
@@ -58,13 +58,17 @@ async def run(dry_run: bool) -> None:
         print("[dry-run] 적재 안 함. 실제 실행은 --dry-run 제거.")
         return
 
-    async with async_session() as session:
-        n = await backfill_publish_topics(session, after_id=0, limit=PAGE)
-        await session.commit()
+    total = 0
+    after = 0
+    while True:
+        async with async_session() as session:
+            n, after = await backfill_publish_topics(session, after_id=after, limit=PAGE)
+            await session.commit()
+        total += n
+        if n < PAGE:
+            break
 
-    print(f"\n[ok] outbox 적재 {n}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
-    if n >= PAGE:
-        print(f"[warn] PAGE({PAGE}) 도달 — 주제가 더 있을 수 있음. after_id 페이징 추가 필요.")
+    print(f"\n[ok] outbox 적재 {total}건 — 발행 워커가 drain(flag on 시) 하며 rev 부여.")
 
 
 def main() -> None:

scripts/ci/boot_smoke.py

@@ -106,7 +106,7 @@ async def main() -> None:
             "SELECT count(*) FROM pg_indexes WHERE indexname='uq_attempt_session_question'"))).scalar()
         mx = (await s.execute(text("SELECT max(version) FROM schema_migrations"))).scalar()
     print(f"SCHEMA OK — max_migration={mx} documents={docs} purge_col={purge} cand_qwen={cand} attempt_uq={uq}")
-    assert docs and purge == 1 and cand == 0 and uq == 1 and mx == 361, "FAIL: 기대 스키마 상태 불일치"
+    assert docs and purge == 1 and cand == 0 and uq == 1 and mx == 378, "FAIL: 기대 스키마 상태 불일치"
 
     # ── 5) /health 직접 호출 ──────────────────────────────────────────────
     health = await main.health_check()

scripts/concept_links_backfill.sql

@@ -0,0 +1,23 @@
+-- concept_links_backfill.sql — 개념↔문항 임베딩 링크 재생성 (Stage B, 멱등·재실행 안전).
+-- 정찰 확정: bge-m3 1024d 코사인, per-concept top-k=10, threshold 0.62 → ~2362링크·284/289개념·964문항.
+-- 재튜닝 시 DELETE(embedding 소스만) 후 재삽입 = ref 링크(후속) 불변. 개념 doc = 가스기사 태그.
+DELETE FROM study_concept_links WHERE link_source = 'embedding';
+INSERT INTO study_concept_links (concept_doc_id, question_id, link_source, score)
+WITH cd AS (
+  SELECT id, embedding FROM documents
+  WHERE user_tags::text LIKE '%@library/가스기사/%'
+    AND deleted_at IS NULL AND embedding IS NOT NULL
+),
+ranked AS (
+  SELECT cd.id AS concept_doc_id, q.id AS question_id,
+         1 - (q.embedding <=> cd.embedding) AS score,
+         row_number() OVER (PARTITION BY cd.id ORDER BY q.embedding <=> cd.embedding) AS rn
+  FROM cd
+  JOIN study_questions q
+    ON q.study_topic_id = 4 AND q.embedding IS NOT NULL
+   AND q.deleted_at IS NULL AND q.is_active
+)
+SELECT concept_doc_id, question_id, 'embedding', score
+FROM ranked
+WHERE rn <= 10 AND score >= 0.62
+ON CONFLICT (concept_doc_id, question_id, link_source) DO NOTHING;

scripts/guide_clause_persist.py

@@ -0,0 +1,100 @@
+#!/usr/bin/env python3
+"""기술지침(KOSHA guide) 절-KB persist: 번호섹션(# 1. 목적 / ## 4.1) 단위 분해 + 제본.
+ASME/법령과 동일 clause-KB 모델(doc_kind='clause', parent_id=지침, 검색제외, /book 리더 공용).
+Usage: python3 guide_clause_persist.py <id|all> [--commit]
+"""
+import asyncio, os, re, sys, hashlib, statistics
+
+CAP = 12000; PAGE_TOK = 11000
+EN, KO = 0.217, 0.529
+# 번호섹션 헤더: '# 1. 목 적', '## 4.1 누출...'  (번호 1~3자리=연도(4자리) 배제)
+ART_RE = re.compile(r'^#{1,6}\s*(\d{1,3}(?:\.\d{1,3})*)\.?\s+(\S.*)$')
+TOP_RE = re.compile(r'^\d{1,3}$')
+# 외부 표준/법규 참조(대부분 dangling): ASME B16.5 · KS B 1501 · 규칙 제N조
+EXT_RE = re.compile(r'(ASME\s+[A-Z][0-9.]+|KS\s+[A-Z]\s*[0-9]+|ISO\s+[0-9]+|제\d+조)')
+
+def tok(s):
+    ko = sum(1 for c in s if '가' <= c <= '힣'); return int((len(s)-ko)*EN + ko*KO)
+
+def build_sections(text):
+    lines = text.split('\n'); off = []; a = 0
+    for ln in lines: off.append(a); a += len(ln) + 1
+    bounds = []; seen = set()
+    for i, ln in enumerate(lines):
+        m = ART_RE.match(ln)
+        if not m: continue
+        code, name = m.group(1), m.group(2).strip()
+        if not TOP_RE.match(code): continue       # top-level 번호섹션만 경계
+        if code in seen: continue
+        if len(name) < 1: continue
+        seen.add(code); bounds.append((off[i], code, name))
+    out = []
+    for idx, (start, code, name) in enumerate(bounds):
+        end = bounds[idx+1][0] if idx+1 < len(bounds) else len(text)
+        body = text[start:end].strip()
+        ext = sorted(set(EXT_RE.findall(body)))[:8]
+        out.append(dict(code=code, part='본문', order=0, title=f"{code}. {name}"[:120],
+                        body=body, tok=tok(body), links=[], ext=ext))
+    # over-CAP 페이지네이션 + 순번
+    final, order = [], 0
+    for c in out:
+        if c['tok'] <= CAP:
+            final.append({**c, 'order': order}); order += 1; continue
+        pages, cur, ct = [], [], 0
+        for ln in c['body'].split('\n'):
+            lt = tok(ln)+1
+            if ct+lt > PAGE_TOK and cur: pages.append('\n'.join(cur)); cur=[ln]; ct=lt
+            else: cur.append(ln); ct+=lt
+        if cur: pages.append('\n'.join(cur))
+        for pi, pb in enumerate(pages):
+            final.append(dict(code=c['code'] if pi==0 else f"{c['code']}·p{pi+1}", part='본문',
+                              order=order, title=c['title'] if pi==0 else f"{c['title']} (p{pi+1})",
+                              body=pb, tok=tok(pb), links=[], ext=[]))
+            order += 1
+    return final
+
+async def process_one(conn, gid, commit, verbose=True):
+    row = await conn.fetchrow("SELECT title, md_content, ai_domain, data_origin FROM documents WHERE id=$1", gid)
+    if not row: return ('notfound', 0)
+    if not row['md_content']: return ('nullmd', 0)
+    secs = build_sections(row['md_content'])
+    if len(secs) < 2: return ('few', len(secs))     # 섹션 2 미만 = 번호구조 아님
+    toks = [c['tok'] for c in secs]
+    if verbose:
+        print(f"guide={gid} «{(row['title'] or '')[:40]}» 섹션={len(secs)} median={int(statistics.median(toks))} max={max(toks)}")
+        print("  샘플:", [c['title'][:26] for c in secs[:7]])
+    if not commit: return ('dry', len(secs))
+    async with conn.transaction():
+        await conn.execute("DELETE FROM clause_links WHERE src_doc_id IN (SELECT id FROM documents WHERE parent_id=$1 AND doc_kind='clause')", gid)
+        await conn.execute("DELETE FROM documents WHERE parent_id=$1 AND doc_kind='clause'", gid)
+        for c in secs:
+            fh = hashlib.sha256(f"{gid}:{c['code']}:{c['body']}".encode()).hexdigest()
+            cid = await conn.fetchval("""
+                INSERT INTO documents (file_format,file_hash,title,md_content,parent_id,doc_kind,
+                  clause_code,clause_part,clause_order,ai_domain,data_origin,
+                  md_status,review_status,conversion_status,preview_status)
+                VALUES ('md',$1,$2,$3,$4,'clause',$5,$6,$7,$8,$9,'success','approved','none','none') RETURNING id
+            """, fh, c['title'], c['body'], gid, c['code'], c['part'], c['order'], row['ai_domain'], row['data_origin'] or 'external')
+            await conn.execute("INSERT INTO document_tags(doc_id,tag,tag_kind) VALUES ($1,'기술지침','kind') ON CONFLICT DO NOTHING", cid)
+        n = await conn.fetchval("SELECT count(*) FROM documents WHERE parent_id=$1 AND doc_kind='clause'", gid)
+        print(f"  COMMITTED: {n} 섹션 for guide {gid}")
+    return ('committed', len(secs))
+
+async def main():
+    import asyncpg
+    arg = sys.argv[1]; commit = '--commit' in sys.argv
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    if arg == 'all':
+        gs = await conn.fetch("SELECT id FROM documents WHERE material_type='guide' AND doc_kind='standard' "
+                              "AND deleted_at IS NULL AND md_content IS NOT NULL ORDER BY id")
+        agg = {}; tot = 0
+        for i, r in enumerate(gs):
+            st, n = await process_one(conn, r['id'], commit, verbose=False)
+            agg[st] = agg.get(st, 0)+1; tot += n if st in ('dry','committed') else 0
+            if commit and (i+1) % 40 == 0: print(f"  …{i+1}/{len(gs)} (누적섹션 {tot})")
+        print(f"BATCH {'COMMIT' if commit else 'DRY'} guides={len(gs)} status={agg} 총섹션={tot}")
+    else:
+        await process_one(conn, int(arg), commit, verbose=True)
+    await conn.close()
+
+asyncio.run(main())

scripts/law_clause_persist.py

@@ -0,0 +1,146 @@
+#!/usr/bin/env python3
+"""법령 조-KB persist: 법령을 조(條) 단위 개별 문서로 분해 + 조↔조 백링크 + 장(章) 태그.
+ASME clause-KB와 동일 모델(doc_kind='clause', parent_id=법령, embedding NULL, 검색제외).
+법령 추출 노이즈(조 앞 ### 메타 반복) 트림. Usage: python3 law_clause_persist.py <law_id> [--commit]
+"""
+import asyncio, os, re, sys, hashlib, statistics
+
+CAP = 12000; PAGE_TOK = 11000
+EN, KO = 0.217, 0.529
+# 조 헤더: '### 제3조의2(가스안전관리...) 본문'
+ART_RE = re.compile(r'^#{0,6}\s*(제\d+조(?:의\d+)?)\s*\(([^)]*)\)\s*(.*)$')
+CHAP_RE = re.compile(r'^#{1,6}\s*(제\d+장(?:의\d+)?)\s*(.*)$')         # 장 = part
+# 같은-법 조 멘션(백링크)
+MENTION_RE = re.compile(r'제\d+조(?:의\d+)?')
+# 타법 참조: 「법명」 ... 제N조
+EXTLAW_RE = re.compile(r'「([^」]+)」')
+
+def tok(s):
+    ko = sum(1 for c in s if '가' <= c <= '힣'); return int((len(s)-ko)*EN + ko*KO)
+def art_code(c): return c  # '제3조의2'
+
+def build_articles(text):
+    lines = text.split('\n'); off = []; a = 0
+    for ln in lines: off.append(a); a += len(ln) + 1
+    arts = []      # (line_idx, code, name, part)
+    cur_part = None
+    for i, ln in enumerate(lines):
+        ch = CHAP_RE.match(ln)
+        if ch and not ART_RE.match(ln):
+            cur_part = (ch.group(1) + (' ' + ch.group(2).strip() if ch.group(2).strip() else '')).strip()
+            continue
+        m = ART_RE.match(ln)
+        if m:
+            arts.append((i, m.group(1), m.group(2).strip(), cur_part))
+    # 본문 슬라이스 + 다음 조 앞 메타 노이즈 트림
+    out = []
+    for idx, (li, code, name, part) in enumerate(arts):
+        end_li = arts[idx+1][0] if idx+1 < len(arts) else len(lines)
+        body_lines = lines[li:end_li]
+        # 트림: 끝에서부터 '### {짧은 메타}' (조번호/조문/날짜/제목, [개정] 제N조 아님) 제거
+        while len(body_lines) > 1:
+            last = body_lines[-1].strip()
+            if last == '':
+                body_lines.pop(); continue
+            mh = re.match(r'^#{1,6}\s+(.*)$', last)
+            if mh:
+                c = mh.group(1).strip()
+                if not c.startswith('[') and not c.startswith('제') and (
+                        c in ('조문', 'N') or re.fullmatch(r'\d+', c) or re.fullmatch(r'\d{8}', c) or len(c) <= 30):
+                    body_lines.pop(); continue
+            break
+        body = '\n'.join(body_lines).strip()
+        links = sorted(set(MENTION_RE.findall(body)) - {code})
+        ext = sorted(set(EXTLAW_RE.findall(body)))[:6]
+        out.append(dict(code=code, part=part or '본칙', order=0,
+                        title=f"{code}({name})" if name else code,
+                        body=body, tok=tok(body), links=links, ext=ext))
+    # 페이지네이션(over-CAP) + 순번
+    final, order = [], 0
+    for c in out:
+        if c['tok'] <= CAP:
+            final.append({**c, 'order': order}); order += 1; continue
+        # 11K 토큰 라인 단위 분할
+        pages, cur, ct = [], [], 0
+        for ln in c['body'].split('\n'):
+            lt = tok(ln)+1
+            if ct+lt > PAGE_TOK and cur: pages.append('\n'.join(cur)); cur=[ln]; ct=lt
+            else: cur.append(ln); ct+=lt
+        if cur: pages.append('\n'.join(cur))
+        for pi, pb in enumerate(pages):
+            final.append(dict(code=c['code'] if pi==0 else f"{c['code']}·p{pi+1}", part=c['part'],
+                              order=order, title=c['title'] if pi==0 else f"{c['title']} (p{pi+1}/{len(pages)})",
+                              body=pb, tok=tok(pb), links=c['links'] if pi==0 else [], ext=[]))
+            order += 1
+    return final
+
+async def process_one(conn, law, commit, verbose=True):
+    row = await conn.fetchrow("SELECT title, coalesce(md_content, extracted_text) AS md_content, ai_domain, data_origin FROM documents WHERE id=$1", law)
+    if not row: return ('notfound', 0, 0)
+    if not row['md_content']: return ('nullmd', 0, 0)
+    arts = build_articles(row['md_content'])
+    if not arts: return ('noart', 0, 0)
+    toks = [c['tok'] for c in arts]
+    nlink = sum(len(c['links']) for c in arts)
+    if verbose:
+        parts = {}
+        for c in arts: parts[c['part']] = parts.get(c['part'], 0)+1
+        print(f"law={law} «{(row['title'] or '')[:34]}» 조문={len(arts)} median={int(statistics.median(toks))} "
+              f"max={max(toks)} 장={len(parts)} 백링크={nlink}")
+        print("  샘플:", [c['title'][:22] for c in arts[:6]])
+    if not commit:
+        return ('dry', len(arts), nlink)
+    async with conn.transaction():
+        await conn.execute(
+            "DELETE FROM clause_links WHERE src_doc_id IN (SELECT id FROM documents WHERE parent_id=$1 AND doc_kind='clause')", law)
+        await conn.execute("DELETE FROM documents WHERE parent_id=$1 AND doc_kind='clause'", law)
+        code2id = {}
+        for c in arts:
+            fh = hashlib.sha256(f"{law}:{c['code']}:{c['body']}".encode()).hexdigest()
+            cid = await conn.fetchval("""
+                INSERT INTO documents (file_format,file_hash,title,md_content,parent_id,doc_kind,
+                  clause_code,clause_part,clause_order,ai_domain,data_origin,
+                  md_status,review_status,conversion_status,preview_status)
+                VALUES ('md',$1,$2,$3,$4,'clause',$5,$6,$7,$8,$9,'success','approved','none','none') RETURNING id
+            """, fh, c['title'], c['body'], law, c['code'], c['part'], c['order'],
+                 row['ai_domain'], row['data_origin'] or 'external')
+            code2id[c['code']] = cid
+            await conn.execute("INSERT INTO document_tags(doc_id,tag,tag_kind) VALUES ($1,$2,'chapter') ON CONFLICT DO NOTHING", cid, c['part'])
+        # 조↔조 백링크 (같은 법 내부; 타법 참조는 dangling)
+        edges = []
+        for c in arts:
+            src = code2id[c['code']]
+            for dst in c['links']:
+                edges.append((src, dst, code2id.get(dst), None, None, None))
+        if edges:
+            await conn.executemany(
+                "INSERT INTO clause_links(src_doc_id,dst_code,dst_doc_id,anchor,ctx,char_off) VALUES ($1,$2,$3,$4,$5,$6)", edges)
+        n = await conn.fetchval("SELECT count(*) FROM documents WHERE parent_id=$1 AND doc_kind='clause'", law)
+        print(f"  COMMITTED: {n} 조문 + {len(edges)} 백링크 for law {law}")
+    return ('committed', n, len(edges))
+
+
+async def main():
+    import asyncpg
+    arg = sys.argv[1]; commit = '--commit' in sys.argv
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    if arg == 'all':
+        laws = await conn.fetch("SELECT lm.document_id AS id FROM legal_meta lm "
+                                "JOIN documents d ON d.id=lm.document_id "
+                                "WHERE lm.law_doc_kind='primary' AND lm.version_status='current' "
+                                "AND coalesce(d.md_content, d.extracted_text) IS NOT NULL "
+                                "ORDER BY lm.document_id")
+        agg = {}; tot_art = tot_link = 0; zero = []
+        for i, r in enumerate(laws):
+            st, na, nl = await process_one(conn, r['id'], commit, verbose=False)
+            agg[st] = agg.get(st, 0) + 1
+            tot_art += na; tot_link += nl
+            if st == 'noart': zero.append(r['id'])
+            if commit and (i + 1) % 30 == 0: print(f"  …{i+1}/{len(laws)} (누적 조 {tot_art})")
+        print(f"BATCH {'COMMIT' if commit else 'DRY'} laws={len(laws)} status={agg} 총조문={tot_art} 총백링크={tot_link}")
+        if zero: print(f"  0-조(추출구조 이질) {len(zero)}건: {zero[:20]}")
+    else:
+        await process_one(conn, int(arg), commit, verbose=True)
+    await conn.close()
+
+asyncio.run(main())

scripts/paper_citation_analyze.py

@@ -0,0 +1,51 @@
+#!/usr/bin/env python3
+"""논문 인용그래프 가능성 측정(read-only) — 본문 DOI로 코퍼스내 인용 엣지 추정.
+own_doi = 헤더(앞 2500자) 첫 DOI / cited = References 이후(또는 전체) DOI. owner 맵 → 엣지.
+"""
+import asyncio, os, re, sys
+
+DOI_RE = re.compile(r'10\.\d{4,9}/[^\s"<>)\]\},;]+')
+REF_RE = re.compile(r'(references|참고문헌|bibliography|reference\s*list)', re.I)
+
+def norm(d): return d.rstrip('.').lower()
+
+async def main():
+    import asyncpg
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    rows = await conn.fetch("SELECT id, title, coalesce(md_content, extracted_text) AS txt FROM documents "
+                            "WHERE material_type='paper' AND doc_kind='standard' AND deleted_at IS NULL "
+                            "AND coalesce(md_content, extracted_text) IS NOT NULL")
+    owner = {}        # doi -> paper id (헤더 DOI = 그 논문 소유)
+    cited = {}        # paper id -> set(cited doi)
+    n_own = n_refsec = 0
+    for r in rows:
+        txt = r['txt']
+        head = txt[:2500]
+        hdois = [norm(d) for d in DOI_RE.findall(head)]
+        if hdois:
+            owner.setdefault(hdois[0], r['id']); n_own += 1
+        m = REF_RE.search(txt)
+        body = txt[m.start():] if m else ''
+        if m: n_refsec += 1
+        cds = set(norm(d) for d in DOI_RE.findall(body))
+        if cds: cited[r['id']] = cds
+    # 엣지: paper -> owner(cited doi)
+    edges = []
+    for pid, cds in cited.items():
+        for d in cds:
+            o = owner.get(d)
+            if o and o != pid: edges.append((pid, o, d))
+    cited_papers = set(e[0] for e in edges)
+    target_papers = set(e[1] for e in edges)
+    print(f"papers={len(rows)} 헤더DOI보유={n_own} References보유={n_refsec} owner_map={len(owner)}")
+    print(f"인용엣지(코퍼스내)={len(edges)} 인용하는논문={len(cited_papers)} 피인용논문={len(target_papers)}")
+    # 피인용 top
+    from collections import Counter
+    top = Counter(e[1] for e in edges).most_common(6)
+    if top:
+        idmap = {r['id']: r['title'] for r in rows}
+        print("피인용 top:")
+        for pid, c in top: print(f"  {c}회 ← {(idmap.get(pid) or '')[:48]}")
+    await conn.close()
+
+asyncio.run(main())

scripts/paper_openalex_match.py

@@ -0,0 +1,39 @@
+#!/usr/bin/env python3
+"""OpenAlex 고신뢰 매치율 측정 — References 보유 논문(학술 추정) 표본."""
+import asyncio, os, re
+
+def toks(s):
+    return set(re.findall(r'[a-z0-9]+', (s or '').lower()))
+def sim(a, b):
+    ta, tb = toks(a), toks(b)
+    if not ta or not tb: return 0.0
+    return len(ta & tb) / len(ta | tb)
+
+async def main():
+    import asyncpg, httpx
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    rows = await conn.fetch("SELECT id, title FROM documents WHERE material_type='paper' "
+                            "AND doc_kind='standard' AND deleted_at IS NULL AND title IS NOT NULL "
+                            "AND coalesce(md_content,extracted_text) ~* 'references|참고문헌' "
+                            "ORDER BY id LIMIT 40")
+    hi = mid = lo = 0; hits = []
+    async with httpx.AsyncClient(timeout=20) as client:
+        for r in rows:
+            title = re.sub(r'\s+', ' ', r['title']).strip()
+            try:
+                resp = await client.get("https://api.openalex.org/works",
+                    params={"search": title[:200], "per_page": 1, "mailto": "hyun49196@gmail.com"})
+                res = (resp.json().get("results") or [])
+                if not res: lo += 1; continue
+                s = sim(title, res[0].get("title"))
+                if s >= 0.6: hi += 1; hits.append((s, title[:40], (res[0].get('title') or '')[:40], res[0].get('cited_by_count'), len(res[0].get('referenced_works') or [])))
+                elif s >= 0.4: mid += 1
+                else: lo += 1
+            except Exception: lo += 1
+    print(f"표본={len(rows)} 고신뢰(≥0.6)={hi} 중간(0.4~0.6)={mid} 저신뢰/무매치={lo}")
+    print("고신뢰 매치 샘플:")
+    for s, a, b, cb, rf in hits[:8]:
+        print(f"  sim={s:.2f} cited={cb} refs={rf} | {a} ≈ {b}")
+    await conn.close()
+
+asyncio.run(main())

scripts/paper_openalex_test.py

@@ -0,0 +1,30 @@
+#!/usr/bin/env python3
+"""OpenAlex 보강 타당성 테스트 — 소수 논문 제목으로 매칭/메타 확인 (외부 API)."""
+import asyncio, os, re
+
+async def main():
+    import asyncpg, httpx
+    conn = await asyncpg.connect(os.environ['DATABASE_URL'].replace('+asyncpg', ''))
+    rows = await conn.fetch("SELECT id, title FROM documents WHERE material_type='paper' "
+                            "AND doc_kind='standard' AND deleted_at IS NULL AND title IS NOT NULL "
+                            "AND length(title) > 15 ORDER BY id LIMIT 6")
+    async with httpx.AsyncClient(timeout=20) as client:
+        for r in rows:
+            title = re.sub(r'\s+', ' ', r['title']).strip()
+            try:
+                resp = await client.get("https://api.openalex.org/works",
+                    params={"search": title[:200], "per_page": 1, "mailto": "hyun49196@gmail.com"})
+                js = resp.json()
+                res = (js.get("results") or [])
+                if not res:
+                    print(f"[{r['id']}] NO MATCH | {title[:50]}"); continue
+                w = res[0]
+                oid = (w.get("id") or "").split("/")[-1]
+                print(f"[{r['id']}] {title[:46]}")
+                print(f"   → OA {oid} | {(w.get('title') or '')[:46]} | {w.get('publication_year')} | "
+                      f"cited_by={w.get('cited_by_count')} | refs={len(w.get('referenced_works') or [])} | doi={w.get('doi')}")
+            except Exception as e:
+                print(f"[{r['id']}] ERROR {type(e).__name__}: {e}")
+    await conn.close()
+
+asyncio.run(main())

services/mineru/server.py

@@ -59,6 +59,11 @@ MAX_IMAGES_PER_DOC = int(os.getenv("MINERU_MAX_IMAGES_PER_DOC", "200"))
 MAX_BYTES_PER_IMAGE = int(os.getenv("MINERU_MAX_BYTES_PER_IMAGE", str(10 * 1024 * 1024)))
 MAX_PAGES_HARD = int(os.getenv("MINERU_MAX_PAGES_HARD", "200"))  # 1-shot max_pages 안전장치
 
+# self-timeout — 변환/워밍이 vLLM 행으로 _engine_lock 을 영구 점유해 서비스가 wedge 되는 것을 차단.
+# (클라이언트 marker_worker 는 300s 로 포기하나 서버측 inflight 는 자동 취소 안 됨 → 서버 자체 상한 필요.)
+PARSE_TIMEOUT_S = float(os.getenv("MINERU_PARSE_TIMEOUT_S", "600"))
+WARMUP_TIMEOUT_S = float(os.getenv("MINERU_WARMUP_TIMEOUT_S", "1200"))  # 최초 모델 다운로드(~2.4GB) 여유
+
 _PRELOAD = os.getenv("MINERU_PRELOAD", "1") != "0"
 
 # ---- 엔진 상태 ---------------------------------------------------------------
@@ -68,6 +73,15 @@ _warmup_error: str | None = None
 _engine_lock = asyncio.Lock()
 
 
+def _is_engine_fatal(exc: BaseException) -> bool:
+    """OOM/CUDA 류 = 엔진 상태 오염 가능 → 재워밍 강제 대상(타임아웃은 호출측에서 별도 판정)."""
+    s = f"{type(exc).__name__} {exc}".lower()
+    return any(
+        k in s
+        for k in ("out of memory", "oom", "cuda", "cublas", "device-side", "illegal memory")
+    )
+
+
 async def _run_mineru(pdf_bytes: bytes, lang: str) -> tuple[str, list[dict]]:
     """슬라이스된 PDF bytes → (markdown, 이미지 dict 리스트). **async 엔진 경로.**
 
@@ -148,7 +162,7 @@ async def _ensure_warmup() -> None:
             page.insert_text((72, 72), "MinerU warmup.")
             warmup_bytes = doc.tobytes()
             doc.close()
-            await _run_mineru(warmup_bytes, MINERU_LANG)
+            await asyncio.wait_for(_run_mineru(warmup_bytes, MINERU_LANG), timeout=WARMUP_TIMEOUT_S)
             _warmup_done = True
             _warmup_error = None
             logger.info(f"[mineru-service] warmup done engine_version={_engine_version}")
@@ -274,6 +288,7 @@ def _serialize_images(images: list[dict], src_path: str) -> tuple[list[ConvertIm
 
 @app.post("/convert", response_model=ConvertResponse)
 async def convert(req: ConvertRequest):
+    global _warmup_done
     p = _resolve_path(req.file_path)
     if p is None or not p.is_file():
         raise HTTPException(404, detail={"code": "file_not_found", "message": req.file_path})
@@ -288,10 +303,18 @@ async def convert(req: ConvertRequest):
     async with _engine_lock:                # 실제 변환 직렬화(단일 GPU)
         start = time.monotonic()
         try:
-            md_text, raw_images = await _run_mineru(pdf_bytes, MINERU_LANG)
+            md_text, raw_images = await asyncio.wait_for(
+                _run_mineru(pdf_bytes, MINERU_LANG), timeout=PARSE_TIMEOUT_S
+            )
         except HTTPException:
             raise
         except Exception as exc:
+            # 타임아웃(엔진 행) 또는 OOM/CUDA 류면 엔진 오염 가능 → 다음 요청이 재워밍하도록 리셋.
+            # 재워밍까지 실패하면 _ensure_warmup 이 _warmup_error 설정 → /ready 503 → healthcheck
+            # 재시작으로 escalate(영구 degradation 차단). 일시 OOM 이면 재워밍 성공 후 정상화.
+            if isinstance(exc, (asyncio.TimeoutError, TimeoutError)) or _is_engine_fatal(exc):
+                _warmup_done = False
+                logger.error("[mineru-service] engine reset (timeout/fatal) path=%s: %s", p, exc)
             logger.exception(f"[mineru-service] conversion failed path={p}: {exc}")
             raise HTTPException(422, detail={"code": "conversion_failed",
                 "message": f"{type(exc).__name__}: {exc}"}) from exc

tests/summarize_units/test_mapreduce_helpers.py

@@ -0,0 +1,80 @@
+"""summarize_units PR2 헬퍼 단위테스트 — map/reduce 프롬프트 조립 순수함수.
+
+핵심 불변식:
+  - render_map_slice: 유닛 위치(1-based)/섹션 라벨 + 본문 그대로 (손실 0).
+  - build_reduce_units_block: 어떤 입력에도 반환 블록 est_tokens <= budget (캡 초과 0
+    검증 게이트의 reduce 측). 절단은 detail 만 — 라벨/TLDR/불일치/순서 보존.
+
+pytest + 단독 실행 양쪽 지원:
+  PYTHONPATH=. pytest tests/summarize_units/ -q
+"""
+from __future__ import annotations
+
+from app.services.summarize_units import (
+    SummarizeUnit,
+    build_reduce_units_block,
+    estimate_tokens,
+    render_map_slice,
+)
+
+
+def _result(idx: int, detail: str, *, tldr: str = "요약", inc: list | None = None) -> dict:
+    return {
+        "index": idx,
+        "titles": [f"섹션{idx}"],
+        "tldr": tldr,
+        "detail": detail,
+        "inconsistencies": inc or [],
+    }
+
+
+# ---------- render_map_slice ----------
+
+def test_render_map_slice_label_and_body():
+    unit = SummarizeUnit(index=2, section_titles=["개요", None, "본론"], text="본문입니다")
+    out = render_map_slice(unit, total_units=5)
+    assert out.startswith("[유닛 3/5 — 섹션: 개요 · 본론]\n")
+    assert out.endswith("본문입니다")
+
+
+def test_render_map_slice_untitled():
+    unit = SummarizeUnit(index=0, section_titles=[None], text="x")
+    assert "(무제 구간)" in render_map_slice(unit, total_units=1)
+
+
+# ---------- build_reduce_units_block ----------
+
+def test_reduce_block_within_budget_untouched():
+    results = [_result(i, "가" * 100) for i in range(3)]
+    block, truncated = build_reduce_units_block(results, budget_tokens=11_000)
+    assert not truncated
+    # 순서/라벨/TLDR 보존
+    assert block.index("[유닛 1/3") < block.index("[유닛 2/3") < block.index("[유닛 3/3")
+    assert "TLDR: 요약" in block
+    assert "가" * 100 in block
+
+
+def test_reduce_block_truncates_to_budget():
+    # 유닛 8개 × 한글 detail 5,000자 ≈ 21K tok — budget 5,000 으로 절단 강제
+    results = [_result(i, "가" * 5_000) for i in range(8)]
+    block, truncated = build_reduce_units_block(results, budget_tokens=5_000)
+    assert truncated
+    assert estimate_tokens(block) <= 5_000
+    # 라벨(유닛 순서)은 절단 후에도 보존
+    assert "[유닛 1/8" in block
+
+
+def test_reduce_block_hard_cut_floor():
+    # min_detail_chars floor 에 막혀 비례 절단으로 불충분한 극단 케이스 — 하드 컷 발동
+    results = [_result(i, "가" * 300) for i in range(50)]
+    block, truncated = build_reduce_units_block(results, budget_tokens=500)
+    assert truncated
+    assert estimate_tokens(block) <= 500
+
+
+def test_reduce_block_preserves_inconsistencies():
+    results = [
+        _result(0, "가" * 50, inc=[{"kind": "version_drift", "desc": "개정판 차이"}]),
+    ]
+    block, _ = build_reduce_units_block(results, budget_tokens=10_000)
+    assert "불일치(version_drift): 개정판 차이" in block

tests/summarize_units/test_units.py

@@ -0,0 +1,180 @@
+"""summarize_units 단위테스트 (presegment PR1 — 순수함수·fixture).
+
+핵심 불변식:
+  - estimate_tokens = PR0 캘리브레이션(한글 0.529 · 기타 0.217 tok/char) 정확 재현.
+  - greedy_pack: 순서 보존·인접만·cap 준수·단독 초과 leaf=over_cap 전용 유닛·텍스트 손실 0
+    (구 deep_summary head/mid/tail 가운데 폐기 버그의 반대 성질).
+  - gate 3-way: 0=auto / (0,40]=hybrid / >40=whole (경계 포함).
+  - plan_summarize_units: trigger 이하=single(현행 단일콜 유지=무회귀) / 초과=map_reduce.
+
+pytest + 단독 실행 양쪽 지원:
+  PYTHONPATH=. .venv/bin/pytest tests/summarize_units/ -q
+"""
+from __future__ import annotations
+
+from app.services.hier_decomp.builder import HierNode
+from app.services.summarize_units import (
+    CAP_TOKENS,
+    TRIGGER_TOKENS,
+    SummarizeUnit,
+    estimate_tokens,
+    extract_leaves,
+    gate,
+    greedy_pack,
+    over_pct,
+    plan_summarize_units,
+)
+
+
+def _leaf(idx: int, text: str, title: str | None = None) -> HierNode:
+    return HierNode(idx=idx, parent_idx=None, level=1, node_type=None,
+                    section_title=title, heading_path=title, text=text)
+
+
+# ---------- estimate_tokens ----------
+
+def test_estimate_tokens_korean_calibration():
+    # 한글 1000자 → 529 tok (PR0: 0.529 tok/char)
+    assert estimate_tokens("가" * 1000) == 529
+
+
+def test_estimate_tokens_english_calibration():
+    # 비한글 1000자 → 217 tok (PR0: 0.217 tok/char)
+    assert estimate_tokens("a" * 1000) == 217
+
+
+def test_estimate_tokens_mixed_and_empty():
+    assert estimate_tokens("") == 0
+    mixed = "가" * 100 + "a" * 100
+    assert estimate_tokens(mixed) == round(100 * 0.529 + 100 * 0.217)
+
+
+# ---------- greedy_pack ----------
+
+def test_greedy_pack_adjacency_and_cap():
+    # 4000tok 짜리 한글 leaf 4개 (4000/0.529 ≈ 7562자) → cap 12000 이면 [3개, 1개]... 아니
+    # 4000*3=12000 = cap 정확 경계(<=cap 허용) → [1,2,3] + [4]
+    body = "가" * 7562  # ≈ 3999~4000 tok
+    leaves = [_leaf(i, body, f"s{i}") for i in range(4)]
+    units = greedy_pack(leaves, cap=12_000)
+    assert len(units) == 2
+    assert [len(u.section_titles) for u in units] == [3, 1]
+    # 순서 보존
+    assert units[0].section_titles == ["s0", "s1", "s2"]
+    assert units[1].section_titles == ["s3"]
+    # cap 준수
+    assert all(u.est_tokens <= 12_000 for u in units)
+
+
+def test_greedy_pack_oversized_leaf_gets_own_unit():
+    small = "가" * 1000            # ≈ 529 tok
+    big = "가" * 30_000            # ≈ 15,870 tok > CAP
+    leaves = [_leaf(0, small, "a"), _leaf(1, big, "mega"), _leaf(2, small, "b")]
+    units = greedy_pack(leaves, cap=CAP_TOKENS)
+    assert len(units) == 3
+    assert units[1].over_cap and units[1].section_titles == ["mega"]
+    assert not units[0].over_cap and not units[2].over_cap
+    # 인접성: 초과 leaf 가 앞뒤 pack 을 넘나들며 합쳐지지 않음
+    assert units[0].section_titles == ["a"] and units[2].section_titles == ["b"]
+
+
+def test_greedy_pack_no_text_loss():
+    leaves = [_leaf(i, f"본문{i} " + "가" * 500, f"s{i}") for i in range(7)]
+    units = greedy_pack(leaves, cap=1_000)
+    joined = "\n\n".join(u.text for u in units)
+    for leaf in leaves:
+        assert leaf.text in joined  # 커버리지 — 중간 폐기 0
+
+
+def test_greedy_pack_empty():
+    assert greedy_pack([]) == []
+
+
+# ---------- over_pct + gate ----------
+
+def test_over_pct_and_gate_boundaries():
+    assert gate(0.0) == "auto"
+    assert gate(0.01) == "hybrid"
+    assert gate(40.0) == "hybrid"
+    assert gate(40.01) == "whole"
+    assert gate(100.0) == "whole"
+
+
+def test_over_pct_computation():
+    # leaf: 6000tok + 18000tok(초과) → over% = 18000/24000 = 75%
+    l_small = _leaf(0, "가" * round(6000 / 0.529), "a")
+    l_big = _leaf(1, "가" * round(18000 / 0.529), "b")
+    pct = over_pct([l_small, l_big], cap=CAP_TOKENS)
+    assert 74.0 < pct < 76.0
+    assert over_pct([], cap=CAP_TOKENS) == 0.0
+    assert over_pct([l_small], cap=CAP_TOKENS) == 0.0
+
+
+# ---------- plan_summarize_units (fixture md) ----------
+
+def _md_doc(sections: int, chars_per_section: int, ch: str = "가") -> str:
+    parts = []
+    for i in range(sections):
+        parts.append(f"# 제{i+1}장 섹션{i}\n\n" + ch * chars_per_section)
+    return "\n\n".join(parts)
+
+
+def test_plan_small_doc_stays_single():
+    md = _md_doc(3, 1000)  # ≈ 3×529 tok ≪ trigger
+    plan = plan_summarize_units(md)
+    assert plan.mode == "single" and plan.tier is None and plan.units == []
+    assert plan.total_est_tokens <= TRIGGER_TOKENS
+
+
+def test_plan_large_doc_auto_tier():
+    # 섹션 20개 × ≈4000tok = ≈80K tok > trigger, 전 섹션 < cap → auto
+    md = _md_doc(20, 7562)
+    plan = plan_summarize_units(md)
+    assert plan.mode == "map_reduce"
+    assert plan.tier == "auto" and plan.over_pct == 0.0
+    assert len(plan.units) >= 2
+    assert all(u.est_tokens <= CAP_TOKENS for u in plan.units)
+
+
+def test_plan_mega_section_whole_tier():
+    # 작은 섹션 2 + 초대형 1(≈53K tok — 전체의 >40%) → whole
+    md = (_md_doc(2, 7562)
+          + "\n\n# 메가섹션\n\n" + "가" * 100_000)
+    plan = plan_summarize_units(md)
+    assert plan.mode == "map_reduce"
+    assert plan.tier == "whole" and plan.over_pct > 40.0
+    assert any(u.over_cap for u in plan.units)
+
+
+def test_plan_hybrid_tier():
+    # 정상 섹션 15개(≈60K tok) + 초과 섹션 1개(≈15.9K tok) → over% ≈ 21% → hybrid
+    md = _md_doc(15, 7562) + "\n\n# 초과섹션\n\n" + "가" * 30_000
+    plan = plan_summarize_units(md)
+    assert plan.mode == "map_reduce"
+    assert plan.tier == "hybrid"
+    assert 0.0 < plan.over_pct <= 40.0
+    over_units = [u for u in plan.units if u.over_cap]
+    assert len(over_units) == 1  # hybrid 시 클로드 대상 = 이 유닛들만
+
+
+def test_plan_headingless_giant_is_whole():
+    # 헤딩 없는 거대 EN 문서 — leaf 1개 전체 초과 → over% 100 → whole (PR0: EN 책 다수)
+    md = "x" * 200_000  # ≈ 43K tok > trigger, 단일 leaf > cap
+    plan = plan_summarize_units(md)
+    assert plan.mode == "map_reduce" and plan.tier == "whole"
+
+
+def test_plan_deterministic():
+    md = _md_doc(10, 7562)
+    p1, p2 = plan_summarize_units(md), plan_summarize_units(md)
+    assert p1 == p2
+
+
+if __name__ == "__main__":
+    import sys
+    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_")]
+    for fn in fns:
+        fn()
+        print(f"ok {fn.__name__}")
+    print(f"{len(fns)} passed (standalone)")
+    sys.exit(0)

tests/test_deep_summary_mapreduce.py

@@ -0,0 +1,266 @@
+"""presegment PR2 — deep_summary_worker map-reduce/HOLD 배선 단위테스트.
+
+worker-process 레벨(DB 필요)의 큐 상태 전이는 라이브 E2E 로 검증하고, 여기서는
+새 메커니즘의 seam 을 단위 검증한다 (test_fair_share.py 선례):
+  - _hold_awaiting_split: payload 마킹 commit 후 StageDeferred(HOLD_RETRY_MINUTES).
+  - _process_map_reduce: 유닛별 map → reduce → doc 필드 기록 / 모든 콜 캡 준수 /
+    payload.presegment.map_results 유닛 단위 persist(멱등 재개) / 실패 유닛 raise /
+    drain 보류(StageDeferred) 시 완료 유닛 보존.
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+from types import SimpleNamespace
+
+import pytest
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "app"))
+
+from ai.envelope import EscalationEnvelope  # noqa: E402
+from models.queue import StageDeferred  # noqa: E402
+from services.summarize_units import (  # noqa: E402
+    CAP_TOKENS,
+    estimate_tokens,
+    plan_summarize_units,
+)
+import workers.deep_summary_worker as dsw  # noqa: E402
+
+
+# ─── fixtures ────────────────────────────────────────────────────────────────
+
+# 30 절 × 한글 2,000자 ≈ 31.7K tok (> TRIGGER 25K) · 절당 ≈ 1,060 tok (< CAP) → auto
+GIANT_AUTO_MD = "\n".join(f"# 절 {i}\n" + ("가" * 2_000) for i in range(30))
+# 헤딩 1개 + 한글 60,000자 단일 섹션 ≈ 31.7K tok (> CAP) → over% 100 → whole
+GIANT_WHOLE_MD = "# 통짜\n" + ("가" * 60_000)
+
+MAP_JSON = (
+    '{"mode": "single", "tldr": "유닛 요약", "detail": "유닛 상세.",'
+    ' "inconsistencies": [{"kind": "version_drift", "desc": "개정판 차이"}],'
+    ' "confidence": 0.9}'
+)
+REDUCE_JSON = (
+    '{"mode": "single", "tldr": "전체 요약", "detail": "최종 상세.",'
+    ' "inconsistencies": [], "confidence": 0.8}'
+)
+
+
+class FakeSession:
+    """commit 시점의 queue_row.payload 를 **객체 참조**로 박제 — SQLAlchemy 의 committed
+    스냅샷과 동일하게, 이후 in-place 변경이 과거 커밋 객체에 소급 반영되는 aliasing
+    (60254 라이브에서 unit 0 만 persist 된 버그)을 검증 시점 직렬화로 탐지한다."""
+
+    def __init__(self, row=None):
+        self.commits = 0
+        self._row = row
+        self.snapshots: list = []
+
+    async def commit(self):
+        self.commits += 1
+        if self._row is not None:
+            self.snapshots.append(self._row.payload)  # 참조 박제 — 복사 금지(의도)
+
+
+class FakeClient:
+    """deep 슬롯 보유 클라이언트 — call_deep_or_defer 가 call_deep 을 타게 한다."""
+
+    def __init__(self, responses=None, fail_indexes=frozenset(), defer_from=None):
+        self.ai = SimpleNamespace(
+            deep=SimpleNamespace(model="qwen-macbook", context_char_limit=260_000)
+        )
+        self.prompts: list[str] = []
+        self._fail_indexes = fail_indexes  # 이 순번(0-based) 콜은 파싱 불가 응답
+        self._defer_from = defer_from  # 이 순번부터 연결 실패(StageDeferred 변환 대상)
+
+    async def call_deep(self, prompt: str, system=None) -> str:
+        import httpx
+
+        idx = len(self.prompts)
+        if self._defer_from is not None and idx >= self._defer_from:
+            raise httpx.ConnectError("macbook down")
+        self.prompts.append(prompt)
+        if idx in self._fail_indexes:
+            return "정상 JSON 아님"
+        if "유닛 요약 (총" in prompt:  # reduce 프롬프트 마커
+            return REDUCE_JSON
+        return MAP_JSON
+
+    async def close(self):
+        pass
+
+
+def _doc():
+    return SimpleNamespace(
+        id=999,
+        extracted_text=GIANT_AUTO_MD,
+        ai_detail_summary=None,
+        ai_inconsistencies=None,
+        ai_analysis_tier="triage",
+        ai_processed_at=None,
+    )
+
+
+def _envelope():
+    return EscalationEnvelope(
+        from_stage="classify",
+        escalation_reasons=("long_context",),
+        risk_flags=(),
+        distilled_context="4B 요지",
+        original_pointers={"doc_ids": [999]},
+    )
+
+
+@pytest.fixture
+def _patch_telemetry(monkeypatch):
+    events: list[dict] = []
+
+    async def fake_record(**kwargs):
+        events.append(kwargs)
+
+    monkeypatch.setattr(dsw, "record_analyze_event", fake_record)
+    return events
+
+
+# ─── _hold_awaiting_split ────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_hold_marks_payload_and_defers():
+    plan = plan_summarize_units(GIANT_WHOLE_MD)
+    assert plan.mode == "map_reduce" and plan.tier == "whole"
+
+    session, row = FakeSession(), SimpleNamespace(payload={"envelope": {"x": 1}})
+    with pytest.raises(StageDeferred) as ei:
+        await dsw._hold_awaiting_split(session, row, plan, document_id=999)
+
+    assert ei.value.retry_after_minutes == dsw.HOLD_RETRY_MINUTES
+    assert session.commits == 1  # 마킹이 defer 전에 commit — consumer 재읽기에서 보존
+    preseg = row.payload["presegment"]
+    assert preseg["awaiting_split"] is True
+    assert preseg["tier"] == "whole"
+    assert preseg["units"] == len(plan.units)
+    assert row.payload["envelope"] == {"x": 1}  # 기존 payload 병합 보존
+
+
+# ─── _process_map_reduce — 정상 경로 ────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_map_reduce_end_to_end(monkeypatch, _patch_telemetry):
+    plan = plan_summarize_units(GIANT_AUTO_MD)
+    assert plan.mode == "map_reduce" and plan.tier == "auto"
+    n = len(plan.units)
+    assert n >= 2  # greedy-pack 이 실제로 유닛을 나눴는지
+
+    client = FakeClient()
+    monkeypatch.setattr(dsw, "AIClient", lambda: client)
+    doc = _doc()
+    row = SimpleNamespace(payload={"envelope": {"x": 1}})
+    session = FakeSession(row)
+
+    await dsw._process_map_reduce(
+        doc, row, _envelope(), "generic", plan, session,
+        defer_on_deep_unavailable=False,
+    )
+
+    # 콜 수 = 유닛 map n + reduce 1
+    assert len(client.prompts) == n + 1
+    # 검증 게이트: 모든 콜 est_tokens <= CAP + 오버헤드(정책 템플릿+envelope ~3K)
+    for p in client.prompts:
+        assert estimate_tokens(p) <= CAP_TOKENS + 3_000
+    # doc 기록 = reduce 출력, 불일치 = map 유닛 합본 dedup
+    assert doc.ai_detail_summary == "최종 상세."
+    assert doc.ai_analysis_tier == "deep"
+    assert doc.ai_inconsistencies == [{"kind": "version_drift", "desc": "개정판 차이"}]
+    # 유닛 단위 persist — 유닛마다 commit
+    assert row.payload["presegment"]["units"] == n
+    assert len(row.payload["presegment"]["map_results"]) == n
+    assert session.commits == n
+    # ★aliasing 회귀 방지: 각 commit 이 박제한 payload 객체를 사후에 봤을 때
+    # map_results 가 1,2,...,n 로 단조 증가해야 한다. in-place 변경(구 버그)이면
+    # 모든 스냅샷이 같은 dict 를 공유해 [n,n,...,n] 으로 보인다 = SQLAlchemy 가
+    # committed 스냅샷과 new 가 같다고 판정해 UPDATE 를 스킵하는 것과 등가.
+    per_commit_units = [
+        len(s["presegment"]["map_results"]) for s in session.snapshots
+    ]
+    assert per_commit_units == list(range(1, n + 1))
+    # telemetry 1건 (reduce 기준)
+    events = _patch_telemetry
+    assert len(events) == 1 and events[0]["error_code"] is None
+
+
+# ─── 멱등 재개 ───────────────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_map_reduce_resume_skips_done_units(monkeypatch, _patch_telemetry):
+    plan = plan_summarize_units(GIANT_AUTO_MD)
+    n = len(plan.units)
+
+    client = FakeClient()
+    monkeypatch.setattr(dsw, "AIClient", lambda: client)
+    done_unit = {
+        "index": 0, "titles": ["절 0"], "tldr": "이전 요약", "detail": "이전 상세.",
+        "inconsistencies": [],
+    }
+    row = SimpleNamespace(payload={
+        "envelope": {"x": 1},
+        "presegment": {"map_results": {"0": done_unit}},
+    })
+    doc, session = _doc(), FakeSession()
+
+    await dsw._process_map_reduce(
+        doc, row, _envelope(), "generic", plan, session,
+        defer_on_deep_unavailable=False,
+    )
+
+    # 유닛 0 은 재호출 안 함 — map (n-1) + reduce 1
+    assert len(client.prompts) == n
+    assert row.payload["presegment"]["map_results"]["0"]["detail"] == "이전 상세."
+    assert doc.ai_detail_summary == "최종 상세."
+
+
+# ─── map 유닛 실패 → raise (성공분 persist) ─────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_map_unit_parse_failure_raises_but_persists_good_units(
+    monkeypatch, _patch_telemetry
+):
+    plan = plan_summarize_units(GIANT_AUTO_MD)
+    n = len(plan.units)
+
+    client = FakeClient(fail_indexes={1})  # 두 번째 map 콜만 파싱 불가
+    monkeypatch.setattr(dsw, "AIClient", lambda: client)
+    doc, session = _doc(), FakeSession()
+    row = SimpleNamespace(payload={"envelope": {"x": 1}})
+
+    with pytest.raises(ValueError, match="map 유닛"):
+        await dsw._process_map_reduce(
+            doc, row, _envelope(), "generic", plan, session,
+            defer_on_deep_unavailable=False,
+        )
+
+    # 성공 유닛(n-1)은 persist — 재시도 시 실패 1건만 재호출
+    assert len(row.payload["presegment"]["map_results"]) == n - 1
+    assert "1" not in row.payload["presegment"]["map_results"]
+    assert doc.ai_detail_summary is None  # doc 은 미기록
+    assert _patch_telemetry == []  # 가짜 완료 이벤트 없음
+
+
+# ─── drain 보류 — 완료 유닛 보존 + StageDeferred 전파 ───────────────────────
+
+@pytest.mark.asyncio
+async def test_map_defer_propagates_and_keeps_progress(monkeypatch, _patch_telemetry):
+    plan = plan_summarize_units(GIANT_AUTO_MD)
+
+    client = FakeClient(defer_from=1)  # 첫 유닛 성공 후 맥북 절단
+    monkeypatch.setattr(dsw, "AIClient", lambda: client)
+    doc, session = _doc(), FakeSession()
+    row = SimpleNamespace(payload={"envelope": {"x": 1}})
+
+    with pytest.raises(StageDeferred):
+        await dsw._process_map_reduce(
+            doc, row, _envelope(), "generic", plan, session,
+            defer_on_deep_unavailable=True,  # drain 시멘틱 — 보류 전파
+        )
+
+    assert len(row.payload["presegment"]["map_results"]) == 1
+    assert doc.ai_detail_summary is None

tests/test_queue_overview.py

@@ -4,6 +4,8 @@ services/queue_overview 의 SQL 수집부와 분리된 순수 판정 함수
 (stage_machine_map / build_machines / build_summarize_eta / build_trend /
 build_totals / compute_eta_minutes / rows_to_* / display_title) 를
 mock 행으로 검증한다. 통합(실 SQL)은 배포 후 라이브 smoke 로 확인.
+
+2026-07-02 컷오버 후 2노드(나스+맥미니) 기준 — 구 3노드 레인은 제거됨.
 """
 
 from datetime import datetime
@@ -18,7 +20,6 @@ from services.queue_overview import (
     compute_eta_minutes,
     display_title,
     rows_to_stage_stats,
-    rows_to_summarize_split,
     stage_machine_map,
 )
 
@@ -36,186 +37,115 @@ def _stage(**kw) -> dict:
     return base
 
 
-def _split(macbook: dict | None = None, macmini: dict | None = None) -> dict:
-    """summarize 풀 완료 실적 split — 미지정 0."""
-    zero = {"done_1h": 0, "done_today": 0, "done_15m": 0}
-    return {
-        "macbook": {**zero, **(macbook or {})},
-        "macmini": {**zero, **(macmini or {})},
-    }
-
-
 def _machine(machines: list[dict], key: str) -> dict:
     return next(m for m in machines if m["key"] == key)
 
 
 # ─── stage→machine 귀속 맵 ────────────────────────────────────────────────────
 
-def test_stage_machine_map_deep_enabled():
-    smap = stage_machine_map(deep_enabled=True)
+def test_stage_machine_map_two_nodes():
+    smap = stage_machine_map()
     for s in ("extract", "embed", "chunk", "markdown", "preview", "thumbnail", "fulltext", "stt"):
-        assert smap[s] == "gpu"
+        assert smap[s] == "nas"
     assert smap["classify"] == "macmini"
     assert smap["summarize"] == "macmini"
-    assert smap["deep_summary"] == "macbook"
-
-
-def test_stage_machine_map_deep_disabled():
-    """deep 슬롯 부재 시 deep_summary 도 macmini 귀속."""
-    smap = stage_machine_map(deep_enabled=False)
     assert smap["deep_summary"] == "macmini"
 
 
 # ─── 머신 카드 귀속 합산 ──────────────────────────────────────────────────────
 
-def test_gpu_stage_counts_attribution():
+def test_nas_stage_counts_attribution():
     stats = {
         "extract": _stage(pending=3, processing=1, done_1h=5, done_today=9, done_15m=1),
         "stt": _stage(failed=2, done_1h=1, done_today=2),
     }
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
-    gpu = _machine(machines, "gpu")
-    assert (gpu["pending"], gpu["processing"], gpu["failed"]) == (3, 1, 2)
-    assert (gpu["done_1h"], gpu["done_today"]) == (6, 11)
-    # gpu 의 stages 는 정적 8종 전부 (집계 0 이어도 표시)
-    assert gpu["stages"] == [
+    machines = build_machines(stats, [])
+    nas = _machine(machines, "nas")
+    assert (nas["pending"], nas["processing"], nas["failed"]) == (3, 1, 2)
+    assert (nas["done_1h"], nas["done_today"]) == (6, 11)
+    # nas 의 stages 는 정적 8종 전부 (집계 0 이어도 표시)
+    assert nas["stages"] == [
         "extract", "embed", "chunk", "markdown",
         "preview", "thumbnail", "fulltext", "stt",
     ]
 
 
-def test_summarize_pool_split_attribution():
-    """summarize pending/failed = macmini 귀속, 완료 실적은 split 로 분리 —
-    stage-level summarize done 수치는 카드에 이중 합산되지 않는다."""
+def test_macmini_llm_stages_attribution():
+    """classify/summarize/deep_summary 전부 macmini 귀속 (단일 생성 LLM 허브)."""
     stats = {
         "classify": _stage(done_1h=2, done_today=3),
         "summarize": _stage(pending=7, failed=1, done_1h=10, done_today=20),
+        "deep_summary": _stage(pending=2, processing=1, done_1h=3, done_today=4),
     }
-    split = _split(macbook={"done_1h": 4, "done_today": 8}, macmini={"done_1h": 6, "done_today": 12})
-    machines = build_machines(stats, split, [], deep_enabled=True)
+    machines = build_machines(stats, [])
     macmini = _machine(machines, "macmini")
-    macbook = _machine(machines, "macbook")
-
-    assert macmini["pending"] == 7 and macmini["failed"] == 1
-    assert macmini["done_1h"] == 2 + 6          # classify + macmini 몫 (10 아님)
-    assert macmini["done_today"] == 3 + 12
-    assert macbook["done_1h"] == 4 and macbook["done_today"] == 8
-    assert macbook["pending"] == 0              # 풀 pending 은 macmini 만
+    assert macmini["pending"] == 9 and macmini["failed"] == 1
+    assert macmini["processing"] == 1
+    assert macmini["done_1h"] == 2 + 10 + 3
+    assert macmini["done_today"] == 3 + 20 + 4
+    assert macmini["stages"] == ["classify", "summarize", "deep_summary"]
+    assert _machine(machines, "nas")["pending"] == 0
 
 
-def test_summarize_by_machine_projection():
-    """build_summarize_by_machine = split 의 done_1h/done_today 를 머신별로 투영
-    (done_15m 은 제외 — 내부 state 판정 전용)."""
-    from services.queue_overview import build_summarize_by_machine
-    split = _split(
-        macbook={"done_1h": 226, "done_today": 312, "done_15m": 60},
-        macmini={"done_1h": 37, "done_today": 94, "done_15m": 9},
-    )
-    sbm = build_summarize_by_machine(split)
-    assert sbm == {
-        "macmini": {"done_1h": 37, "done_today": 94},
-        "macbook": {"done_1h": 226, "done_today": 312},
-    }
-    assert "done_15m" not in sbm["macbook"]
-
-
-def test_compose_overview_includes_summarize_by_machine():
-    """compose_overview 응답 계약에 summarize_by_machine 포함 (FE 레인 분담 재료)."""
-    now_kst = datetime(2026, 6, 13, 13, 0, tzinfo=KST)
-    stats = {"summarize": _stage(pending=1317, done_1h=264)}
-    split = _split(macbook={"done_1h": 226, "done_today": 312}, macmini={"done_1h": 37, "done_today": 94})
-    ov = compose_overview(stats, split, {}, {}, [], deep_enabled=True, now_kst=now_kst)
-    assert ov["summarize_by_machine"]["macbook"]["done_1h"] == 226
-    assert ov["summarize_by_machine"]["macmini"]["done_today"] == 94
-
-
-def test_deep_disabled_deep_summary_counts_to_macmini():
-    stats = {"deep_summary": _stage(pending=2, processing=1, done_1h=3, done_today=4)}
-    machines = build_machines(stats, _split(), [], deep_enabled=False)
-    macmini = _machine(machines, "macmini")
-    macbook = _machine(machines, "macbook")
-    assert macmini["pending"] == 2 and macmini["processing"] == 1
-    assert macmini["done_1h"] == 3 and macmini["done_today"] == 4
-    assert macbook["stages"] == [] and macbook["pending"] == 0
-    assert _machine(machines, "macmini")["stages"] == ["classify", "summarize", "deep_summary"]
-
-
-def test_deferred_pending_always_on_macbook_card():
-    """보류(deferred_until 미래)는 summarize+deep_summary 합산으로 macbook 카드 귀속.
-    deep 슬롯 유무와 무관 (보류 = 맥북 불가 신호)."""
+def test_deferred_pending_on_macmini_card():
+    """보류(deferred_until 미래)는 summarize+deep_summary 합산으로 macmini 카드 귀속
+    (보류 = LLM 백오프 신호)."""
     stats = {
         "summarize": _stage(pending=5, deferred_pending=2),
         "deep_summary": _stage(pending=1, deferred_pending=1),
     }
-    for deep_enabled in (True, False):
-        machines = build_machines(stats, _split(), [], deep_enabled=deep_enabled)
-        assert _machine(machines, "macbook")["deferred_pending"] == 3
-        assert _machine(machines, "gpu")["deferred_pending"] == 0
-        assert _machine(machines, "macmini")["deferred_pending"] == 0
+    machines = build_machines(stats, [])
+    assert _machine(machines, "macmini")["deferred_pending"] == 3
+    assert _machine(machines, "nas")["deferred_pending"] == 0
 
 
 # ─── state 판정 ───────────────────────────────────────────────────────────────
 
-def test_macbook_state_active_wins_over_deferred_while_working():
+def test_macmini_state_active_wins_over_deferred_while_working():
     """가동 > 보류 (사용자 피드백 2026-06-11): 일하고 있으면 백오프 잔여가 있어도 '가동'.
 
     보류 건수는 deferred_pending 필드가 별도로 전달 — 카드 라인이 표시.
     """
-    stats = {"summarize": _stage(pending=1, deferred_pending=1)}
-    split = _split(macbook={"done_15m": 3})
-    machines = build_machines(stats, split, [], deep_enabled=True)
-    mb = _machine(machines, "macbook")
-    assert mb["state"] == "active"
-    assert mb["deferred_pending"] == 1
+    stats = {"summarize": _stage(pending=1, deferred_pending=1, done_15m=3)}
+    machines = build_machines(stats, [])
+    mm = _machine(machines, "macmini")
+    assert mm["state"] == "active"
+    assert mm["deferred_pending"] == 1
 
 
-def test_macbook_state_deferred_only_when_not_working():
+def test_macmini_state_deferred_only_when_not_working():
     """일이 멈춰 있고(처리 0·최근 완료 0) 백오프만 쌓인 상태에서만 '보류'."""
     stats = {"summarize": _stage(pending=1, deferred_pending=1)}
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
-    assert _machine(machines, "macbook")["state"] == "deferred"
+    machines = build_machines(stats, [])
+    assert _machine(machines, "macmini")["state"] == "deferred"
 
 
-def test_macbook_state_active_on_recent_qwen_done():
-    split = _split(macbook={"done_15m": 1})
-    machines = build_machines({}, split, [], deep_enabled=True)
-    assert _machine(machines, "macbook")["state"] == "active"
-
-
-def test_macbook_state_idle():
-    machines = build_machines({}, _split(), [], deep_enabled=True)
-    assert _machine(machines, "macbook")["state"] == "idle"
-
-
-def test_gpu_state_active_on_processing():
-    stats = {"extract": _stage(processing=1)}
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
-    assert _machine(machines, "gpu")["state"] == "active"
-
-
-def test_gpu_state_active_on_recent_done():
-    stats = {"embed": _stage(done_15m=2)}
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
-    assert _machine(machines, "gpu")["state"] == "active"
-
-
-def test_gpu_state_idle_when_old_done_only():
-    stats = {"embed": _stage(done_1h=5, done_today=9)}     # 15분 내 완료 없음
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
-    assert _machine(machines, "gpu")["state"] == "idle"
-
-
-def test_macmini_state_not_active_on_macbook_pool_done():
-    """summarize 풀 완료가 전부 macbook 몫이면 macmini 는 active 아님 (귀속 기준)."""
-    stats = {"summarize": _stage(done_15m=1)}
-    split = _split(macbook={"done_15m": 1})
-    machines = build_machines(stats, split, [], deep_enabled=True)
+def test_macmini_state_idle():
+    machines = build_machines({}, [])
     assert _machine(machines, "macmini")["state"] == "idle"
 
 
+def test_nas_state_active_on_processing():
+    stats = {"extract": _stage(processing=1)}
+    machines = build_machines(stats, [])
+    assert _machine(machines, "nas")["state"] == "active"
+
+
+def test_nas_state_active_on_recent_done():
+    stats = {"embed": _stage(done_15m=2)}
+    machines = build_machines(stats, [])
+    assert _machine(machines, "nas")["state"] == "active"
+
+
+def test_nas_state_idle_when_old_done_only():
+    stats = {"embed": _stage(done_1h=5, done_today=9)}     # 15분 내 완료 없음
+    machines = build_machines(stats, [])
+    assert _machine(machines, "nas")["state"] == "idle"
+
+
 def test_macmini_state_active_on_summarize_processing():
     stats = {"summarize": _stage(processing=1)}
-    machines = build_machines(stats, _split(), [], deep_enabled=True)
+    machines = build_machines(stats, [])
     assert _machine(machines, "macmini")["state"] == "active"
 
 
@@ -228,21 +158,18 @@ def test_current_summarize_to_macmini_max_two():
         {"stage": "summarize", "document_id": 3, "title": "문서C", "original_filename": None, "file_path": None},
         {"stage": "extract", "document_id": 4, "title": "문서D", "original_filename": None, "file_path": None},
     ]
-    machines = build_machines({}, _split(), rows, deep_enabled=True)
+    machines = build_machines({}, rows)
     macmini = _machine(machines, "macmini")
-    gpu = _machine(machines, "gpu")
+    nas = _machine(machines, "nas")
     assert [c["document_id"] for c in macmini["current"]] == [1, 2]    # 최대 2건
     assert macmini["current"][0] == {"document_id": 1, "title": "문서A", "stage": "summarize"}
-    assert [c["document_id"] for c in gpu["current"]] == [4]
-    assert _machine(machines, "macbook")["current"] == []
+    assert [c["document_id"] for c in nas["current"]] == [4]
 
 
-def test_current_deep_summary_follows_deep_slot():
+def test_current_deep_summary_to_macmini():
     rows = [{"stage": "deep_summary", "document_id": 9, "title": "심층", "original_filename": None, "file_path": None}]
-    enabled = build_machines({}, _split(), rows, deep_enabled=True)
-    disabled = build_machines({}, _split(), rows, deep_enabled=False)
-    assert _machine(enabled, "macbook")["current"][0]["document_id"] == 9
-    assert _machine(disabled, "macmini")["current"][0]["document_id"] == 9
+    machines = build_machines({}, rows)
+    assert _machine(machines, "macmini")["current"][0]["document_id"] == 9
 
 
 def test_display_title_fallback_chain():
@@ -344,32 +271,15 @@ def test_rows_to_stage_stats_conversion():
     assert stats["summarize"]["deferred_pending"] == 2
 
 
-def test_rows_to_summarize_split_conversion():
-    rows = [
-        (True, 4, 8, 1),       # is_macbook
-        (False, 6, 12, 0),
-    ]
-    split = rows_to_summarize_split(rows)
-    assert split["macbook"] == {"done_1h": 4, "done_today": 8, "done_15m": 1}
-    assert split["macmini"] == {"done_1h": 6, "done_today": 12, "done_15m": 0}
-
-
-def test_rows_to_summarize_split_empty():
-    split = rows_to_summarize_split([])
-    assert split["macbook"]["done_1h"] == 0 and split["macmini"]["done_today"] == 0
-
-
 def test_compose_overview_contract_shape():
     """응답 dict 의 키가 FE 계약 shape 과 정확히 일치하는지 고정."""
     out = compose_overview(
         {"summarize": _stage(pending=1)},
-        _split(),
         {}, {}, [],
-        deep_enabled=True,
         now_kst=datetime(2026, 6, 11, 14, 30, tzinfo=KST),
     )
     assert set(out.keys()) == {"machines", "stages", "summarize_eta", "trend_24h", "totals"}
-    assert [m["key"] for m in out["machines"]] == ["gpu", "macmini", "macbook"]
+    assert [m["key"] for m in out["machines"]] == ["nas", "macmini"]
     for m in out["machines"]:
         assert set(m.keys()) == {
             "key", "label", "state", "stages", "pending", "processing", "failed",
@@ -381,7 +291,7 @@ def test_compose_overview_contract_shape():
     assert set(out["trend_24h"][0].keys()) == {"hour", "inflow", "done"}
     assert set(out["totals"].keys()) == {"pending", "processing", "failed"}
     # 머신 label 고정 (raw 모델명 노출 금지 — label 만)
-    assert [m["label"] for m in out["machines"]] == ["GPU 서버", "맥미니", "맥북 M5 Max"]
+    assert [m["label"] for m in out["machines"]] == ["나스", "맥미니"]
 
 
 # ─── build_stages (단계별 현황 — 2026-06-11 사용자 피드백: 완료 가시화) ──────

tests/test_rerank_protocol.py

@@ -0,0 +1,54 @@
+"""rerank 프로토콜 정규화 단위 테스트 — 2노드 이관 P1-4 (llama.cpp /v1/rerank).
+
+순수 함수(ai/rerank_protocol.py)만 대상 — HTTP/DB 의존 없음.
+실행: PYTHONPATH=app pytest tests/test_rerank_protocol.py
+"""
+
+import json
+from pathlib import Path
+
+from ai.rerank_protocol import normalize_llamacpp_rerank
+
+FIXTURES = Path(__file__).parent / "fixtures"
+
+
+def test_normalize_llamacpp_shape_and_desc_sort():
+    payload = {
+        "model": "bge-reranker-v2-m3",
+        "results": [
+            {"index": 0, "relevance_score": 0.12},
+            {"index": 1, "relevance_score": 2.21},
+            {"index": 2, "relevance_score": -1.5},
+        ],
+    }
+    out = normalize_llamacpp_rerank(payload)
+    # TEI 계약: [{"index","score"}] score 내림차순
+    assert [r["index"] for r in out] == [1, 0, 2]
+    assert all(set(r) == {"index", "score"} for r in out)
+    assert out[0]["score"] == 2.21
+
+
+def test_normalize_llamacpp_missing_fields_skipped():
+    payload = {
+        "results": [
+            {"index": 0},  # relevance_score 없음 → 버림
+            {"relevance_score": 1.0},  # index 없음 → 버림
+            {"index": 3, "relevance_score": 0.5},
+        ]
+    }
+    assert normalize_llamacpp_rerank(payload) == [{"index": 3, "score": 0.5}]
+
+
+def test_normalize_llamacpp_empty_and_absent_results():
+    assert normalize_llamacpp_rerank({}) == []
+    assert normalize_llamacpp_rerank({"results": []}) == []
+
+
+def test_tei_fixture_shape_is_already_contract():
+    """TEI 캡처 fixture(Phase 2B G0-1 spec 박제)의 실응답이 정규화 없이 계약 형태임을 확인."""
+    doc = json.loads((FIXTURES / "tei_rerank_response.json").read_text())
+    captured = doc["captured_responses"]["baseline_bge_v2_m3"]["raw"]
+    assert isinstance(captured, list) and captured
+    assert {"index", "score"} <= set(captured[0])
+    # spec 문자열도 계약과 일치 (score desc 정렬 포함)
+    assert "index" in doc["response_shape"] and "score" in doc["response_shape"]