Initial commit: Korean stock value-investing AI pipeline

- 19개 마이크로서비스 (news-collector, score-engine, ta-engine, dart-collector, aux-signal, us-market, graph-engine, telegram-bot, dashboard-api, kis-api 등) - 가치투자 스코어링 + 10공식 앙상블 보팅 (매직포뮬러/F-Score/Altman/PEG/ 모멘텀/Beneish/GP-A/G-Score/Amihud/BAB) - 뉴스 수집→형태소→임베딩→중복제거→AI분석 파이프라인 - 기술적분석 + GAT 그래프신경망 + 미증시 동조 시그널 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-20 21:33:56 +09:00
commit 6d3b0bacc0
60 changed files with 17818 additions and 0 deletions
@@ -0,0 +1 @@
+{"id":"669c2fcf-b774-40a1-b874-21bbb5027474","name":"📈 AI 예측 성과 추적 (매일 16:10)","nodes":[{"id":"n1","name":"매일 16:10","type":"n8n-nodes-base.scheduleTrigger","typeVersion":1.1,"position":[100,300],"parameters":{"rule":{"interval":[{"field":"cronExpression","expression":"10 16 * * 1-5"}]}}},{"id":"n2","name":"추천종목 랭킹조회","type":"n8n-nodes-base.httpRequest","typeVersion":4.2,"position":[350,300],"parameters":{"method":"GET","url":"http://score-engine:8686/recommendations","options":{"timeout":30000}}},{"id":"n3","name":"오늘 성과 브리핑","type":"n8n-nodes-base.httpRequest","typeVersion":4.2,"position":[600,300],"parameters":{"method":"POST","url":"http://score-engine:8686/briefing/send","options":{"timeout":30000}}}],"connections":{"매일 16:10":{"main":[[{"node":"추천종목 랭킹조회","type":"main","index":0}]]},"추천종목 랭킹조회":{"main":[[{"node":"오늘 성과 브리핑","type":"main","index":0}]]}},"active":true}
@@ -0,0 +1,12 @@
+*.log
+logs/
+data/
+__pycache__/
+.git/
+node_modules/
+*.csv
+*.json
+*.db
+*.sqlite
+volumes/
+pg_backup/
@@ -0,0 +1,36 @@
+# 비밀/환경
+.env
+.env.*
+env-additions.txt
+*.secret
+secrets/
+
+# DB 덤프/백업
+pg_backup/
+*.dump
+*.sql.gz
+*.tar.gz
+
+# 마운트/모델/캐시
+mnt/
+models/
+**/__pycache__/
+*.pyc
+*.pyo
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+
+# IDE/OS
+.DS_Store
+.idea/
+.vscode/
+*.swp
+*~
+
+# Claude Code 로컬 설정
+.claude/
+
+# 로그
+*.log
+logs/
@@ -0,0 +1,577 @@
+# Trading AI — Claude Code 프로젝트 가이드
+
+> **역할**: 워렌 버핏 스타일 한국 주식 AI 투자 분석 전문가로 행동할 것.
+> 가치투자 관점(ROE·영업이익률·부채비율·FCF)을 최우선으로 판단한다.
+> 이 파일은 매 대화 시작 시 자동 로드됨 — 파일 탐색 없이 이 내용만으로 작업 시작.
+
+---
+
+## Claude 행동 원칙
+
+### 1. 먼저 생각, 그 다음 코딩
+- 가정은 명시적으로 밝힐 것. 불확실하면 질문.
+- 해석이 여러 개면 나열하고 고를 것 — 혼자 결정 금지.
+- 더 단순한 방법이 있으면 말할 것. 불필요한 복잡성에 반론.
+
+### 2. 단순함 우선
+- 요청한 것만 구현. 추측성 기능·추상화·유연성 추가 금지.
+- "나중에 필요할 수도" 코드 금지. 200줄이 50줄로 가능하면 다시 짤 것.
+- 불가능한 시나리오를 위한 에러 핸들링 금지.
+
+### 3. 정밀한 변경
+- 요청한 부분만 수정. 인접 코드 "개선" 금지.
+- 기존 스타일 유지 (내 방식이 달라도).
+- 내 변경으로 생긴 불필요한 import/변수/함수만 제거. 기존 dead code는 언급만.
+
+### 4. 검증 기준 명시
+- 다단계 작업 시 각 단계 완료 조건을 먼저 정의.
+- Docker 서비스 변경 → 로그 확인 필수.
+- DB 변경 → 쿼리로 확인 필수.
+
+### 이 프로젝트 특이사항
+- sudo 비밀번호 입력 불가 → sudo 필요 작업은 `! sudo <cmd>` 형태로 유저에게 요청.
+- n8n 마이그레이션 충돌 이력 있음 → n8n 스키마 변경 시 반드시 백업 먼저.
+- NAS fstab은 한 줄 유지 필수 (nfsvers=3).
+- 상장폐지 필터(`is_active=true`) 빠지면 추천 결과 오염됨.
+
+---
+
+## 시스템 개요
+
+뉴스 수집 → 형태소분석 → 임베딩 → 중복제거 → AI분석 → 재무스코어링 → 기술분석 → 매수/매도 추천 → 텔레그램 알림
+
+---
+
+## 인프라
+
+| 항목 | 값 |
+|------|-----|
+| 서버 OS | Ubuntu 22.04 |
+| GPU | GPU0: RTX 3060 12GB (EXAONE 추론) / GPU1: RTX 3070 8GB (bge-m3 임베딩) |
+| NAS | `192.168.0.36:/volume1/trading → /mnt/nas` (nfsvers=3 필수) |
+| DB 연결 | `postgres:5432`, DB명 `trading_ai`, user `kyu` |
+| 환경변수 | `/home/kyu/trading/.env` |
+| Docker | `docker compose` (파일: `/home/kyu/trading/docker-compose.yml`) |
+| PostgreSQL 데이터 | `/mnt/nas/postgresql/data` (NAS 저장) |
+
+---
+
+## 서비스 목록
+
+| 서비스 | 컨테이너 | 포트 | 내부IP | 역할 |
+|--------|---------|------|--------|------|
+| news-collector | trading-news-collector | 8787 | 172.30.0.16 | 뉴스수집+형태소+임베딩+Ollama분석 |
+| bareunaapi | trading-bareunaapi | 5757 | 172.30.0.12 | 한국어 형태소분석 FastAPI 래퍼 |
+| bareun | trading-bareun | 5656/9902 | 172.30.0.15 | 바른 NLP gRPC 서버 |
+| dart-collector | trading-dart-collector | 8888 | 172.30.0.17 | DART 공시/재무제표 수집 |
+| score-engine | trading-score-engine | 8686 | 172.30.0.19 | 종합 투자 스코어 계산+텔레그램 |
+| ta-engine | trading-ta-engine | 8484 | 172.30.0.23 | 기술적분석 (MA/RSI/MACD/볼밴/스토캐스틱) |
+| dashboard-api | trading-dashboard-api | 8989 | 172.30.0.22 | 대시보드 REST API |
+| kis-api | trading-kis-api | 8585 | 172.30.0.18 | 네이버모바일 주가수집 + 매매시그널 |
+| ollama | trading-ollama | 11434 | 172.30.0.13 | EXAONE 3.5 7.8B 추론(GPU0) + bge-m3 임베딩(GPU1) |
+| n8n | trading-n8n | 5678 | 172.30.0.20 | 워크플로우 자동화 |
+| n8n-worker | trading-n8n-worker | - | 172.30.0.21 | n8n Queue Worker |
+| postgres | trading-postgres | 5432 | - | 메인 DB |
+| redis | trading-redis | 6379 | 172.30.0.10 | 캐시/중복제거 |
+| qdrant | trading-qdrant | 6333 | 172.30.0.11 | 뉴스 벡터 유사도 필터 |
+| aux-signal | trading-aux-signal | 8282 | 172.30.0.25 | 보조 데이터: 네이버 컨센서스·기관/외국인 수급 + ECOS 매크로(USD/KRW, 국고채10년) |
+| us-market | trading-us-market | 8383 | 172.30.0.24 | 미증시→한국 동조 시그널 (섹터ETF 14개 ±5점, 개별 페어 회귀 ±10점, 자동 페어발굴) |
+| graph-engine | trading-graph-engine | 9090 | 172.30.0.27 | GAT 그래프 신경망 (12피처 노드 × 가격상관/섹터/뉴스공기 엣지) → graph_score |
+| telegram-bot | trading-telegram-bot | - | 172.30.0.26 | 텔레그램 명령 처리 (/buy /sell /stock /deep /market) |
+
+---
+
+## 파일 구조 & 핵심 함수
+
+### `/home/kyu/trading/news-collector/main.py` (21KB)
+뉴스 수집 + AI 분석 파이프라인. 포트 8787.
+
+| 함수/클래스 | 역할 |
+|------------|------|
+| `class S` | 전역 상태 (DB pool, Redis, Qdrant client) |
+| `nhash(title, url)` | SHA256[:16] 뉴스 중복 해시 |
+| `is_korean(text)` | 한글 포함 여부 체크 |
+| `parse_rss_date(date_str)` | RSS pubDate → ISO timestamp |
+| `@app.post("/collect/market")` | 네이버 금융 메인 뉴스 수집 |
+| `@app.post("/collect/stocks")` | 종목별 네이버 뉴스 수집 |
+| `@app.post("/collect/rss")` | 18개 RSS 피드 수집 |
+| `@app.get("/sources")` | RSS 소스 목록 |
+
+**파이프라인 순서** (함수 내부):
+1. 바른API `/analyze` → 형태소분석 + 종목감지
+2. Ollama `bge-m3` → 1024차원 임베딩
+3. Qdrant → 코사인 유사도 ≥0.92 중복 제거
+4. Ollama `exaone3.5:7.8b` (GPU0) → 호재/악재/중립 + intensity(1~5) + catalyst분류
+5. PostgreSQL `news_analysis` 저장
+
+**EXAONE 프롬프트**: 버핏 관점 (실적/수주/배당/리스크/모멘텀/기타 catalyst 분류)
+
+**RSS 소스 (28개)**: 한국경제, 매일경제, 머니투데이, 이데일리, 연합뉴스, 조선비즈, 헤럴드경제, 아시아경제, 파이낸셜뉴스, SBS Biz, 뉴스1, 뉴시스 등 + 네이버금융 직접 크롤링
+
+**스케줄러**:
+- 평일 RSS: 8-18시 5분마다 (`rss_weekday`)
+- 주말 RSS: 8-22시 15분마다 (`rss_weekend`) — 누적 학습용
+- 평일 마켓: 9-17시 10분마다 (`market`, 네이버 금융 메인)
+- raw 분석: 24시간 30분마다 200건 batch (`process_raw`) — 백로그 소화용으로 확장
+- 주간 raw 백필: 일요일 02:00 전체 종목 × 50페이지 (`historical_raw_weekly`)
+
+**수집/분석 분리** (대규모 백필용):
+- `POST /collect/historical-raw?count=0&max_pages=100` — raw만 빠르게 수집 (LLM 스킵)
+- `POST /process/raw?batch_size=200` — news_raw → news_analysis 점진 분석
+- `GET /raw/stats` — 백필 진행률 (total/unprocessed/processed)
+
+---
+
+### `/home/kyu/trading/bareunaapi/main.py` (9.5KB)
+한국어 형태소분석 FastAPI 래퍼. 포트 5757.
+
+| 함수/클래스 | 역할 |
+|------------|------|
+| `class AppState` | bareun gRPC stub + Redis + 종목사전 |
+| `class AnalyzeRequest / AnalyzeResponse` | 분석 요청/응답 모델 |
+| `news_hash(title, url)` | Redis 중복제거용 해시 |
+| `extract_morphemes(text)` | 바른API → 명사 추출 |
+| `extract_stocks(text)` | 텍스트에서 종목코드 감지 |
+| `build_filtered(nouns, stocks)` | 필터링된 텍스트 구성 |
+| `scan_finance_terms(text)` | finance_dict.py 금융용어 매칭 |
+| `_analyze(req)` | 핵심 분석 로직 |
+| `@app.post("/analyze")` | 단건 분석 |
+| `@app.post("/analyze/batch")` | 배치 분석 |
+| `@app.get("/stocks")` | 로드된 종목 목록 |
+| `@app.post("/stocks/refresh")` | KRX→Naver 종목 새로고침 |
+| `@app.delete("/dedup/flush")` | Redis 중복제거 캐시 초기화 |
+
+**관련 파일**:
+- `bareunaapi/finance_dict.py` — 주식/금융 전문 용어 사전
+- `bareunaapi/stock_loader.py` — KRX→Naver 폴백, ~3768 종목 로드
+
+---
+
+### `/home/kyu/trading/dart-collector/main.py` (34KB)
+DART 공시/재무제표 수집. 포트 8888.
+
+| 함수/클래스 | 역할 |
+|------------|------|
+| `class Stats` | 수집 통계 |
+| `get_corp_code(stock_code)` | 종목코드 → DART corp_code 변환 |
+| `get_corp_name(stock_code)` | 종목코드 → 기업명 |
+| `calc_financial_ratios(key_items, prev_revenue)` | ROE/영업이익률/부채비율/FCF/매출성장률 계산 |
+| `@app.post("/collect/disclosures")` | 공시 수집 |
+| `@app.post("/collect/financials")` | 재무제표 수집 (300개 종목) |
+| `@app.post("/collect/major")` | 주요 대형주 수집 |
+| `@app.post("/collect/corps")` | DART 기업목록 갱신 |
+| `@app.get("/corps")` | 기업목록 조회 |
+| `@app.get("/corps/{stock_code}")` | 특정 기업 조회 |
+| `@app.get("/financials/{stock_code}")` | 종목 재무데이터 |
+| `@app.get("/stats")` | 수집 통계 |
+
+---
+
+### `/home/kyu/trading/score-engine/main.py` (27KB)
+종합 투자 스코어 계산 + 텔레그램 알림. 포트 8686.
+
+| 함수/클래스 | 역할 |
+|------------|------|
+| `get_recommendation(score)` | 점수→등급 변환 |
+| `calc_fundamental_score(fin, per, pbr)` | 재무점수 산출 (-100~100) |
+| `calc_magic_formula(fin, market_cap)` | 그린블라트 매직 포뮬러 (ROC + EY) 0~30점 |
+| `calc_piotroski_score(curr, prev)` | 피오트로스키 F-Score 7신호 → -15~+15 |
+| `calc_altman_z(fin, market_cap)` | 알트만 Z-Score 단순화 → 매수/매도/관망 신호 |
+| `calc_peg(curr, prev, per)` | 린치 GARP — PER/이익성장률 → 신호 |
+| `calc_momentum(conn, code)` | AQR 12-1개월 가격 모멘텀 → 신호 |
+| `calc_beneish_simplified(curr, prev)` | Beneish M-Score 단순화 (분식 의심도) |
+| `aggregate_signals(signals)` | 6공식 보팅 다수결 → (요약, 카운트) |
+| `calc_trend_score(conn, code)` | 5년 ROE 일관성·추세 -30~+30 |
+| `calc_dcf(fin, market_cap)` | 간이 DCF 내재가치 + 안전마진 |
+| `calc_earnings_quality(fin)` | CFO/영업이익 비율 (분식의심 패널티) |
+| `is_value_investable(fin, per, pbr, market_cap)` | 버핏 필터 통과 여부 |
+| `@app.post("/score/calculate")` | 종목 점수 계산 실행 |
+| `@app.post("/briefing/send")` | 텔레그램 브리핑 전송 |
+| `@app.get("/ranking")` | 전체 종목 랭킹 |
+| `@app.get("/recommendations")` | 추천 종목 목록 |
+| `@app.get("/stock/{code}")` | 특정 종목 점수 상세 |
+| `@app.get("/backtest")` | 추천 성과 백테스트 (수익률/승률/샤프/MDD/알파) |
+| `@app.post("/learn-weights")` | 공식별 신호 → 7d 수익률 회귀로 가중치 학습 |
+| `@app.get("/learn-weights")` | 현재 적용 중인 공식 가중치 조회 |
+| `@app.post("/learn-pricing")` | D+E: 점수→30d 수익률 선형회귀 + Random Forest 학습 |
+| `@app.get("/predict-price/{code}")` | 학습된 모델로 종목 30일 후 예상 수익률·가격 |
+| `@app.get("/sector/concentration")` | 섹터 집중도 + 30% 초과 경고 |
+
+**스코어링 공식 (실제 가중치)**:
+```
+종합점수 = 펀더멘털통합×0.24 + 뉴스(catalyst가중)×0.18 + 기술×0.15
+        + DART공시×0.10 + 외국인수급×0.14 + 공매도×0.06
+        + 가격모멘텀×0.03 + DCF안전마진×0.10
+        + 시장레짐 보정(±5~10)
+        + 앙상블 보팅(±18) ← 6공식 매수/매도 합 × 학습가중치
+```
+
+**10공식 앙상블 보팅** — 각 공식이 독립 신호 발신, 학습 가중치로 결합:
+| 공식 | 출처/논문 | 데이터 | 매수 신호 |
+|------|----------|--------|----------|
+| 매직포뮬러 | Greenblatt | ROC + EY | magic_score ≥ 20 |
+| F-Score | Piotroski 2000 | 7신호 (전년 대비) | f_score ≥ 6 |
+| 알트만 Z | Altman 1968 | 6.72×ROA + 1.05×(시총/부채) | Z ≥ 2.6 |
+| PEG | Lynch GARP | PER / 이익성장률 | PEG ≤ 1.5 |
+| 12-1 모멘텀 | AQR (Carhart 1997) | (P_-21 / P_-252) - 1 | ≥ 10% |
+| Beneish | Beneish 1999 | TATA·SGI·CFO/NI | 의심도 < 50 + CFO/NI > 1 |
+| **GP/A** | Novy-Marx 2013 | 영업이익/총자산 (대체) | ≥ 15% |
+| **G-Score** | Mohanram 2005 | 5신호 vs 섹터 중앙값 | ≥ 4 |
+| **Amihud** | Amihud 2002 | avg(\|return\|/거래대금) | ≥ 100 (소형 알파) |
+| **베타(BAB)** | Frazzini-Pedersen 2014 | 종목 vs KOSPI 60일 회귀 | β < 0.7 (저베타 알파) |
+
+**학습 가중치** (`weight_config` 테이블):
+- `POST /learn-weights?days=90` — 백테스트로 공식별 매수vs매도 그룹 7d 수익률 차이(edge) 측정
+- edge가 큰 공식일수록 가중치 ↑ (정규화 후 합 = 6, 균등 시 각 1.0)
+- 다음 점수 계산부터 자동 적용. 표본 부족 시 default(균등 1.0)
+
+**펀더멘털통합** = `clip(buffett_score + trend_score + earnings_quality + magic_score + f_score_adj, -100, 100)`
+- `buffett_score`: ROE / 영업이익률 / 부채비율 / 매출성장 / PER / PBR / FCF / 배당
+- `trend_score`: 5년 ROE 일관성·추세 (-30~+30)
+- `earnings_quality`: CFO/영업이익 ≥1 가산, <0.7 분식의심 패널티
+- `magic_score`: ROC(영업이익/총자산) + EY(영업이익/EV) 임계값 합산 (0~30)
+- `f_score_adj`: F-Score ≥6 +15, 5 +8, 4 +3, ≤2 -15 (가치함정 회피)
+
+**매직 포뮬러 임계값** (한국 시장 보정):
+- ROC: ≥25% +15 / ≥15% +10 / ≥8% +5
+- EY:  ≥15% +15 / ≥10% +10 / ≥6% +5
+
+**피오트로스키 F-Score 7신호** (현재/전년 사업보고서 비교, 9중 2개는 데이터 부재로 생략):
+1. ROA(NI/총자산) > 0  2. CFO > 0  3. ΔROA > 0  4. CFO > NI
+5. Δ부채비율 < 0  6. Δ영업이익률 > 0  7. Δ자산회전율 > 0
+
+**추천 등급** (점수 + 다수공식 동의 강제):
+- 강력매수: 점수 ≥70 AND 매수보팅 ≥3
+- 매수관심: 점수 ≥40 AND 매수보팅 ≥1 AND 매도보팅 <2
+- 매도관심: 점수 ≤-30 OR 매도보팅 ≥3
+- 강력매도: 점수 ≤-60 OR 매도보팅 ≥4
+- 관망: 그 외
+
+**버핏 가치투자 필터** (`is_value_investable`):
+- `operating_profit > 0` 영업적자 제외
+- ROE ≥ 10%
+- 부채비율 ≤ 200%
+- PER ≤ 60
+- 시총 ≥ 100억
+- `dart_corps.is_active=true` (상장폐지 제외)
+
+---
+
+### `/home/kyu/trading/ta-engine/main.py` (37KB)
+기술적 분석 엔진. 포트 8484.
+
+| 함수/클래스 | 역할 |
+|------------|------|
+| `_ema_series(values, period)` | EMA 계산 |
+| `_ma(closes, n)` | 단순이동평균 |
+| `_rsi(closes, period=14)` | RSI |
+| `_macd(closes)` | MACD (12/26/9) |
+| `_bollinger(closes, period=20)` | 볼린저밴드 |
+| `_stochastic(highs, lows, closes, period=14)` | 스토캐스틱 K/D |
+| `_vol_ratio(volumes, period=20)` | 거래량비율 |
+| `calc_indicators(ohlcv)` | 전체 지표 계산 |
+| `calc_tech_score(ind)` | 기술점수 (-100~100) |
+| `calc_price_targets(price, ind, sig)` | 목표가/손절가 계산 |
+| `analyze_position(price, buy_price, qty, ...)` | 포지션 분석 |
+| `@app.get("/technical/{code}")` | 종목 기술분석 |
+| `@app.get("/ranking")` | 기술점수 랭킹 |
+| `@app.get("/buy-candidates")` | 매수 후보 |
+| `@app.post("/analyze/all")` | 전체 종목 분석 |
+| `@app.post("/analyze/{code}")` | 특정 종목 분석 |
+| `@app.post("/position")` | 포지션 분석 |
+| `@app.get("/report/{code}")` | 종합 리포트 |
+
+**목표가 계산**:
+- 진입가: `min(현재가, MA20×0.99)`
+- T1: `max(볼밴상단, 현재가×1.05)`
+- T2: `max((현재가+52주고가)/2, 현재가×1.10)`
+- T3: `max(52주고가×0.97, 현재가×1.20)`
+- 손절가: `min(MA60×0.98, 현재가×0.95)`
+
+---
+
+### `/home/kyu/trading/dashboard-api/main.py` (19KB)
+대시보드 REST API. 포트 8989.
+
+| 엔드포인트 | 역할 |
+|-----------|------|
+| `GET /api/summary` | 전체 요약 통계 |
+| `GET /api/recent` | 최근 뉴스 분석 |
+| `GET /api/ranking` | 종합 랭킹 |
+| `GET /api/recommendations` | 추천 종목 |
+| `GET /api/avoid` | 회피 종목 |
+| `GET /api/signals` | 매매 시그널 |
+| `GET /api/technical/{code}` | 기술분석 데이터 |
+| `GET /api/buy-candidates` | 매수 후보 |
+| `GET /api/alerts` | 알림 목록 |
+| `GET /api/timeline` | 타임라인 |
+| `GET /api/stock/{code}` | 종목 상세 |
+| `GET /api/search` | 종목 검색 |
+| `POST /api/position` | 포지션 등록 |
+| `GET /api/report/{code}` | 종목 리포트 |
+| `GET /api/fundamentals` | 전체 재무데이터 |
+| `GET /api/fundamentals/{code}` | 종목 재무데이터 |
+| `GET /api/name/{code}` | 종목명 조회 |
+| `GET /` | 대시보드 HTML |
+
+---
+
+### `/home/kyu/trading/kis-api/main.py` (11KB)
+네이버 모바일 주가수집 + 매매시그널. 포트 8585.
+
+| 함수/엔드포인트 | 역할 |
+|---------------|------|
+| `class Stats` | 수집 통계 |
+| `calc_signal(p, news_sc, dart_sc)` | 뉴스+공시점수 → 매매시그널 |
+| `GET /price/{code}` | 단일 종목 주가 |
+| `GET /prices` | 전체 주가 목록 |
+| `GET /signals` | 전체 시그널 |
+| `GET /signals/{code}` | 종목별 시그널 |
+| `POST /collect` | 주가 수집 실행 |
+
+**Redis DB 할당**:
+- db=0: n8n Queue
+- db=1: bareunaapi 중복제거
+- db=3: kis-api/score-engine (`price:{code}`, `prices:last_update`)
+- db=4: news-collector (`news:naver:{hash}`)
+- db=5: ta-engine (`ta:{code}`)
+
+---
+
+### `/home/kyu/trading/aux-signal/main.py`
+보조 데이터 수집. 포트 8282.
+
+| 엔드포인트 | 역할 |
+|-----------|------|
+| `POST /collect/naver` | 네이버 integration API → 컨센서스 + 기관/외국인 일별 수급 |
+| `POST /collect/macro` | 한국은행 ECOS → USD/KRW 환율, 국고채 10년 |
+| `GET /consensus/{code}` | 종목 컨센서스 (목표가/투자의견) |
+| `GET /flow/{code}` | 기관/외국인 일별 순매수 |
+| `GET /macro/latest` | 최신 매크로 지표 |
+| `GET /macro/{indicator}` | 지표별 시계열 |
+
+스케줄: 영업일 18:00 네이버 수급/컨센서스, 08:30 ECOS 매크로.
+
+---
+
+### `/home/kyu/trading/us-market/main.py`
+미국증시 동조 시그널. 포트 8383.
+
+| 엔드포인트 | 역할 |
+|-----------|------|
+| `POST /collect` | Finnhub/AlphaVantage → 미국 ETF·개별주 가격 |
+| `POST /collect/backfill` | 신규 페어용 과거 데이터 백필 |
+| `POST /collect/yfinance-backfill` | yfinance 폴백 백필 |
+| `POST /signal/calculate` | 페어 회귀로 한국 종목별 미증시 시그널 산출 |
+| `GET /signal/{kr_code}` | 한국 종목 KR 코드별 시그널 |
+| `GET /signal/latest` | 전체 종목 최신 시그널 |
+| `GET /pairs` | 등록된 KR↔US 페어 |
+| `POST /pairs/recalc-beta` | 60일 회귀 베타 재계산 |
+| `POST /pairs/discover` | KOSPI200 × S&P500 상관 자동 발굴 |
+| `GET /etfs` / `GET /etfs/{etf}/latest` | 섹터 ETF 현황 |
+
+**기여 점수**:
+- 섹터 ETF 동조 (SOXX/XBI/LIT 등 14개) → 같은 한국 섹터 ±5점
+- 개별 페어 60일 회귀 베타 (NVDA↔SK하이닉스 등) → ±10점
+
+스케줄: 매일 KST 07:30 수집, 08:00 시그널 계산.
+
+---
+
+### `/home/kyu/trading/graph-engine/main.py`
+GAT 그래프 신경망 (PyTorch). 포트 9090.
+
+| 엔드포인트 | 역할 |
+|-----------|------|
+| `POST /graph/build` | 노드/엣지 재구성 |
+| `POST /train` | 학습 (6mo rolling) |
+| `POST /predict` | 전체 추론 → `stock_scores.graph_score` |
+| `GET /predict/{code}` | 종목별 예측 값 |
+| `GET /status` | 모델/학습 상태 |
+
+- **노드**: 활성 종목, 12피처 (1d/5d/20d 수익률, vol_ratio, RSI, tech_score, ROE, 영업이익률, 부채비율, 7d 뉴스, 미증시 overnight, log_mcap)
+- **엣지**: ① 60일 가격 상관 |corr|≥0.4  ② 동일 섹터  ③ 뉴스 공기 ≥3회
+- **스케줄**: 일요 06:00 학습 / 매일 08:30 추론
+- **모델 저장**: `/mnt/nas/models/graph/`
+
+---
+
+### `/home/kyu/trading/telegram-bot/main.py`
+텔레그램 명령 처리 봇 (python-telegram-bot).
+
+| 명령 | 역할 |
+|------|------|
+| `/start /help` | 도움말 |
+| `/buy /buys` | 강력매수/매수관심 톱 N |
+| `/sell /sells` | 매도관심/강력매도 톱 N |
+| `/stock <코드>` | 종목 점수·신호 요약 |
+| `/deep <코드>` | RAG+EXAONE 심층분석 결과 |
+| `/market` | 시장 요약(섹터·매크로) |
+
+권한: `TELEGRAM_CHAT_ID` 일치 채팅만 응답.
+
+---
+
+## 데이터베이스 스키마 (PostgreSQL)
+
+### `news_analysis`
+```
+id, title, url, source, published_at, hash(16자 UNIQUE)
+sentiment(호재/악재/중립), intensity(1~5), primary_stock
+affected_stocks(JSONB), reason, investment_action(매수관심/매도관심/관망)
+keywords(JSONB), stock_names(JSONB), stock_codes(JSONB)
+similar_count, catalyst, analyzed_at, created_at
+```
+
+### `stock_technical`
+```
+stock_code(UNIQUE), stock_name, price
+ma5, ma20, ma60, ma120, rsi, macd, macd_signal, macd_hist
+bb_upper, bb_mid, bb_lower, pct_b, stoch_k, stoch_d
+vol_ratio, tech_score(-100~100), signal(매수/매도/관망)
+obv, obv_trend, vwap20, ichimoku(JSONB)  ← 새 지표
+signals(JSONB), targets(JSONB), analyzed_at
+```
+
+### `dart_corps`
+```
+stock_code(PK), corp_code, corp_name, modify_date, is_active
+※ is_active=true 필터 필수 (상장폐지 제외)
+```
+
+### `dart_financials`
+```
+stock_code, corp_code, corp_name, bsns_year, reprt_code
+revenue, operating_profit, net_income
+total_assets, total_liabilities, total_equity, operating_cashflow
+roe, operating_margin, net_margin, debt_ratio, revenue_growth, fcf_ratio
+UNIQUE(stock_code, bsns_year, reprt_code)
+※ F-Score는 전년 11011 사업보고서 필요 → years≥2 백필
+```
+
+### `stock_scores` (주요 컬럼)
+```
+stock_code, score_date, total_score, recommendation
+news_score, dart_score, technical_score, foreign_score, short_score, price_score
+trend_score, intrinsic_value, margin_of_safety, earnings_quality
+magic_score, f_score, roc_pct, earnings_yield_pct
+altman_z, peg, momentum_pct, beneish_score
+gpa_pct, g_score, amihud_illiq, market_beta  ← 학술 논문 기반 4개
+signals(JSONB), buy_votes, sell_votes
+position_size_pct, volatility_60d, market_regime_adj, sector
+top_reasons, UNIQUE(stock_code, score_date)
+```
+
+### `weight_config`
+```
+config_date(PK), weights(JSONB), period_days, sample_size
+※ 최신 row가 calculate_daily_scores에서 자동 로드 → ensemble_bonus 가중
+```
+
+### `pricing_model` (D+E 학습 결과 저장)
+```
+model_date(PK), linear_coef, linear_intercept, linear_r2,
+rf_features(JSONB), rf_r2, sample_size, period_days, created_at
+※ /learn-pricing 호출 시 갱신, /predict-price 호출 시 최신 row 사용
+```
+
+### `news_raw` (수집/분석 분리용 임시 저장소)
+```
+id, stock_code, title, url, url_hash(UNIQUE), source,
+published_at_text, collected_at, processed(BOOL), processed_at
+※ 크롤러는 raw만 빠르게 저장 → 야간 cron(process_raw)이 batch로 LLM 분석
+```
+
+### 뷰
+- `v_recent_signals` — 최근 24시간 호재/악재 (intensity DESC)
+- `v_stock_news_count` — 종목별 7일 뉴스 감성 카운트
+
+---
+
+## 환경변수 키 목록 (.env)
+
+```
+POSTGRES_HOST=postgres  POSTGRES_PORT=5432  POSTGRES_DB=trading_ai  POSTGRES_USER=kyu
+REDIS_MAX_MEMORY=2gb  REDIS_MAXMEMORY_POLICY=allkeys-lru
+OLLAMA_NUM_PARALLEL=4  (EXAONE→GPU0, bge-m3→GPU1 자동 분배)
+QDRANT_COLLECTION=news_vectors  QDRANT_VECTOR_SIZE=1024
+TELEGRAM_CHAT_ID=8690666445
+KIS_IS_PAPER=true  (모의투자 모드)
+ECOS_API_KEY=...   (한국은행 매크로 — ecos.bok.or.kr/api 무료)
+```
+
+---
+
+## 자주 쓰는 명령어
+
+```bash
+# 전체 상태
+docker compose ps
+
+# 특정 서비스 로그 (실시간)
+docker logs trading-news-collector --tail 50 -f
+docker logs trading-score-engine --tail 50 -f
+docker logs trading-n8n --tail 30
+
+# 재빌드+재배포
+docker compose build <service> && docker compose up -d <service>
+
+# PostgreSQL 쿼리 (DB명: trading_ai, user: kyu)
+docker exec trading-postgres psql -U kyu -d trading_ai -c "SELECT ..."
+
+# Redis
+source /home/kyu/trading/.env && docker exec trading-redis redis-cli -a $REDIS_PASSWORD
+
+# n8n 마이그레이션 확인
+docker exec trading-postgres psql -U kyu -d trading_ai -c "SELECT COUNT(*) FROM n8n.migrations;"
+
+# 추천 종목 조회
+docker exec trading-postgres psql -U kyu -d trading_ai -c "
+  SELECT s.stock_code, d.corp_name, s.total_score, s.recommendation,
+         s.magic_score, s.f_score, s.roc_pct, s.earnings_yield_pct
+  FROM stock_scores s JOIN dart_corps d ON d.stock_code=s.stock_code
+  WHERE d.is_active=true AND s.score_date=CURRENT_DATE
+  ORDER BY s.total_score DESC LIMIT 10;"
+
+# 다년치 사업보고서 백필 (F-Score / PEG용)
+curl -X POST 'http://localhost:8888/collect/financials?count=3000&years=10&annual_only=true'
+
+# 섹터 정보 채우기 (dart_corps.sector)
+curl -X POST http://localhost:8888/collect/sectors
+
+# 점수 수동 재계산 (전체 활성종목 대상)
+curl -X POST http://localhost:8686/score/calculate
+
+# 백테스트 기반 공식 가중치 학습
+curl -X POST 'http://localhost:8686/learn-weights?days=90'
+
+# 현재 가중치 조회
+curl -s http://localhost:8686/learn-weights
+
+# 종목별 6공식 신호 확인
+docker exec trading-postgres psql -U kyu -d trading_ai -c "
+  SELECT stock_code, stock_name, total_score, signals
+  FROM stock_scores WHERE score_date=CURRENT_DATE
+  ORDER BY total_score DESC LIMIT 10;"
+```
+
+---
+
+## 주의사항
+
+- **상장폐지 필터**: 모든 추천 쿼리에 `JOIN dart_corps d ON d.stock_code=... WHERE d.is_active=true` 필수
+- **fstab**: NAS 마운트는 반드시 한 줄로 작성, `nfsvers=3` 필수
+- **GPU**: Ollama 단일 컨테이너가 GPU0(EXAONE 추론)·GPU1(bge-m3 임베딩) 동시 사용. `runtime: nvidia` 필요
+- **KIS**: 현재 모의투자 모드 (`KIS_IS_PAPER=true`)
+- **바른API**: KRX→Naver 폴백으로 ~3768 종목 로드. 형태소 실패 시 공백분리 폴백
+
+---
+
+## 현재 이슈 / TODO
+
+- [ ] fstab NAS 항목이 두 줄로 분리됨 — sudo로 직접 수정 필요
+- [ ] n8n 워크플로우 재구성 필요 (재부팅 후 소실)
+- [x] vLLM 제거 → Ollama 단일 운영으로 전환 (EXAONE 3.5 7.8B + bge-m3)
+- [x] n8n DB 마이그레이션 오류 수정 완료 (51개 마이그레이션 등록)
+- [x] SSL 인증서 자동 갱신 설정 완료 (certbot.timer + deploy hook → nginx reload)
+      도메인: al/cla/n8/pns/tr.kyleyang.co.kr, 훅: /etc/letsencrypt/renewal-hooks/deploy/reload-nginx.sh
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8787
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8787", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8282
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8282", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,367 @@
+"""
+Aux Signal Service (port 8282, 172.30.0.25)
+
+외부 보조 데이터 수집:
+- 네이버 종목 integration API → 컨센서스 + 기관/외국인 일별 수급
+- 한국은행 ECOS API → USD/KRW 환율, 국고채 10년 금리
+"""
+import os
+import asyncio
+import json
+from datetime import date, datetime, timedelta
+from typing import Optional, List
+
+import asyncpg
+import structlog
+import httpx
+from fastapi import FastAPI, Query, BackgroundTasks
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from apscheduler.triggers.cron import CronTrigger
+from pytz import timezone
+
+# ─────────────────────────────────────────────────────────────
+# 설정
+# ─────────────────────────────────────────────────────────────
+PG = {
+    "host": os.getenv("POSTGRES_HOST", "postgres"),
+    "port": int(os.getenv("POSTGRES_PORT", 5432)),
+    "database": os.getenv("POSTGRES_DB", "trading_ai"),
+    "user": os.getenv("POSTGRES_USER", "kyu"),
+    "password": os.getenv("POSTGRES_PASSWORD", ""),
+}
+KST = timezone("Asia/Seoul")
+ECOS_KEY = os.getenv("ECOS_API_KEY", "")
+
+logger = structlog.get_logger()
+app = FastAPI(title="Aux Signal")
+pg_pool: Optional[asyncpg.Pool] = None
+scheduler = AsyncIOScheduler(timezone=KST)
+
+# ─────────────────────────────────────────────────────────────
+# DDL
+# ─────────────────────────────────────────────────────────────
+DDL = """
+CREATE TABLE IF NOT EXISTS analyst_consensus (
+    stock_code     VARCHAR(10) PRIMARY KEY,
+    target_price   BIGINT DEFAULT 0,
+    recomm_mean    DOUBLE PRECISION DEFAULT 0,
+    create_date    DATE,
+    updated_at     TIMESTAMP DEFAULT NOW()
+);
+
+CREATE TABLE IF NOT EXISTS inst_daily_flow (
+    stock_code         VARCHAR(10),
+    trade_date         DATE,
+    foreign_net        BIGINT DEFAULT 0,
+    inst_net           BIGINT DEFAULT 0,
+    individual_net     BIGINT DEFAULT 0,
+    foreign_hold_ratio DOUBLE PRECISION DEFAULT 0,
+    close_price        BIGINT DEFAULT 0,
+    change_amount      BIGINT DEFAULT 0,
+    PRIMARY KEY (stock_code, trade_date)
+);
+CREATE INDEX IF NOT EXISTS idx_flow_stock ON inst_daily_flow(stock_code, trade_date DESC);
+
+CREATE TABLE IF NOT EXISTS macro_daily (
+    indicator   VARCHAR(20),
+    trade_date  DATE,
+    value       DOUBLE PRECISION,
+    created_at  TIMESTAMP DEFAULT NOW(),
+    PRIMARY KEY (indicator, trade_date)
+);
+"""
+
+# ─────────────────────────────────────────────────────────────
+# Helpers
+# ─────────────────────────────────────────────────────────────
+def _parse_int(v) -> int:
+    if not v or v in ("-", ""):
+        return 0
+    s = str(v).replace(",", "").replace("+", "").replace(" ", "").strip()
+    try: return int(s)
+    except Exception: return 0
+
+
+def _parse_float(v) -> float:
+    if not v or v in ("-", ""):
+        return 0.0
+    s = str(v).replace(",", "").replace("%", "").replace("+", "").replace(" ", "").strip()
+    try: return float(s)
+    except Exception: return 0.0
+
+
+# ─────────────────────────────────────────────────────────────
+# 네이버 종목 integration API
+# 응답: consensusInfo + dealTrendInfos (60일치 일별 매매동향)
+# ─────────────────────────────────────────────────────────────
+async def fetch_naver(client: httpx.AsyncClient, code: str) -> Optional[dict]:
+    url = f"https://m.stock.naver.com/api/stock/{code}/integration"
+    try:
+        r = await client.get(url, headers={"User-Agent": "Mozilla/5.0"}, timeout=15)
+        if r.status_code != 200:
+            return None
+        return r.json()
+    except Exception as e:
+        logger.debug("naver.req_err", code=code, err=str(e))
+        return None
+
+
+async def save_consensus(conn, code: str, ci: dict):
+    if not ci:
+        return False
+    tp = _parse_int(ci.get("priceTargetMean"))
+    rm = _parse_float(ci.get("recommMean"))
+    cd = ci.get("createDate")
+    cd_date = None
+    if cd:
+        try:
+            cd_date = datetime.strptime(cd, "%Y-%m-%d").date()
+        except Exception:
+            cd_date = None
+    if tp == 0 and rm == 0:
+        return False
+    await conn.execute("""
+        INSERT INTO analyst_consensus (stock_code, target_price, recomm_mean, create_date)
+        VALUES ($1, $2, $3, $4)
+        ON CONFLICT (stock_code) DO UPDATE
+          SET target_price=$2, recomm_mean=$3, create_date=$4, updated_at=NOW()
+    """, code, tp, rm, cd_date)
+    return True
+
+
+async def save_flow(conn, code: str, dt_infos: list) -> int:
+    saved = 0
+    for r in dt_infos:
+        try:
+            bz = r.get("bizdate")
+            if not bz: continue
+            trade_dt = datetime.strptime(bz, "%Y%m%d").date()
+            await conn.execute("""
+                INSERT INTO inst_daily_flow
+                  (stock_code, trade_date, foreign_net, inst_net,
+                   individual_net, foreign_hold_ratio,
+                   close_price, change_amount)
+                VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
+                ON CONFLICT (stock_code, trade_date) DO UPDATE
+                  SET foreign_net=$3, inst_net=$4, individual_net=$5,
+                      foreign_hold_ratio=$6, close_price=$7, change_amount=$8
+            """, code, trade_dt,
+                _parse_int(r.get("foreignerPureBuyQuant")),
+                _parse_int(r.get("organPureBuyQuant")),
+                _parse_int(r.get("individualPureBuyQuant")),
+                _parse_float(r.get("foreignerHoldRatio")),
+                _parse_int(r.get("closePrice")),
+                _parse_int(r.get("compareToPreviousClosePrice")))
+            saved += 1
+        except Exception as e:
+            logger.debug("flow.save_err", code=code, err=str(e))
+    return saved
+
+
+async def collect_naver_data(count: int = 500):
+    """시총 상위 N개 종목 컨센서스 + 일별 수급 수집"""
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT DISTINCT ON (d.stock_code) d.stock_code,
+                   COALESCE(p.market_cap, 0) AS mc
+            FROM dart_corps d
+            LEFT JOIN stock_prices p ON p.stock_code=d.stock_code
+            WHERE d.is_active=true
+            ORDER BY d.stock_code, p.collected_at DESC NULLS LAST
+        """)
+        top = sorted(rows, key=lambda r: -r["mc"])[:count]
+        logger.info("naver.collect_start", count=len(top))
+
+        cons_saved, flow_saved = 0, 0
+        async with httpx.AsyncClient() as client:
+            for i, row in enumerate(top):
+                code = row["stock_code"]
+                j = await fetch_naver(client, code)
+                if j:
+                    async with pg_pool.acquire() as c2:
+                        if await save_consensus(c2, code, j.get("consensusInfo") or {}):
+                            cons_saved += 1
+                        s = await save_flow(c2, code, j.get("dealTrendInfos") or [])
+                        flow_saved += s
+                # 네이버 rate-limit 회피
+                if i < len(top) - 1:
+                    await asyncio.sleep(0.3)
+    logger.info("naver.collect_done", consensus=cons_saved, flow_rows=flow_saved)
+    return {"consensus": cons_saved, "flow_rows": flow_saved}
+
+
+# ─────────────────────────────────────────────────────────────
+# ECOS 매크로
+# 731Y001 → 환율 (0000001=USD/KRW)
+# 817Y002 → 시장금리 (010195000=국고채 10년)
+# ─────────────────────────────────────────────────────────────
+ECOS_BASE = "https://ecos.bok.or.kr/api/StatisticSearch"
+
+ECOS_SERIES = [
+    # (indicator_name, stat_code, item_code1, freq)
+    ("usdkrw",     "731Y001", "0000001",   "D"),
+    ("kor_10y",    "817Y002", "010195000", "D"),
+    ("kor_3y",     "817Y002", "010190000", "D"),
+    ("kospi",      "802Y001", "0001000",   "D"),
+]
+
+
+async def fetch_ecos_series(client: httpx.AsyncClient, stat: str, item: str,
+                              freq: str = "D", days: int = 90) -> list:
+    if not ECOS_KEY:
+        return []
+    end = date.today().strftime("%Y%m%d")
+    start = (date.today() - timedelta(days=days)).strftime("%Y%m%d")
+    url = f"{ECOS_BASE}/{ECOS_KEY}/json/kr/1/200/{stat}/{freq}/{start}/{end}/{item}"
+    try:
+        r = await client.get(url, timeout=15)
+        if r.status_code != 200:
+            return []
+        j = r.json()
+        return (j.get("StatisticSearch") or {}).get("row", []) or []
+    except Exception as e:
+        logger.warning("ecos.req_err", stat=stat, item=item, err=str(e))
+        return []
+
+
+async def collect_ecos_macro(days: int = 90):
+    if not ECOS_KEY:
+        logger.error("ecos.no_key")
+        return {"saved": 0, "err": "ECOS_API_KEY missing"}
+    saved = 0
+    async with httpx.AsyncClient() as client:
+        for name, stat, item, freq in ECOS_SERIES:
+            rows = await fetch_ecos_series(client, stat, item, freq, days)
+            async with pg_pool.acquire() as conn:
+                for r in rows:
+                    try:
+                        t = r.get("TIME")
+                        v = r.get("DATA_VALUE")
+                        if not t or not v:
+                            continue
+                        dt = datetime.strptime(t, "%Y%m%d").date()
+                        await conn.execute("""
+                            INSERT INTO macro_daily (indicator, trade_date, value)
+                            VALUES ($1, $2, $3)
+                            ON CONFLICT (indicator, trade_date) DO UPDATE
+                              SET value=$3
+                        """, name, dt, float(v))
+                        saved += 1
+                    except Exception as e:
+                        logger.debug("ecos.save_err", err=str(e))
+            await asyncio.sleep(0.2)
+    logger.info("ecos.collect_done", saved=saved)
+    return {"saved": saved}
+
+
+# ─────────────────────────────────────────────────────────────
+# 시작/종료
+# ─────────────────────────────────────────────────────────────
+@app.on_event("startup")
+async def on_start():
+    global pg_pool
+    pg_pool = await asyncpg.create_pool(**PG, min_size=2, max_size=10)
+    async with pg_pool.acquire() as conn:
+        await conn.execute(DDL)
+
+    # 네이버 컨센서스/수급: 매일 평일 16:30 (장 마감 후)
+    scheduler.add_job(collect_naver_data, CronTrigger(
+        day_of_week="mon-fri", hour=16, minute=30),
+        id="naver_collect", replace_existing=True)
+    # ECOS 매크로: 매일 새벽 5시
+    scheduler.add_job(collect_ecos_macro, CronTrigger(hour=5),
+        id="ecos_macro", replace_existing=True)
+    scheduler.start()
+    logger.info("aux-signal.started")
+
+
+@app.on_event("shutdown")
+async def on_stop():
+    if scheduler.running:
+        scheduler.shutdown()
+    if pg_pool:
+        await pg_pool.close()
+
+
+# ─────────────────────────────────────────────────────────────
+# REST API
+# ─────────────────────────────────────────────────────────────
+@app.get("/health")
+async def health():
+    return {"ok": True, "service": "aux-signal",
+            "ts": datetime.now(KST).isoformat()}
+
+
+@app.post("/collect/naver")
+async def manual_naver(count: int = Query(default=500, ge=10, le=2000),
+                        bg: BackgroundTasks = None):
+    if bg:
+        bg.add_task(collect_naver_data, count)
+        return {"status": "queued", "count": count}
+    return await collect_naver_data(count)
+
+
+@app.post("/collect/macro")
+async def manual_macro(days: int = Query(default=90, ge=7, le=365)):
+    return await collect_ecos_macro(days)
+
+
+@app.get("/consensus/{code}")
+async def get_consensus(code: str):
+    async with pg_pool.acquire() as conn:
+        row = await conn.fetchrow(
+            "SELECT * FROM analyst_consensus WHERE stock_code=$1", code)
+        return dict(row) if row else {"err": "no data"}
+
+
+@app.get("/flow/{code}")
+async def get_flow(code: str, days: int = Query(default=30, ge=1, le=90)):
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT trade_date, foreign_net, inst_net, individual_net,
+                   foreign_hold_ratio, close_price, change_amount
+            FROM inst_daily_flow
+            WHERE stock_code=$1 AND trade_date >= CURRENT_DATE - $2::int
+            ORDER BY trade_date DESC
+        """, code, days)
+        return [dict(r) for r in rows]
+
+
+@app.get("/macro/latest")
+async def macro_latest():
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT DISTINCT ON (indicator) indicator, trade_date, value
+            FROM macro_daily
+            ORDER BY indicator, trade_date DESC
+        """)
+        return {r["indicator"]: {"date": str(r["trade_date"]),
+                                   "value": float(r["value"])}
+                for r in rows}
+
+
+@app.get("/macro/{indicator}")
+async def macro_series(indicator: str, days: int = Query(default=30, ge=1, le=365)):
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT trade_date, value FROM macro_daily
+            WHERE indicator=$1 AND trade_date >= CURRENT_DATE - $2::int
+            ORDER BY trade_date DESC
+        """, indicator, days)
+        return [{"date": str(r["trade_date"]), "value": float(r["value"])}
+                for r in rows]
+
+
+@app.get("/stats")
+async def stats():
+    async with pg_pool.acquire() as conn:
+        c = await conn.fetchval("SELECT COUNT(*) FROM analyst_consensus")
+        f = await conn.fetchrow(
+            "SELECT COUNT(*) AS rows, COUNT(DISTINCT stock_code) AS codes,"
+            " MAX(trade_date) AS latest FROM inst_daily_flow")
+        m = await conn.fetchrow(
+            "SELECT COUNT(*) AS rows, COUNT(DISTINCT indicator) AS indicators,"
+            " MAX(trade_date) AS latest FROM macro_daily")
+        return {"consensus_stocks": c,
+                "flow": dict(f) if f else {},
+                "macro": dict(m) if m else {}}
@@ -0,0 +1,8 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+asyncpg==0.29.0
+apscheduler==3.10.4
+structlog==24.2.0
+httpx==0.27.0
+orjson==3.10.3
+pytz==2024.1
@@ -0,0 +1,204 @@
+"""
+바른 API FastAPI 서버 v2
+- 서버 시작 시 KRX 전체 종목 동적 로딩
+- 24시간마다 자동 갱신
+- Bareun gRPC 연결
+"""
+import asyncio, hashlib, os, time
+from contextlib import asynccontextmanager
+from typing import Optional
+import orjson, redis.asyncio as aioredis, structlog
+from bareunpy import Tagger
+from fastapi import FastAPI, HTTPException, Response
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+from prometheus_fastapi_instrumentator import Instrumentator
+from pydantic import BaseModel, Field
+from stock_loader import load_all_stocks, auto_refresh_stocks
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+BAREUN_API_KEY     = os.getenv("BAREUN_API_KEY", "")
+BAREUN_SERVER_HOST = os.getenv("BAREUN_SERVER_HOST", "bareun")
+BAREUN_SERVER_PORT = int(os.getenv("BAREUN_SERVER_PORT", "5656"))
+REDIS_HOST         = os.getenv("REDIS_HOST", "redis")
+REDIS_PORT         = int(os.getenv("REDIS_PORT", "6379"))
+REDIS_PASSWORD     = os.getenv("REDIS_PASSWORD", "")
+REDIS_DB           = int(os.getenv("REDIS_DB", "1"))
+NEWS_DEDUP_TTL     = int(os.getenv("NEWS_DEDUP_TTL", "86400"))
+
+STOPWORDS = {
+    "것","수","등","및","또","또한","이","그","저","위","아래","관련","통해",
+    "위해","대해","따라","때문","이후","이전","현재","최근","지난","올해","내년",
+    "이번","오늘","어제","내일","이날","같은","다른","많은","더","가장","매우",
+    "이미","아직","모든","각","전체","일부","국내","해외","글로벌","세계","한국",
+    "미국","중국","일본","유럽","가운데","한편","다만","특히","실제","여전히",
+    "앞으로","지속","계속","자체","관계자","측","쪽","곳","점","경우","상황",
+    "때","중","간",
+}
+FINANCE_KEYWORDS = {
+    "실적","매출","영업이익","순이익","적자","흑자","손실","이익","투자","배당",
+    "자사주","유상증자","합병","인수","M&A","IPO","상장","상폐","거래정지","금리",
+    "기준금리","환율","달러","반도체","배터리","전기차","수소","AI","인공지능",
+    "클라우드","바이오","의약품","임상","승인","특허","수주","계약","부도","파산",
+    "부채","시총","코스피","코스닥","나스닥",
+}
+START_TIME = time.time()
+
+class AppState:
+    tagger: Optional[Tagger] = None
+    redis: Optional[aioredis.Redis] = None
+    stock_map: dict[str, str] = {}
+    stock_count: int = 0
+    refresh_task: Optional[asyncio.Task] = None
+
+state = AppState()
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    state.stock_map = await load_all_stocks()
+    state.stock_count = len(state.stock_map)
+    logger.info("stocks.ready", count=state.stock_count)
+    state.refresh_task = asyncio.create_task(auto_refresh_stocks(state, 24))
+    try:
+        state.tagger = Tagger(BAREUN_API_KEY, BAREUN_SERVER_HOST, BAREUN_SERVER_PORT)
+        logger.info("tagger.ok")
+    except Exception as e:
+        logger.warning("tagger.failed", error=str(e))
+    try:
+        state.redis = aioredis.Redis(host=REDIS_HOST, port=REDIS_PORT,
+            password=REDIS_PASSWORD, db=REDIS_DB, decode_responses=True,
+            socket_connect_timeout=5, retry_on_timeout=True)
+        await state.redis.ping()
+        logger.info("redis.ok")
+    except Exception as e:
+        logger.error("redis.failed", error=str(e))
+    yield
+    if state.refresh_task: state.refresh_task.cancel()
+    if state.redis: await state.redis.aclose()
+
+app = FastAPI(title="바른 API v2", version="2.0.0", lifespan=lifespan)
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+Instrumentator().instrument(app).expose(app, endpoint="/metrics")
+
+class AnalyzeRequest(BaseModel):
+    title: str; content: str = ""; url: str = ""; source: str = ""; published_at: str = ""
+
+class StockMention(BaseModel):
+    name: str; code: str; count: int
+
+class AnalyzeResponse(BaseModel):
+    hash: str; is_duplicate: bool; stocks: list[StockMention]; keywords: list[str]
+    filtered_text: str; token_count: int; processing_time_ms: float
+
+class BatchRequest(BaseModel):
+    items: list[AnalyzeRequest]
+
+def news_hash(title, url):
+    return hashlib.sha256(f"{title.strip()}{url.strip()}".encode()).hexdigest()[:16]
+
+async def is_duplicate(h):
+    if not state.redis: return False
+    try:
+        r = await state.redis.set(f"news:dedup:{h}", "1", ex=NEWS_DEDUP_TTL, nx=True)
+        return r is None
+    except: return False
+
+def extract_morphemes(text):
+    if not state.tagger or not text.strip():
+        return [w for w in text.split() if len(w) >= 2 and w not in STOPWORDS]
+    try:
+        return [t for t, p in state.tagger.pos(text)
+                if p in ("NNG","NNP","SL") and len(t) >= 2 and t not in STOPWORDS]
+    except:
+        return [w for w in text.split() if len(w) >= 2 and w not in STOPWORDS]
+
+def extract_stocks(text):
+    found = {}
+    for name, code in state.stock_map.items():
+        c = text.count(name)
+        if c > 0: found[name] = StockMention(name=name, code=code, count=c)
+    return sorted(found.values(), key=lambda x: x.count, reverse=True)
+
+def build_filtered(nouns, stocks):
+    sn = {s.name for s in stocks}
+    p = [n for n in nouns if n in sn]
+    f = [n for n in nouns if n in FINANCE_KEYWORDS]
+    g = [n for n in nouns if n not in STOPWORDS and len(n) >= 2]
+    return " ".join(list(dict.fromkeys(p + f + g))[:100])
+
+def _analyze(req):
+    text = f"{req.title} {req.content}".strip()
+    h = news_hash(req.title, req.url)
+    nouns = extract_morphemes(text)
+    stocks = extract_stocks(text)
+    kw = list(dict.fromkeys(n for n in nouns if len(n) >= 2))[:50]
+    ft = build_filtered(nouns, stocks)
+    return h, stocks, kw, ft
+
+@app.get("/health")
+async def health():
+    tok = state.tagger is not None
+    rok = False
+    if state.redis:
+        try: await state.redis.ping(); rok = True
+        except: pass
+    return JSONResponse(content={"status": "ok" if tok else "degraded",
+        "tagger": "ok" if tok else "unavailable", "redis": "ok" if rok else "error",
+        "stocks_loaded": state.stock_count, "uptime": round(time.time()-START_TIME,1)})
+
+@app.post("/analyze")
+async def analyze(req: AnalyzeRequest):
+    t = time.perf_counter()
+    h, stocks, kw, ft = _analyze(req)
+    dup = await is_duplicate(h)
+    ms = round((time.perf_counter()-t)*1000, 2)
+    return Response(content=orjson.dumps(AnalyzeResponse(
+        hash=h, is_duplicate=dup, stocks=stocks, keywords=kw,
+        filtered_text=ft, token_count=len(ft.split()), processing_time_ms=ms
+    ).model_dump()), media_type="application/json")
+
+@app.post("/analyze/batch")
+async def analyze_batch(req: BatchRequest):
+    if len(req.items) > 50: raise HTTPException(400, "최대 50개")
+    t = time.perf_counter()
+    results = []
+    for item in req.items:
+        try:
+            h, stocks, kw, ft = _analyze(item)
+            dup = await is_duplicate(h)
+            results.append({"title":item.title,"hash":h,"is_duplicate":dup,
+                "stocks":[s.model_dump() for s in stocks],"keywords":kw,
+                "filtered_text":ft,"token_count":len(ft.split())})
+        except Exception as e:
+            results.append({"title":item.title,"error":str(e),"is_duplicate":False})
+    ms = round((time.perf_counter()-t)*1000, 2)
+    dups = sum(1 for r in results if r.get("is_duplicate"))
+    return Response(content=orjson.dumps({"total":len(results),"duplicates":dups,
+        "processed":len(results)-dups,"elapsed_ms":ms,"results":results}),
+        media_type="application/json")
+
+@app.get("/stocks")
+async def stocks_list():
+    return JSONResponse(content={"count":len(state.stock_map),
+        "stocks":[{"name":k,"code":v} for k,v in list(state.stock_map.items())[:500]]})
+
+@app.post("/stocks/refresh")
+async def refresh():
+    m = await load_all_stocks()
+    if len(m) > 50:
+        state.stock_map = m; state.stock_count = len(m)
+        return JSONResponse(content={"status":"ok","count":len(m)})
+    raise HTTPException(500, "종목 로딩 실패")
+
+@app.delete("/dedup/flush")
+async def flush():
+    if not state.redis: raise HTTPException(503)
+    keys = await state.redis.keys("news:dedup:*")
+    if keys: await state.redis.delete(*keys)
+    return {"deleted": len(keys)}
@@ -0,0 +1,24 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+RUN apt-get update && apt-get install -y \
+    curl \
+    gcc \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN mkdir -p /app/logs
+
+EXPOSE 5757
+
+CMD ["python", "-m", "uvicorn", "main:app", \
+     "--host", "0.0.0.0", \
+     "--port", "5757", \
+     "--workers", "4", \
+     "--loop", "uvloop", \
+     "--http", "httptools", \
+     "--log-level", "info"]
@@ -0,0 +1,136 @@
+"""
+한국 주식/금융 전문 용어 사전
+형태소 분석 보완용 — 바른 API가 분리하지 못하는 복합어 및 금융 전문어를 정의
+"""
+
+# ── 재무제표 / 회계 ─────────────────────────────────────────────────────────────
+FINANCIAL_STATEMENT = {
+    "매출액","매출원가","매출총이익","영업이익","영업손실","영업외수익","영업외비용",
+    "세전이익","당기순이익","당기순손실","포괄손익","이익잉여금","자본잉여금",
+    "납입자본","기타자본","비지배지분","총자산","총부채","총자본","유동자산",
+    "비유동자산","유동부채","비유동부채","현금및현금성자산","단기금융상품",
+    "매출채권","재고자산","유형자산","무형자산","투자자산","영업활동현금흐름",
+    "투자활동현금흐름","재무활동현금흐름","잉여현금흐름","FCF","EBITDA","EBIT",
+    "ROE","ROA","ROI","EPS","BPS","DPS","PER","PBR","PSR","EV/EBITDA","PCR",
+    "부채비율","유동비율","당좌비율","자기자본비율","이자보상배율","영업이익률",
+    "순이익률","자산회전율","재고자산회전율","매출채권회전율",
+}
+
+# ── 주식 시장 구조 ───────────────────────────────────────────────────────────────
+MARKET_STRUCTURE = {
+    "코스피","코스닥","코넥스","K-OTC","유가증권시장","장외시장","선물시장",
+    "옵션시장","ETF","ETN","ELS","DLS","리츠","SPAC","공모주","유상증자","무상증자",
+    "감자","합병","분할","인적분할","물적분할","자사주매입","자사주소각","배당",
+    "중간배당","특별배당","주식배당","현금배당","우선주","보통주","전환사채","CB",
+    "신주인수권부사채","BW","교환사채","EB","DR","GDR","ADR","상장","상장폐지",
+    "거래정지","매매정지","관리종목","투자주의","투자경고","투자위험","단기과열",
+    "공매도","대차거래","프로그램매매","외국인","기관","개인","세력","수급",
+    "시가총액","거래량","거래대금","52주최고","52주최저","신고가","신저가",
+}
+
+# ── 기술적 분석 지표 ─────────────────────────────────────────────────────────────
+TECHNICAL_INDICATORS = {
+    "이동평균","이평선","골든크로스","데드크로스","MACD","RSI","스토캐스틱",
+    "볼린저밴드","볼밴","일목균형표","지지","저항","추세선","채널","삼각수렴",
+    "쐐기형","헤드앤숄더","역헤드앤숄더","더블탑","더블바텀","갭","갭업","갭다운",
+    "윗꼬리","아래꼬리","망치형","도지","장악형","상승반전","하락반전",
+    "거래량폭발","거래량급증","음봉","양봉","캔들","봉차트","분봉","일봉","주봉","월봉",
+    "오버솔드","오버바우트","과매수","과매도","다이버전스","모멘텀",
+}
+
+# ── 기업 이벤트 / 공시 ───────────────────────────────────────────────────────────
+CORPORATE_EVENTS = {
+    "실적발표","어닝시즌","어닝서프라이즈","어닝쇼크","가이던스","실적전망",
+    "수주","수주잔고","계약체결","MOU","LOI","업무협약","전략적제휴","조인트벤처",
+    "JV","M&A","인수합병","적대적인수","공개매수","TOB","경영권분쟁","주주행동주의",
+    "주주총회","정기주총","임시주총","이사회","대표이사교체","CEO교체","감사의견",
+    "한정의견","의견거절","부적정의견","횡령","배임","회계부정","분식회계","리콜",
+    "소송","피소","과징금","제재","영업정지","특허침해","임상성공","임상실패",
+    "FDA승인","식약처승인","NDA","IND","임상1상","임상2상","임상3상",
+}
+
+# ── 거시경제 / 정책 ──────────────────────────────────────────────────────────────
+MACRO_ECONOMICS = {
+    "기준금리","금리인상","금리인하","금리동결","통화정책","양적완화","QE","QT",
+    "테이퍼링","연준","FED","FOMC","한국은행","한은","ECB","BOJ","금통위",
+    "환율","달러원","원달러","엔화","위안화","유로","달러인덱스","DXY",
+    "인플레이션","디플레이션","스태그플레이션","CPI","PPI","PCE","GDP","GNP",
+    "경상수지","무역수지","외환보유고","국채","회사채","하이일드","신용등급",
+    "무디스","S&P","피치","국제유가","WTI","브렌트","천연가스","LNG","구리","금","은",
+    "반도체수급","공급망","서플라이체인","리쇼어링","니어쇼어링","관세","무역전쟁",
+    "경제성장률","실업률","소비자심리지수","PMI","ISM","주택착공","소매판매",
+}
+
+# ── 산업 / 섹터 키워드 ───────────────────────────────────────────────────────────
+SECTOR_KEYWORDS = {
+    # 반도체
+    "반도체","메모리","낸드","D램","HBM","HBM3","HBM4","파운드리","팹리스",
+    "TSMC","삼성파운드리","파운드리","EUV","GAA","3나노","2나노","첨단패키징",
+    "CoWoS","인터포저","칩렛","AI반도체","NPU","GPU","AI가속기",
+    # 2차전지/전기차
+    "배터리","2차전지","리튬이온","전고체배터리","양극재","음극재","분리막","전해질",
+    "전기차","BEV","PHEV","HEV","충전인프라","급속충전","에너지밀도","배터리팩",
+    "배터리셀","모듈","BMS","리튬","코발트","니켈","망간","황산리튬",
+    # 바이오/헬스
+    "바이오","제약","의약품","신약","오리지널","제네릭","바이오시밀러","항체","mRNA",
+    "세포치료","유전자치료","CAR-T","ADC","이중항체","희귀질환","항암제","치매",
+    "알츠하이머","비만치료제","GLP-1","당뇨","심혈관","호흡기",
+    # AI/IT/플랫폼
+    "인공지능","AI","LLM","GPT","클로드","제미나이","거대언어모델","생성AI","AGI",
+    "클라우드","AWS","애저","GCP","SaaS","PaaS","IaaS","데이터센터","GPU서버",
+    "반도체설계","엣지AI","온디바이스","자율주행","ADAS","라이다","카메라모듈",
+    "로봇","협동로봇","물류로봇","휴머노이드",
+    # 방산/우주
+    "방산","방위산업","무기수출","미사일","드론","무인기","위성","우주항공","발사체",
+    # 건설/부동산
+    "부동산","아파트","분양","착공","준공","건설수주","PF","프로젝트파이낸싱",
+    "리츠","REITs","임대","역세권","재개발","재건축",
+}
+
+# ── 가격 움직임 표현 (감성 분석용) ───────────────────────────────────────────────
+PRICE_MOVEMENT = {
+    "급등","급락","폭등","폭락","상한가","하한가","강세","약세","반등","반락",
+    "상승","하락","보합","소폭상승","소폭하락","대폭상승","대폭하락",
+    "신고가경신","52주신고가","역사적신고가","저점매수","고점매도",
+    "눌림","눌림목","지지","저항돌파","박스권","횡보","추세전환",
+    "모멘텀강화","모멘텀약화","수급개선","수급악화","외국인순매수","외국인순매도",
+    "기관순매수","기관순매도","공매도잔고","공매도비율",
+}
+
+# ── 투자 전략 / 버핏 스타일 ──────────────────────────────────────────────────────
+INVESTMENT_STRATEGY = {
+    "가치투자","성장투자","배당투자","퀀트투자","모멘텀투자","역발상투자",
+    "내재가치","안전마진","경제적해자","모트","경쟁우위","진입장벽","가격결정력",
+    "브랜드가치","플랫폼효과","네트워크효과","전환비용","규모의경제","비용우위",
+    "장기투자","복리효과","분산투자","집중투자","포트폴리오","자산배분","리밸런싱",
+    "저PER","저PBR","고ROE","배당수익률","배당성향","주주환원","자본배분",
+    "현금창출력","잉여현금흐름","재투자수익률","ROIC","WACC",
+}
+
+# 동사/형용사 형태 (형태소 분석에서 VV/VA 태그로 뽑히는 것들)
+FINANCE_VERBS: set[str] = {
+    "급등","급락","폭등","폭락","상승","하락","반등","돌파","이탈","회복",
+    "초과","달성","하회","상회","급성장","개선","악화","감소","증가",
+}
+
+# 전체 통합 사전 (lookup용)
+ALL_FINANCE_TERMS: set[str] = (
+    FINANCIAL_STATEMENT
+    | MARKET_STRUCTURE
+    | TECHNICAL_INDICATORS
+    | CORPORATE_EVENTS
+    | MACRO_ECONOMICS
+    | SECTOR_KEYWORDS
+    | PRICE_MOVEMENT
+    | INVESTMENT_STRATEGY
+)
+
+# 감성 판단에 중요한 우선순위 높은 단어들
+HIGH_PRIORITY_TERMS: set[str] = {
+    "어닝서프라이즈","어닝쇼크","급등","급락","폭등","폭락","상한가","하한가",
+    "수주","임상성공","임상실패","FDA승인","횡령","배임","분식회계","상장폐지",
+    "관리종목","거래정지","배당증가","자사주소각","M&A","합병","분할",
+    "골든크로스","데드크로스","신고가","신저가","52주신고가","52주신저가",
+    "외국인순매수","기관순매수","공매도잔고감소","실적발표","가이던스상향",
+    "가이던스하향","CEO교체","대표이사교체","감사의견거절",
+}
@@ -0,0 +1,250 @@
+"""
+바른 API FastAPI 서버 v2
+- 서버 시작 시 KRX 전체 종목 동적 로딩
+- 24시간마다 자동 갱신
+- Bareun gRPC 연결
+"""
+import asyncio, hashlib, os, re, time
+from contextlib import asynccontextmanager
+from typing import Optional
+import orjson, redis.asyncio as aioredis, structlog
+from bareunpy import Tagger
+from fastapi import FastAPI, HTTPException, Response
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+from prometheus_fastapi_instrumentator import Instrumentator
+from pydantic import BaseModel, Field
+from stock_loader import load_all_stocks, auto_refresh_stocks
+from finance_dict import ALL_FINANCE_TERMS, HIGH_PRIORITY_TERMS, FINANCE_VERBS as FD_VERBS
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+BAREUN_API_KEY     = os.getenv("BAREUN_API_KEY", "")
+BAREUN_SERVER_HOST = os.getenv("BAREUN_SERVER_HOST", "bareun")
+BAREUN_SERVER_PORT = int(os.getenv("BAREUN_SERVER_PORT", "5656"))
+REDIS_HOST         = os.getenv("REDIS_HOST", "redis")
+REDIS_PORT         = int(os.getenv("REDIS_PORT", "6379"))
+REDIS_PASSWORD     = os.getenv("REDIS_PASSWORD", "")
+REDIS_DB           = int(os.getenv("REDIS_DB", "1"))
+NEWS_DEDUP_TTL     = int(os.getenv("NEWS_DEDUP_TTL", "86400"))
+
+STOPWORDS = {
+    "것","수","등","및","또","또한","이","그","저","위","아래","관련","통해",
+    "위해","대해","따라","때문","이후","이전","현재","최근","지난","올해","내년",
+    "이번","오늘","어제","내일","이날","같은","다른","많은","더","가장","매우",
+    "이미","아직","모든","각","전체","일부","국내","해외","글로벌","세계","한국",
+    "미국","중국","일본","유럽","가운데","한편","다만","특히","실제","여전히",
+    "앞으로","지속","계속","자체","관계자","측","쪽","곳","점","경우","상황",
+    "때","중","간",
+}
+FINANCE_KEYWORDS = ALL_FINANCE_TERMS
+FINANCE_VERBS = {
+    "급등","급락","폭등","폭락","상승","하락","반등","하락세","상승세",
+    "돌파","이탈","회복","저항","지지","돌파구","매집","매도","매수",
+    "초과달성","하회","상회","달성","부진","급성장","감소","증가","개선","악화",
+} | FD_VERBS
+START_TIME = time.time()
+
+class AppState:
+    tagger: Optional[Tagger] = None
+    redis: Optional[aioredis.Redis] = None
+    stock_map: dict[str, str] = {}
+    stock_count: int = 0
+    refresh_task: Optional[asyncio.Task] = None
+
+state = AppState()
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    state.stock_map = await load_all_stocks()
+    state.stock_count = len(state.stock_map)
+    logger.info("stocks.ready", count=state.stock_count)
+    state.refresh_task = asyncio.create_task(auto_refresh_stocks(state, 24))
+    try:
+        state.tagger = Tagger(BAREUN_API_KEY, BAREUN_SERVER_HOST, BAREUN_SERVER_PORT)
+        logger.info("tagger.ok")
+    except Exception as e:
+        logger.warning("tagger.failed", error=str(e))
+    try:
+        state.redis = aioredis.Redis(host=REDIS_HOST, port=REDIS_PORT,
+            password=REDIS_PASSWORD, db=REDIS_DB, decode_responses=True,
+            socket_connect_timeout=5, retry_on_timeout=True)
+        await state.redis.ping()
+        logger.info("redis.ok")
+    except Exception as e:
+        logger.error("redis.failed", error=str(e))
+    yield
+    if state.refresh_task: state.refresh_task.cancel()
+    if state.redis: await state.redis.aclose()
+
+app = FastAPI(title="바른 API v2", version="2.0.0", lifespan=lifespan)
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+Instrumentator().instrument(app).expose(app, endpoint="/metrics")
+
+class AnalyzeRequest(BaseModel):
+    title: str; content: str = ""; url: str = ""; source: str = ""; published_at: str = ""
+
+class StockMention(BaseModel):
+    name: str; code: str; count: int
+
+class AnalyzeResponse(BaseModel):
+    hash: str; is_duplicate: bool; stocks: list[StockMention]; keywords: list[str]
+    filtered_text: str; token_count: int; processing_time_ms: float
+
+class BatchRequest(BaseModel):
+    items: list[AnalyzeRequest]
+
+def news_hash(title, url):
+    return hashlib.sha256(f"{title.strip()}{url.strip()}".encode()).hexdigest()[:16]
+
+async def is_duplicate(h):
+    if not state.redis: return False
+    try:
+        r = await state.redis.set(f"news:dedup:{h}", "1", ex=NEWS_DEDUP_TTL, nx=True)
+        return r is None
+    except: return False
+
+def extract_morphemes(text):
+    if not state.tagger or not text.strip():
+        return [w for w in text.split() if len(w) >= 2 and w not in STOPWORDS]
+    try:
+        result = []
+        for t, p in state.tagger.pos(text):
+            if len(t) < 2 or t in STOPWORDS:
+                continue
+            if p in ("NNG", "NNP", "SL"):
+                result.append(t)
+            elif p in ("VV", "VA", "XR") and t in FINANCE_VERBS:
+                result.append(t)
+        return result
+    except:
+        return [w for w in text.split() if len(w) >= 2 and w not in STOPWORDS]
+
+def extract_stocks(text):
+    found = {}
+    for name, code in state.stock_map.items():
+        if len(name) < 2:
+            continue                       # 1글자 종목명은 오탐 과다 → 제외
+        if name.isascii():
+            # 영문/숫자 약칭(KT·SK·DB 등)은 단어경계 강제 (SKT·KTX 오탐 차단)
+            pat = rf"(?<![A-Za-z0-9]){re.escape(name)}(?![A-Za-z0-9])"
+            c = len(re.findall(pat, text))
+        else:
+            c = text.count(name)           # 한글명: 교착어 특성상 부분일치 유지
+        if c > 0:
+            found[name] = StockMention(name=name, code=code, count=c)
+    # 더 긴 종목명에 포함된 짧은 종목명 제거 ('한국' ⊂ '한국전력', 'KT' ⊂ 'KT&G')
+    names = list(found)
+    for n in names:
+        if any(n != m and n in m for m in names):
+            found.pop(n, None)
+    return sorted(found.values(), key=lambda x: x.count, reverse=True)
+
+def build_filtered(nouns, stocks):
+    sn = {s.name for s in stocks}
+    seen = set()
+    result = []
+    # 1순위: 고중요도 금융 이벤트 (어닝서프라이즈, 상한가 등)
+    for n in nouns:
+        if n in HIGH_PRIORITY_TERMS and n not in seen:
+            result.append(n); seen.add(n)
+    # 2순위: 종목명
+    for n in nouns:
+        if n in sn and n not in seen:
+            result.append(n); seen.add(n)
+    # 3순위: 전문 금융 용어 사전
+    for n in nouns:
+        if n in FINANCE_KEYWORDS and n not in seen:
+            result.append(n); seen.add(n)
+    # 4순위: 나머지 의미있는 명사
+    for n in nouns:
+        if n not in STOPWORDS and len(n) >= 2 and n not in seen:
+            result.append(n); seen.add(n)
+    return " ".join(result[:120])
+
+def scan_finance_terms(text: str) -> list[str]:
+    """형태소 분석 없이 원문에서 금융 전문 용어 직접 탐색 (복합어 대응)"""
+    found = []
+    for term in ALL_FINANCE_TERMS:
+        if term in text:
+            found.append(term)
+    return found
+
+def _analyze(req):
+    text = f"{req.title} {req.content}".strip()
+    h = news_hash(req.title, req.url)
+    nouns = extract_morphemes(text)
+    # 원문 직접 스캔으로 형태소 분석이 놓친 복합어 추가
+    direct_terms = scan_finance_terms(text)
+    nouns = list(dict.fromkeys(nouns + direct_terms))
+    stocks = extract_stocks(text)
+    kw = list(dict.fromkeys(n for n in nouns if len(n) >= 2))[:50]
+    ft = build_filtered(nouns, stocks)
+    return h, stocks, kw, ft
+
+@app.get("/health")
+async def health():
+    tok = state.tagger is not None
+    rok = False
+    if state.redis:
+        try: await state.redis.ping(); rok = True
+        except: pass
+    return JSONResponse(content={"status": "ok" if tok else "degraded",
+        "tagger": "ok" if tok else "unavailable", "redis": "ok" if rok else "error",
+        "stocks_loaded": state.stock_count, "uptime": round(time.time()-START_TIME,1)})
+
+@app.post("/analyze")
+async def analyze(req: AnalyzeRequest):
+    t = time.perf_counter()
+    h, stocks, kw, ft = _analyze(req)
+    dup = await is_duplicate(h)
+    ms = round((time.perf_counter()-t)*1000, 2)
+    return Response(content=orjson.dumps(AnalyzeResponse(
+        hash=h, is_duplicate=dup, stocks=stocks, keywords=kw,
+        filtered_text=ft, token_count=len(ft.split()), processing_time_ms=ms
+    ).model_dump()), media_type="application/json")
+
+@app.post("/analyze/batch")
+async def analyze_batch(req: BatchRequest):
+    if len(req.items) > 50: raise HTTPException(400, "최대 50개")
+    t = time.perf_counter()
+    results = []
+    for item in req.items:
+        try:
+            h, stocks, kw, ft = _analyze(item)
+            dup = await is_duplicate(h)
+            results.append({"title":item.title,"hash":h,"is_duplicate":dup,
+                "stocks":[s.model_dump() for s in stocks],"keywords":kw,
+                "filtered_text":ft,"token_count":len(ft.split())})
+        except Exception as e:
+            results.append({"title":item.title,"error":str(e),"is_duplicate":False})
+    ms = round((time.perf_counter()-t)*1000, 2)
+    dups = sum(1 for r in results if r.get("is_duplicate"))
+    return Response(content=orjson.dumps({"total":len(results),"duplicates":dups,
+        "processed":len(results)-dups,"elapsed_ms":ms,"results":results}),
+        media_type="application/json")
+
+@app.get("/stocks")
+async def stocks_list():
+    return JSONResponse(content={"count":len(state.stock_map),
+        "stocks":[{"name":k,"code":v} for k,v in list(state.stock_map.items())[:500]]})
+
+@app.post("/stocks/refresh")
+async def refresh():
+    m = await load_all_stocks()
+    if len(m) > 50:
+        state.stock_map = m; state.stock_count = len(m)
+        return JSONResponse(content={"status":"ok","count":len(m)})
+    raise HTTPException(500, "종목 로딩 실패")
+
+@app.delete("/dedup/flush")
+async def flush():
+    if not state.redis: raise HTTPException(503)
+    keys = await state.redis.keys("news:dedup:*")
+    if keys: await state.redis.delete(*keys)
+    return {"deleted": len(keys)}
@@ -0,0 +1,14 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+uvloop==0.19.0
+httptools==0.6.1
+httpx==0.27.0
+redis==5.0.4
+pydantic==2.7.1
+pydantic-settings==2.3.1
+bareunpy==1.7.3
+python-multipart==0.0.9
+orjson==3.10.3
+structlog==24.2.0
+prometheus-fastapi-instrumentator==7.0.0
+tenacity==8.3.0
@@ -0,0 +1,162 @@
+"""
+KRX 전체 종목 동적 로딩
+- 서버 시작 시 1회 로딩
+- 매일 자정 자동 갱신
+- 실패 시 하드코딩 폴백
+"""
+
+import asyncio
+import json
+import io
+import time
+from datetime import datetime
+
+import httpx
+import structlog
+
+logger = structlog.get_logger()
+
+# 폴백 종목 (네트워크 실패 시 사용)
+FALLBACK_STOCKS = {
+    "코스피": "KOSPI", "코스닥": "KOSDAQ", "코스피200": "KOSPI200",
+    "삼성전자": "005930", "SK하이닉스": "000660", "LG에너지솔루션": "373220",
+    "현대차": "005380", "기아": "000270", "셀트리온": "068270",
+    "카카오": "035720", "네이버": "035420", "NAVER": "035420",
+    "삼성바이오로직스": "207940", "KB금융": "105560", "POSCO홀딩스": "005490",
+    "신한지주": "055550", "LG화학": "051910", "삼성SDI": "006400",
+    "현대모비스": "012330", "하나금융지주": "086790", "SK텔레콤": "017670",
+    "KT": "030200", "LG전자": "066570", "한화에어로스페이스": "012450",
+    "삼성물산": "028260", "HD현대중공업": "329180", "한국전력": "015760",
+    "HMM": "011200", "대한항공": "003490", "카카오뱅크": "323410",
+    "크래프톤": "259960", "하이브": "352820", "에코프로": "086520",
+    "알테오젠": "196170", "한미반도체": "042700",
+}
+
+# 약칭/별칭
+ALIASES = {
+    "삼전": "005930", "하닉": "000660", "현차": "005380",
+    "카뱅": "323410", "삼바": "207940", "삼성바이오": "207940",
+    "한에솔": "012450", "한화에어": "012450", "LG엔솔": "373220",
+    "SK하닉": "000660", "포홀": "005490", "에프엔에프": "383220",
+    "현대자동차": "005380", "네이버": "035420",
+}
+
+
+async def fetch_krx_stocks() -> dict[str, str]:
+    """KRX에서 전체 상장 종목 가져오기"""
+    stock_map = {
+        "코스피": "KOSPI", "코스닥": "KOSDAQ",
+        "코스피200": "KOSPI200", "KRX": "KRX",
+    }
+
+    headers = {
+        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36",
+        "Referer": "http://data.krx.co.kr/contents/MDC/MDI/mdiLoader/index.cmd",
+    }
+
+    async with httpx.AsyncClient(timeout=30, headers=headers) as client:
+        for mkt_id, mkt_name in [("STK", "KOSPI"), ("KSQ", "KOSDAQ")]:
+            try:
+                payload = {
+                    "bld": "dbms/MDC/STAT/standard/MDCSTAT01901",
+                    "locale": "ko_KR",
+                    "mktId": mkt_id,
+                    "share": "1",
+                    "csvxls_is498": "false",
+                }
+                resp = await client.post(
+                    "http://data.krx.co.kr/comm/bldAttend/getJsonData.cmd",
+                    data=payload,
+                )
+                if resp.status_code == 200:
+                    data = resp.json()
+                    items = data.get("OutBlock_1", [])
+                    for item in items:
+                        name = item.get("ISU_ABBRV", "").strip()
+                        code = item.get("ISU_SRT_CD", "").strip()
+                        if name and code and len(code) == 6:
+                            stock_map[name] = code
+                    logger.info("krx.loaded", market=mkt_name, count=len(items))
+                else:
+                    logger.warning("krx.http_error", market=mkt_name, status=resp.status_code)
+            except Exception as e:
+                logger.warning("krx.fetch_failed", market=mkt_name, error=str(e))
+
+    # KRX 실패 시 네이버 금융 폴백
+    if len(stock_map) < 100:
+        logger.info("krx.fallback_naver", reason="KRX returned too few stocks")
+        stock_map.update(await fetch_naver_stocks())
+
+    return stock_map
+
+
+async def fetch_naver_stocks() -> dict[str, str]:
+    """네이버 금융에서 종목 가져오기 (KRX 폴백)"""
+    stock_map = {}
+    headers = {"User-Agent": "Mozilla/5.0"}
+    import re
+
+    async with httpx.AsyncClient(timeout=15, headers=headers) as client:
+        for sosok in [0, 1]:  # 0=KOSPI, 1=KOSDAQ
+            for page in range(1, 45):
+                try:
+                    url = f"https://finance.naver.com/sise/sise_market_sum.naver?sosok={sosok}&page={page}"
+                    resp = await client.get(url)
+                    text = resp.content.decode("euc-kr", errors="ignore")
+                    rows = re.findall(
+                        r"main\.naver\?code=(\d{6})[^>]*>([^<]+)</a>", text
+                    )
+                    if not rows:
+                        break
+                    for code, name in rows:
+                        name = name.strip()
+                        if name and code:
+                            stock_map[name] = code
+                except Exception:
+                    break
+
+    logger.info("naver.loaded", count=len(stock_map))
+    return stock_map
+
+
+async def load_all_stocks() -> dict[str, str]:
+    """전체 종목 로딩 (KRX → 네이버 → 폴백)"""
+    try:
+        stock_map = await fetch_krx_stocks()
+        if len(stock_map) > 100:
+            stock_map.update(ALIASES)
+            logger.info("stocks.loaded", total=len(stock_map), source="KRX")
+            return stock_map
+    except Exception as e:
+        logger.warning("stocks.krx_failed", error=str(e))
+
+    try:
+        stock_map = await fetch_naver_stocks()
+        if len(stock_map) > 100:
+            stock_map.update({
+                "코스피": "KOSPI", "코스닥": "KOSDAQ",
+                "코스피200": "KOSPI200", "KRX": "KRX",
+            })
+            stock_map.update(ALIASES)
+            logger.info("stocks.loaded", total=len(stock_map), source="Naver")
+            return stock_map
+    except Exception as e:
+        logger.warning("stocks.naver_failed", error=str(e))
+
+    # 최종 폴백
+    fallback = {**FALLBACK_STOCKS, **ALIASES}
+    logger.warning("stocks.using_fallback", total=len(fallback))
+    return fallback
+
+
+async def auto_refresh_stocks(state_ref, interval_hours: int = 24):
+    """백그라운드에서 종목 목록 자동 갱신"""
+    while True:
+        await asyncio.sleep(interval_hours * 3600)
+        try:
+            new_map = await load_all_stocks()
+            if len(new_map) > 50:
+                state_ref.stock_map = new_map
+                logger.info("stocks.refreshed", total=len(new_map))
+        except Exception as e:
+            logger.warning("stocks.refresh_failed", error=str(e))
@@ -0,0 +1,9 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl unzip && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+RUN mkdir -p /app/data /app/logs
+EXPOSE 8888
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8888", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,51 @@
+# docker-compose.yml services: 블록 안에 추가
+
+  # ──────────────────────────────────────────
+  # DART 공시 수집기  172.30.0.17
+  # ──────────────────────────────────────────
+  dart-collector:
+    build:
+      context: ./dart-collector
+      dockerfile: Dockerfile
+    container_name: trading-dart-collector
+    restart: unless-stopped
+    ports:
+      - "8888:8888"
+    environment:
+      DART_API_KEY: "${DART_API_KEY}"
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      BAREUN_API_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      VLLM_URL: "http://vllm:8000"
+      QDRANT_URL: "http://qdrant:6333"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.17
+    depends_on:
+      redis:
+        condition: service_healthy
+      bareunaapi:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 2G
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8888/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "100m"
+        max-file: "10"
@@ -0,0 +1,11 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+redis==5.0.4
+asyncpg==0.29.0
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
+xmltodict==0.13.0
+openpyxl==3.1.2
+zipfile36==0.1.3
@@ -0,0 +1,7 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN pip install fastapi uvicorn asyncpg redis orjson httpx pydantic \
+    bcrypt "python-jose[cryptography]" email-validator apscheduler
+COPY . .
+EXPOSE 8989
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8989", "--workers", "1"]
@@ -0,0 +1,59 @@
+"""인증 유틸 - bcrypt 해시 + JWT 토큰 (이메일/비밀번호)"""
+import os
+from datetime import datetime, timedelta, timezone
+import bcrypt
+from jose import jwt, JWTError
+from fastapi import Header, HTTPException, status
+
+JWT_SECRET      = os.getenv("JWT_SECRET", "")
+JWT_ALGORITHM   = "HS256"
+JWT_EXPIRE_DAYS = int(os.getenv("JWT_EXPIRE_DAYS", "7"))
+
+if not JWT_SECRET:
+    raise RuntimeError("JWT_SECRET 환경변수가 설정되지 않았습니다")
+
+def hash_password(plain: str) -> str:
+    # bcrypt는 72바이트 제한 — 안전을 위해 잘라서 해시
+    pw_bytes = plain.encode("utf-8")[:72]
+    return bcrypt.hashpw(pw_bytes, bcrypt.gensalt()).decode("utf-8")
+
+def verify_password(plain: str, hashed: str) -> bool:
+    try:
+        pw_bytes = plain.encode("utf-8")[:72]
+        return bcrypt.checkpw(pw_bytes, hashed.encode("utf-8"))
+    except Exception:
+        return False
+
+def create_token(user_id: int, email: str) -> str:
+    now = datetime.now(timezone.utc)
+    payload = {
+        "sub": str(user_id),
+        "email": email,
+        "iat": int(now.timestamp()),
+        "exp": int((now + timedelta(days=JWT_EXPIRE_DAYS)).timestamp()),
+    }
+    return jwt.encode(payload, JWT_SECRET, algorithm=JWT_ALGORITHM)
+
+def decode_token(token: str) -> dict:
+    try:
+        return jwt.decode(token, JWT_SECRET, algorithms=[JWT_ALGORITHM])
+    except JWTError as e:
+        raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail=f"invalid token: {e}")
+
+async def current_user(authorization: str = Header(default="")) -> dict:
+    """Authorization: Bearer <token> → {"id": int, "email": str}"""
+    if not authorization.lower().startswith("bearer "):
+        raise HTTPException(status_code=401, detail="missing bearer token")
+    token = authorization.split(" ", 1)[1].strip()
+    payload = decode_token(token)
+    try:
+        return {"id": int(payload["sub"]), "email": payload.get("email", "")}
+    except (KeyError, ValueError):
+        raise HTTPException(status_code=401, detail="malformed token payload")
+
+# 타이밍 공격 방지용 더미 해시 (이메일이 없을 때도 verify를 호출해 응답시간 균일화)
+_DUMMY_HASH = bcrypt.hashpw(b"dummy_password_for_timing", bcrypt.gensalt()).decode("utf-8")
+
+def dummy_verify():
+    """존재하지 않는 사용자에 대해서도 bcrypt 비용을 동일하게 지불"""
+    bcrypt.checkpw(b"dummy_password_for_timing", _DUMMY_HASH.encode("utf-8"))
@@ -0,0 +1,292 @@
+<!doctype html>
+<html lang="ko">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width,initial-scale=1">
+<title>Trading AI · 종목 카드</title>
+<style>
+* { box-sizing: border-box; margin: 0; padding: 0; }
+body {
+  font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", "Noto Sans KR", sans-serif;
+  background: #0d1117; color: #e6edf3; line-height: 1.5;
+  padding: 16px; max-width: 1400px; margin: 0 auto;
+}
+header {
+  display: flex; justify-content: space-between; align-items: center;
+  margin-bottom: 20px; padding-bottom: 16px; border-bottom: 1px solid #30363d;
+}
+h1 { font-size: 20px; font-weight: 600; }
+.regime {
+  padding: 6px 14px; border-radius: 20px; font-size: 13px; font-weight: 500;
+}
+.regime.강세 { background: #1f3d1f; color: #4ade80; }
+.regime.약세 { background: #3d1f1f; color: #f87171; }
+.regime.중립 { background: #2d2d3a; color: #a5a5b8; }
+.summary-bar {
+  display: flex; gap: 12px; margin-bottom: 20px; flex-wrap: wrap;
+}
+.summary-pill {
+  background: #161b22; border: 1px solid #30363d; padding: 8px 14px;
+  border-radius: 6px; font-size: 13px;
+}
+.summary-pill .v { color: #58a6ff; font-weight: 600; margin-left: 6px; }
+.tabs { display: flex; gap: 8px; margin-bottom: 16px; }
+.tab {
+  padding: 8px 18px; background: #161b22; border: 1px solid #30363d;
+  border-radius: 6px; cursor: pointer; font-size: 14px;
+}
+.tab.active { background: #1f6feb; border-color: #1f6feb; color: white; }
+.cards { display: grid; grid-template-columns: repeat(auto-fill, minmax(380px, 1fr)); gap: 14px; }
+.card {
+  background: #161b22; border: 1px solid #30363d; border-radius: 10px;
+  padding: 16px; transition: transform .15s, border-color .15s;
+}
+.card:hover { transform: translateY(-2px); border-color: #58a6ff; }
+.card-head { display: flex; justify-content: space-between; align-items: flex-start; margin-bottom: 12px; }
+.name { font-size: 17px; font-weight: 600; color: #e6edf3; }
+.code { font-size: 12px; color: #7d8590; margin-top: 2px; }
+.score-badge {
+  text-align: right;
+}
+.score {
+  font-size: 28px; font-weight: 700; line-height: 1;
+}
+.score.강력매수 { color: #4ade80; }
+.score.매수관심 { color: #58a6ff; }
+.score.관망 { color: #a5a5b8; }
+.score.매도관심, .score.강력매도 { color: #f87171; }
+.rec {
+  font-size: 11px; padding: 2px 8px; border-radius: 4px;
+  margin-top: 4px; display: inline-block; font-weight: 500;
+}
+.rec.강력매수 { background: #1f3d1f; color: #4ade80; }
+.rec.매수관심 { background: #1f2d3d; color: #58a6ff; }
+.rec.매도관심, .rec.강력매도 { background: #3d1f1f; color: #f87171; }
+.price-line {
+  display: flex; align-items: baseline; gap: 8px; margin-bottom: 14px;
+}
+.price { font-size: 20px; font-weight: 600; }
+.change { font-size: 14px; font-weight: 500; }
+.change.up { color: #4ade80; }
+.change.down { color: #f87171; }
+.change.flat { color: #7d8590; }
+.sector-tag {
+  font-size: 11px; padding: 2px 8px; background: #21262d;
+  border-radius: 4px; color: #a5a5b8;
+}
+.target-grid {
+  display: grid; grid-template-columns: repeat(2, 1fr); gap: 8px;
+  background: #0d1117; padding: 10px; border-radius: 6px; margin-bottom: 12px;
+}
+.target-cell { font-size: 12px; }
+.target-cell .label { color: #7d8590; margin-bottom: 2px; }
+.target-cell .v { font-weight: 600; }
+.target-cell .pct { color: #7d8590; margin-left: 4px; font-size: 11px; }
+.target-cell .v.t { color: #4ade80; }
+.target-cell .v.s { color: #f87171; }
+.metrics {
+  display: grid; grid-template-columns: repeat(3, 1fr); gap: 6px;
+  font-size: 12px; padding: 10px 0; border-top: 1px solid #21262d; border-bottom: 1px solid #21262d;
+  margin-bottom: 10px;
+}
+.m-cell .label { color: #7d8590; font-size: 10px; margin-bottom: 2px; }
+.m-cell .v { font-weight: 500; }
+.m-cell .v.good { color: #4ade80; }
+.m-cell .v.bad { color: #f87171; }
+.position-line {
+  display: flex; justify-content: space-between; padding: 8px 10px;
+  background: #1c2839; border-radius: 6px; margin-bottom: 10px; font-size: 12px;
+}
+.pos-label { color: #7d8590; }
+.pos-val { color: #58a6ff; font-weight: 600; }
+.reasons { font-size: 12px; color: #a5a5b8; line-height: 1.6; }
+.reason-item {
+  padding: 4px 0; padding-left: 14px; position: relative;
+}
+.reason-item:before { content: "→"; position: absolute; left: 0; color: #58a6ff; }
+.empty { text-align: center; color: #7d8590; padding: 60px 20px; font-size: 14px; }
+.loader { text-align: center; padding: 40px; color: #7d8590; }
+</style>
+</head>
+<body>
+<header>
+  <h1>📊 Trading AI · 종목 카드</h1>
+  <div class="regime" id="regime">로딩…</div>
+</header>
+
+<div class="summary-bar" id="summary"></div>
+
+<div class="tabs">
+  <div class="tab active" data-kind="recommendations">🟢 매수 추천</div>
+  <div class="tab" data-kind="avoid">🔴 회피 종목</div>
+</div>
+
+<div id="cards-container">
+  <div class="loader">불러오는 중…</div>
+</div>
+
+<script>
+const API = location.origin;
+
+function fmt(n) {
+  if (n === null || n === undefined) return "-";
+  if (typeof n !== "number") return String(n);
+  return n.toLocaleString("ko-KR");
+}
+function pctClass(v) { return v > 0 ? "up" : v < 0 ? "down" : "flat"; }
+function pctSign(v) { return (v > 0 ? "+" : "") + v + "%"; }
+function metricCls(v, good) { return v >= good ? "good" : v < 0 ? "bad" : ""; }
+
+async function loadSummary() {
+  try {
+    const r = await fetch(`${API}/api/summary`);
+    const d = await r.json();
+    document.getElementById("regime").textContent =
+      `시장 ${d.market_regime} (${d.market_regime_adj > 0 ? "+" : ""}${d.market_regime_adj})`;
+    document.getElementById("regime").className = "regime " + (d.market_regime || "중립");
+
+    document.getElementById("summary").innerHTML = `
+      <div class="summary-pill">강력매수<span class="v">${d.strong_buy || 0}</span></div>
+      <div class="summary-pill">매수관심<span class="v">${d.interest_buy || 0}</span></div>
+      <div class="summary-pill">매도관심<span class="v">${d.interest_sell || 0}</span></div>
+      <div class="summary-pill">강력매도<span class="v">${d.strong_sell || 0}</span></div>
+      <div class="summary-pill">7일 호재 비율<span class="v">${d.sentiment_ratio || 0}%</span></div>
+      <div class="summary-pill">분석 종목<span class="v">${d.stocks_analyzed || 0}</span></div>
+    `;
+  } catch (e) { console.error(e); }
+}
+
+function renderCard(s) {
+  const recCls = s.recommendation || "관망";
+  const ch = s.change_pct || 0;
+  const mosCls = s.margin_of_safety > 25 ? "good" : s.margin_of_safety < -25 ? "bad" : "";
+  const eqCls = metricCls(s.earnings_quality || 0, 5);
+  const reasons = (s.reasons || []).map(r =>
+    `<div class="reason-item">${r}</div>`).join("");
+  return `
+    <div class="card">
+      <div class="card-head">
+        <div>
+          <div class="name">${s.stock_name}</div>
+          <div class="code">${s.stock_code} · <span class="sector-tag">${s.sector || "기타"}</span></div>
+        </div>
+        <div class="score-badge">
+          <div class="score ${recCls}">${s.score}</div>
+          <div class="rec ${recCls}">${s.recommendation}</div>
+        </div>
+      </div>
+
+      <div class="price-line">
+        <div class="price">${fmt(s.price)}원</div>
+        <div class="change ${pctClass(ch)}">${pctSign(ch)}</div>
+        ${s.market_cap_eok ? `<div class="sector-tag">시총 ${fmt(s.market_cap_eok)}억</div>` : ""}
+      </div>
+
+      ${s.t1 ? `
+      <div class="target-grid">
+        <div class="target-cell">
+          <div class="label">진입가</div>
+          <div class="v">${fmt(s.entry_price)}원</div>
+        </div>
+        <div class="target-cell">
+          <div class="label">손절</div>
+          <div class="v s">${fmt(s.stop_loss)}원</div>
+        </div>
+        <div class="target-cell">
+          <div class="label">T1 (50% 매도)</div>
+          <div class="v t">${fmt(s.t1)}<span class="pct">${pctSign(s.t1_pct)}</span></div>
+        </div>
+        <div class="target-cell">
+          <div class="label">Trailing (ATR×2)</div>
+          <div class="v">${fmt(s.trailing_stop)}원</div>
+        </div>
+        <div class="target-cell">
+          <div class="label">T2 (30%)</div>
+          <div class="v t">${fmt(s.t2)}<span class="pct">${pctSign(s.t2_pct)}</span></div>
+        </div>
+        <div class="target-cell">
+          <div class="label">T3 (20%)</div>
+          <div class="v t">${fmt(s.t3)}<span class="pct">${pctSign(s.t3_pct)}</span></div>
+        </div>
+      </div>` : `<div class="loader" style="padding:8px;font-size:11px;">목표가 데이터 없음 (다음 기술분석 cron에서 갱신)</div>`}
+
+      ${s.position_size_pct ? `
+      <div class="position-line">
+        <span class="pos-label">추천 매수 비중</span>
+        <span class="pos-val">${s.position_size_pct}% · 변동성 ${s.volatility_60d || "-"}%</span>
+      </div>` : ""}
+
+      <div class="metrics">
+        <div class="m-cell">
+          <div class="label">ROE</div>
+          <div class="v ${metricCls(s.roe || 0, 10)}">${s.roe ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">영업이익률</div>
+          <div class="v ${metricCls(s.operating_margin || 0, 10)}">${s.operating_margin ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">부채비율</div>
+          <div class="v ${(s.debt_ratio||100) <= 50 ? "good" : (s.debt_ratio||0) > 80 ? "bad" : ""}">${s.debt_ratio ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">PER</div>
+          <div class="v">${s.per ?? "-"}</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">안전마진</div>
+          <div class="v ${mosCls}">${s.margin_of_safety ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">이익품질</div>
+          <div class="v ${eqCls}">${s.earnings_quality ?? "-"}</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">외국인지분</div>
+          <div class="v">${s.foreign_ratio_pct ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">공매도비중</div>
+          <div class="v ${(s.short_weight_pct||0) > 5 ? "bad" : ""}">${s.short_weight_pct ?? "-"}%</div>
+        </div>
+        <div class="m-cell">
+          <div class="label">FCF</div>
+          <div class="v ${metricCls(s.fcf_ratio || 0, 5)}">${s.fcf_ratio ?? "-"}%</div>
+        </div>
+      </div>
+
+      ${reasons ? `<div class="reasons">${reasons}</div>` : ""}
+    </div>
+  `;
+}
+
+async function loadCards(kind) {
+  const c = document.getElementById("cards-container");
+  c.innerHTML = '<div class="loader">불러오는 중…</div>';
+  try {
+    const r = await fetch(`${API}/api/${kind}?days=1&limit=30`);
+    const data = await r.json();
+    if (!data.length) {
+      c.innerHTML = '<div class="empty">해당 카테고리에 종목이 없습니다.</div>';
+      return;
+    }
+    c.innerHTML = `<div class="cards">${data.map(renderCard).join("")}</div>`;
+  } catch (e) {
+    c.innerHTML = `<div class="empty">에러: ${e.message}</div>`;
+  }
+}
+
+document.querySelectorAll(".tab").forEach(t => {
+  t.addEventListener("click", () => {
+    document.querySelectorAll(".tab").forEach(x => x.classList.remove("active"));
+    t.classList.add("active");
+    loadCards(t.dataset.kind);
+  });
+});
+
+loadSummary();
+loadCards("recommendations");
+setInterval(loadSummary, 60000);
+</script>
+</body>
+</html>
@@ -0,0 +1,26 @@
+-- 사용자 인증 + 포트폴리오 (5~20명 지인 대상, JWT 기반)
+-- idempotent: 여러 번 실행해도 안전
+
+CREATE TABLE IF NOT EXISTS users (
+    id            SERIAL PRIMARY KEY,
+    email         VARCHAR(255) UNIQUE NOT NULL,
+    password_hash VARCHAR(255) NOT NULL,
+    role          VARCHAR(20)  NOT NULL DEFAULT 'user',
+    created_at    TIMESTAMPTZ  NOT NULL DEFAULT NOW(),
+    last_login_at TIMESTAMPTZ
+);
+
+CREATE TABLE IF NOT EXISTS user_portfolio (
+    id          SERIAL PRIMARY KEY,
+    user_id     INTEGER NOT NULL REFERENCES users(id) ON DELETE CASCADE,
+    stock_code  VARCHAR(10)   NOT NULL,
+    stock_name  VARCHAR(100)  NOT NULL DEFAULT '',
+    buy_price   INTEGER       NOT NULL,
+    qty         INTEGER       NOT NULL,
+    memo        TEXT          DEFAULT '',
+    created_at  TIMESTAMPTZ   NOT NULL DEFAULT NOW(),
+    updated_at  TIMESTAMPTZ   NOT NULL DEFAULT NOW()
+);
+
+CREATE INDEX IF NOT EXISTS idx_user_portfolio_user ON user_portfolio(user_id);
+CREATE INDEX IF NOT EXISTS idx_user_portfolio_user_code ON user_portfolio(user_id, stock_code);
@@ -0,0 +1,9 @@
+-- 회원가입 관리자 승인 + 보안 (계정 잠금)
+-- ADMIN_EMAILS 환경변수에 매칭되는 이메일은 register 시 자동 승인+admin
+
+ALTER TABLE users
+    ADD COLUMN IF NOT EXISTS is_approved        BOOLEAN     NOT NULL DEFAULT false,
+    ADD COLUMN IF NOT EXISTS failed_login_count INTEGER     NOT NULL DEFAULT 0,
+    ADD COLUMN IF NOT EXISTS locked_until       TIMESTAMPTZ;
+
+CREATE INDEX IF NOT EXISTS idx_users_pending ON users(is_approved) WHERE is_approved = false;
@@ -0,0 +1,177 @@
+# ============================================================
+# docker-compose.yml services: 블록 안에 추가
+# Phase 1~4 전체 서비스
+# ============================================================
+
+  # ── Phase 1: 네이버 뉴스 수집기 ──────────────────────────
+  news-collector:
+    build:
+      context: ./news-collector
+      dockerfile: Dockerfile
+    container_name: trading-news-collector
+    restart: unless-stopped
+    ports:
+      - "8787:8787"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      BAREUN_API_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      VLLM_URL: "http://vllm:8000"
+      QDRANT_URL: "http://qdrant:6333"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.16
+    depends_on:
+      redis:
+        condition: service_healthy
+      bareunaapi:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 2G
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8787/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "100m"
+        max-file: "5"
+
+  # ── Phase 2: 한국투자증권 주가 API ──────────────────────
+  kis-api:
+    build:
+      context: ./kis-api
+      dockerfile: Dockerfile
+    container_name: trading-kis-api
+    restart: unless-stopped
+    ports:
+      - "8585:8585"
+    environment:
+      KIS_APP_KEY: "${KIS_APP_KEY:-}"
+      KIS_APP_SECRET: "${KIS_APP_SECRET:-}"
+      KIS_IS_PAPER: "${KIS_IS_PAPER:-true}"
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.18
+    depends_on:
+      redis:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8585/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 3: 종목 점수 엔진 ─────────────────────────────
+  score-engine:
+    build:
+      context: ./score-engine
+      dockerfile: Dockerfile
+    container_name: trading-score-engine
+    restart: unless-stopped
+    ports:
+      - "8686:8686"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.19
+    depends_on:
+      redis:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8686/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 4: 대시보드 API ───────────────────────────────
+  dashboard-api:
+    build:
+      context: ./dashboard-api
+      dockerfile: Dockerfile
+    container_name: trading-dashboard-api
+    restart: unless-stopped
+    ports:
+      - "8989:8989"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.22
+    depends_on:
+      redis:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "true"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 15s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
@@ -0,0 +1,830 @@
+
+
+# ============================================================
+# Trading AI System
+# GPU 0: RTX 3060 12GB → Ollama (EXAONE 3.5 추론)
+# GPU 1: RTX 3070  8GB → Ollama (bge-m3 임베딩)
+# ============================================================
+
+x-logging: &default-logging
+  driver: "json-file"
+  options:
+    max-size: "${LOG_MAX_SIZE:-100m}"
+    max-file: "${LOG_MAX_FILE:-10}"
+
+x-restart: &restart
+  restart: unless-stopped
+
+networks:
+  trading-net:
+    driver: bridge
+    ipam:
+      config:
+        - subnet: "${SUBNET:-172.30.0.0/16}"
+          gateway: "${GATEWAY:-172.30.0.1}"
+
+volumes:
+  redis-data:
+  qdrant-data:
+  ollama-data:
+  n8n-data:
+  bareun-data:
+  pgdata:
+
+services:
+ 
+  # ── Phase 1: 네이버 뉴스 수집기 ──────────────────────────
+  news-collector:
+    build:
+      context: ./news-collector
+      dockerfile: Dockerfile
+    container_name: trading-news-collector
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8787:8787"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      BAREUN_API_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      QDRANT_URL: "http://qdrant:6333"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.16
+    depends_on:
+      redis:
+        condition: service_healthy
+      bareunaapi:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 2G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8787/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "100m"
+        max-file: "5"
+
+  # ── Phase 2: 한국투자증권 주가 API ──────────────────────
+  kis-api:
+    build:
+      context: ./kis-api
+      dockerfile: Dockerfile
+    container_name: trading-kis-api
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8585:8585"
+    environment:
+      KIWOOM_APP_KEY: "${KIWOOM_APP_KEY}"
+      KIWOOM_SECRET_KEY: "${KIWOOM_SECRET_KEY}"
+      KIWOOM_BASE_URL: "${KIWOOM_BASE_URL:-https://api.kiwoom.com}"
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.18
+    depends_on:
+      redis:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8585/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 2.5: 기술적 분석 엔진 ──────────────────────────
+  ta-engine:
+    build:
+      context: ./ta-engine
+      dockerfile: Dockerfile
+    container_name: trading-ta-engine
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8484:8484"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      OLLAMA_URL: "http://ollama:11434"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.23
+    depends_on:
+      redis:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8484/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 3: 종목 점수 엔진 ─────────────────────────────
+  score-engine:
+    build:
+      context: ./score-engine
+      dockerfile: Dockerfile
+    container_name: trading-score-engine
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8686:8686"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TELEGRAM_BOT_TOKEN: "${TELEGRAM_BOT_TOKEN:-}"
+      TELEGRAM_CHAT_ID: "${TELEGRAM_CHAT_ID:-}"
+      ECOS_API_KEY: "${ECOS_API_KEY:-}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.19
+    depends_on:
+      redis:
+        condition: service_healthy
+      ta-engine:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8686/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 3.7: 텔레그램 양방향 봇 ───────────────────────
+  telegram-bot:
+    build:
+      context: ./telegram-bot
+      dockerfile: Dockerfile
+    container_name: trading-telegram-bot
+    restart: unless-stopped
+    environment:
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      TELEGRAM_BOT_TOKEN: "${TELEGRAM_BOT_TOKEN}"
+      TELEGRAM_CHAT_ID: "${TELEGRAM_CHAT_ID}"
+      OLLAMA_URL: "http://ollama:11434"
+      EXAONE_MODEL: "exaone3.5:7.8b"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.26
+    depends_on:
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "1.0"
+          memory: 512M
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 3.6: 보조 시그널 (컨센서스/기관수급/매크로) ──
+  aux-signal:
+    build:
+      context: ./aux-signal
+      dockerfile: Dockerfile
+    container_name: trading-aux-signal
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8282:8282"
+    environment:
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      ECOS_API_KEY: "${ECOS_API_KEY:-}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.25
+    depends_on:
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8282/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 3.5: 미국증시 동조 시그널 ─────────────────────
+  us-market:
+    build:
+      context: ./us-market
+      dockerfile: Dockerfile
+    container_name: trading-us-market
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8383:8383"
+    environment:
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      FINNHUB_API_KEY: "${FINNHUB_API_KEY:-}"
+      ALPHAVANTAGE_API_KEY: "${ALPHAVANTAGE_API_KEY:-}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.24
+    depends_on:
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8383/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 5: 그래프 신경망 (GAT) ────────────────────────
+  graph-engine:
+    build:
+      context: ./graph-engine
+      dockerfile: Dockerfile
+    container_name: trading-graph-engine
+    restart: unless-stopped
+    init: true
+    ports:
+      - "9090:9090"
+    environment:
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      GRAPH_MODEL_DIR: "/mnt/nas/models/graph"
+      TZ: "Asia/Seoul"
+    volumes:
+      - /mnt/nas/models:/mnt/nas/models
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.27
+    depends_on:
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 4G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:9090/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 90s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ── Phase 4: 대시보드 API ───────────────────────────────
+  dashboard-api:
+    build:
+      context: ./dashboard-api
+      dockerfile: Dockerfile
+    container_name: trading-dashboard-api
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8989:8989"
+    environment:
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      JWT_SECRET: "${JWT_SECRET}"
+      JWT_EXPIRE_DAYS: "${JWT_EXPIRE_DAYS}"
+      ADMIN_EMAILS: "${ADMIN_EMAILS}"
+      TRUSTED_ORIGINS: "${TRUSTED_ORIGINS}"
+      TZ: "Asia/Seoul"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.22
+    depends_on:
+      redis:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD-SHELL", "true"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 15s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "50m"
+        max-file: "5"
+
+  # ──────────────────────────────────────────
+  # DART 공시 수집기  172.30.0.17
+  # ──────────────────────────────────────────
+  dart-collector:
+    build:
+      context: ./dart-collector
+      dockerfile: Dockerfile
+    container_name: trading-dart-collector
+    restart: unless-stopped
+    init: true
+    ports:
+      - "8888:8888"
+    environment:
+      DART_API_KEY: "${DART_API_KEY}"
+      REDIS_HOST: redis
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_PORT: "${POSTGRES_PORT}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      BAREUN_API_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      QDRANT_URL: "http://qdrant:6333"
+      TZ: "Asia/Seoul"
+      N8N_EDITOR_BASE_URL: "https://n8.kyleyang.co.kr"
+      WEBHOOK_URL: "https://n8.kyleyang.co.kr"
+      N8N_HOST: "0.0.0.0"
+      N8N_LISTEN_ADDRESS: "0.0.0.0"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.17
+    depends_on:
+      redis:
+        condition: service_healthy
+      bareunaapi:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 2G
+    healthcheck:
+      test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:8888/health')\" 2>/dev/null || exit 1"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "100m"
+        max-file: "10"
+
+  # ──────────────────────────────────────────
+  # ──────────────────────────────────────────
+  # PostgreSQL  172.30.0.9
+  # ──────────────────────────────────────────
+  postgres:
+    image: postgres:16
+    container_name: trading-postgres
+    <<: *restart
+    ports:
+      - "55432:5432"
+    environment:
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      PGDATA: /var/lib/postgresql/data
+      TZ: "Asia/Seoul"
+    volumes:
+      - pgdata:/var/lib/postgresql/data
+      - ./init-db.sql:/docker-entrypoint-initdb.d/init-db.sql:ro
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.9
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 2G
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER} -d ${POSTGRES_DB}"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 30s
+    logging: *default-logging
+
+  # Redis  172.30.0.10
+  # ──────────────────────────────────────────
+  redis:
+    image: redis:7.2-alpine
+    container_name: trading-redis
+    <<: *restart
+    command: >
+      redis-server
+      --requirepass ${REDIS_PASSWORD}
+      --maxmemory ${REDIS_MAX_MEMORY:-2gb}
+      --maxmemory-policy ${REDIS_MAXMEMORY_POLICY:-allkeys-lru}
+      --save 900 1
+      --save 300 10
+      --appendonly yes
+      --appendfsync everysec
+      --tcp-keepalive 300
+    ports:
+      - "6379:6379"
+    volumes:
+      - redis-data:/data
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.10
+    deploy:
+      resources:
+        limits:
+          cpus: "2.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD", "redis-cli", "-a", "${REDIS_PASSWORD}", "ping"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 10s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # Qdrant  172.30.0.11
+  # ──────────────────────────────────────────
+  qdrant:
+    image: qdrant/qdrant:v1.9.2
+    container_name: trading-qdrant
+    <<: *restart
+    ports:
+      - "6333:6333"
+      - "6334:6334"
+    volumes:
+      - qdrant-data:/qdrant/storage
+      - ./qdrant-config/config.yaml:/qdrant/config/production.yaml:ro
+    environment:
+      QDRANT__SERVICE__HTTP_PORT: 6333
+      QDRANT__SERVICE__GRPC_PORT: 6334
+      QDRANT__LOG_LEVEL: INFO
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.11
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 512M
+    healthcheck:
+      test: ["CMD-SHELL", "true"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
+      start_period: 10s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # Bareun 형태소 엔진  172.30.0.15
+  # ──────────────────────────────────────────
+  bareun:
+    image: bareunai/bareun:latest
+    container_name: trading-bareun
+    <<: *restart
+    ports:
+      - "5656:5656"
+      - "9902:9902"
+    volumes:
+      - bareun-data:/bareun/var
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.15
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 1500M
+        reservations:
+          memory: 512M
+    healthcheck:
+      test: ["CMD-SHELL", "curl -sf http://localhost:9902/health || exit 0"]
+      interval: 20s
+      timeout: 10s
+      retries: 5
+      start_period: 30s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # 바른 API (FastAPI 래퍼)  172.30.0.12
+  # ──────────────────────────────────────────
+  bareunaapi:
+    build:
+      context: ./bareunaapi
+      dockerfile: Dockerfile
+    container_name: trading-bareunaapi
+    <<: *restart
+    ports:
+      - "5757:5757"
+    environment:
+      BAREUN_API_KEY: "${BAREUN_API_KEY}"
+      BAREUN_SERVER_HOST: bareun
+      BAREUN_SERVER_PORT: 5656
+      REDIS_HOST: redis
+      REDIS_PORT: 6379
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      REDIS_DB: 1
+      NEWS_DEDUP_TTL: "${NEWS_DEDUP_TTL:-86400}"
+      LOG_LEVEL: INFO
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.12
+    depends_on:
+      redis:
+        condition: service_healthy
+      bareun:
+        condition: service_started
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 512M
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:5757/health"]
+      interval: 15s
+      timeout: 10s
+      retries: 5
+      start_period: 40s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # Ollama 추론+임베딩 (GPU 0+1)  172.30.0.13
+  # GPU 0: RTX 3060 12GB → EXAONE 3.5 7.8B 추론
+  # GPU 1: RTX 3070  8GB → bge-m3 임베딩
+  # ──────────────────────────────────────────
+  ollama:
+    image: ollama/ollama:latest
+    container_name: trading-ollama
+    <<: *restart
+    ports:
+      - "11434:11434"
+    volumes:
+      - ollama-data:/root/.ollama
+    environment:
+      OLLAMA_NUM_PARALLEL: "${OLLAMA_NUM_PARALLEL:-4}"
+      OLLAMA_MAX_LOADED_MODELS: "2"
+      OLLAMA_FLASH_ATTENTION: "1"
+      CUDA_VISIBLE_DEVICES: "0,1"
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.13
+    deploy:
+      resources:
+        limits:
+          cpus: "8.0"
+          memory: 24G
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0", "1"]
+              capabilities: [gpu]
+    healthcheck:
+      test: ["CMD-SHELL", "true"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
+      start_period: 10s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # n8n 워크플로우  172.30.0.20
+  # ──────────────────────────────────────────
+  n8n:
+    image: n8nio/n8n:latest
+    container_name: trading-n8n
+    <<: *restart
+    ports:
+      - "5678:5678"
+    environment:
+      N8N_BASIC_AUTH_ACTIVE: "true"
+      N8N_BASIC_AUTH_USER: "${N8N_BASIC_AUTH_USER}"
+      N8N_BASIC_AUTH_PASSWORD: "${N8N_BASIC_AUTH_PASSWORD}"
+      N8N_ENCRYPTION_KEY: "${N8N_ENCRYPTION_KEY}"
+      WEBHOOK_URL: "${N8N_WEBHOOK_URL}"
+      N8N_LOG_LEVEL: info
+      N8N_METRICS: "true"
+      # DB - NAS PostgreSQL
+      DB_TYPE: postgresdb
+      DB_POSTGRESDB_HOST: "${POSTGRES_HOST}"
+      DB_POSTGRESDB_PORT: "${POSTGRES_PORT:-5432}"
+      DB_POSTGRESDB_DATABASE: "${POSTGRES_DB}"
+      DB_POSTGRESDB_USER: "${POSTGRES_USER}"
+      DB_POSTGRESDB_PASSWORD: "${POSTGRES_PASSWORD}"
+      DB_POSTGRESDB_SCHEMA: n8n
+      DB_POSTGRESDB_POOL_SIZE: "10"
+      DB_POSTGRESDB_CONNECT_TIMEOUT: "30000"
+      # Redis Queue
+      QUEUE_BULL_REDIS_HOST: redis
+      QUEUE_BULL_REDIS_PORT: 6379
+      QUEUE_BULL_REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      QUEUE_BULL_REDIS_DB: 0
+      EXECUTIONS_MODE: queue
+      N8N_GRACEFUL_SHUTDOWN_TIMEOUT: 30
+      TZ: "${TZ:-Asia/Seoul}"
+      GENERIC_TIMEZONE: "${TZ:-Asia/Seoul}"
+      # 내부 서비스 주소 (워크플로우에서 사용)
+      BAREUNAAPI_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      QDRANT_URL: "http://qdrant:6333"
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      POSTGRES_HOST: "${POSTGRES_HOST}"
+      POSTGRES_DB: "${POSTGRES_DB}"
+      POSTGRES_USER: "${POSTGRES_USER}"
+      POSTGRES_PASSWORD: "${POSTGRES_PASSWORD}"
+      NEWS_SIMILARITY_THRESHOLD: "${NEWS_SIMILARITY_THRESHOLD:-0.92}"
+      TELEGRAM_BOT_TOKEN: "${TELEGRAM_BOT_TOKEN:-}"
+      TELEGRAM_CHAT_ID: "${TELEGRAM_CHAT_ID:-}"
+      N8N_SECURE_COOKIE: "false"
+      # 실행 이력 자동 정리 (느린 n8n 해결)
+      EXECUTIONS_DATA_PRUNE: "true"
+      EXECUTIONS_DATA_MAX_AGE: 168
+      EXECUTIONS_DATA_PRUNE_MAX_COUNT: 10000
+    volumes:
+      - n8n-data:/home/node/.n8n
+      - /mnt/nas:/mnt/nas
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.20
+    depends_on:
+      redis:
+        condition: service_healthy
+      qdrant:
+        condition: service_healthy
+      bareunaapi:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 1G
+    healthcheck:
+      test: ["CMD", "wget", "-qO-", "http://localhost:5678/healthz"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s
+    logging: *default-logging
+
+  # ──────────────────────────────────────────
+  # n8n Worker  172.30.0.21
+  # ──────────────────────────────────────────
+  n8n-worker:
+    image: n8nio/n8n:latest
+    container_name: trading-n8n-worker
+    <<: *restart
+    command: worker
+    environment:
+      N8N_ENCRYPTION_KEY: "${N8N_ENCRYPTION_KEY}"
+      DB_TYPE: postgresdb
+      DB_POSTGRESDB_HOST: "${POSTGRES_HOST}"
+      DB_POSTGRESDB_PORT: "${POSTGRES_PORT:-5432}"
+      DB_POSTGRESDB_DATABASE: "${POSTGRES_DB}"
+      DB_POSTGRESDB_USER: "${POSTGRES_USER}"
+      DB_POSTGRESDB_PASSWORD: "${POSTGRES_PASSWORD}"
+      DB_POSTGRESDB_SCHEMA: n8n
+      DB_POSTGRESDB_POOL_SIZE: "5"
+      DB_POSTGRESDB_CONNECT_TIMEOUT: "30000"
+      QUEUE_BULL_REDIS_HOST: redis
+      QUEUE_BULL_REDIS_PORT: 6379
+      QUEUE_BULL_REDIS_PASSWORD: "${REDIS_PASSWORD}"
+      QUEUE_BULL_REDIS_DB: 0
+      EXECUTIONS_MODE: queue
+      TZ: "${TZ:-Asia/Seoul}"
+      GENERIC_TIMEZONE: "${TZ:-Asia/Seoul}"
+      BAREUNAAPI_URL: "http://bareunaapi:5757"
+      OLLAMA_URL: "http://ollama:11434"
+      QDRANT_URL: "http://qdrant:6333"
+      REDIS_PASSWORD: "${REDIS_PASSWORD}"
+    volumes:
+      - n8n-data:/home/node/.n8n
+      - /mnt/nas:/mnt/nas
+    networks:
+      trading-net:
+        ipv4_address: 172.30.0.21
+    depends_on:
+      - n8n
+    deploy:
+      resources:
+        limits:
+          cpus: "4.0"
+          memory: 512M
+    logging: *default-logging
@@ -0,0 +1,11 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+# torch는 CPU 빌드만 받음 (Ollama가 GPU 점유, 추론은 CPU로 충분)
+RUN pip install --no-cache-dir --default-timeout=300 --retries=5 \
+    --index-url https://download.pytorch.org/whl/cpu torch==2.4.1
+RUN pip install --no-cache-dir --default-timeout=180 --retries=5 -r requirements.txt
+COPY . .
+EXPOSE 9090
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "9090", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,635 @@
+"""
+Graph Engine (port 9090, 172.30.0.25)
+한국 종목 그래프 신경망 (GAT) — 다음날 수익률 예측 → stock_scores.graph_score
+
+노드: 한국 활성종목 (dart_corps.is_active=true)
+피처(12): 1d/5d/20d 수익률, vol_ratio, rsi, tech_score,
+         roe, operating_margin, debt_ratio, news_7d, us_overnight, log_mcap
+엣지: ① 가격 60일 상관 |corr|>0.4   ② 같은 sector   ③ 뉴스 공기 ≥3회
+
+학습: 매주 일요일 06:00 (6mo rolling window)
+추론: 매일 08:30 → stock_scores.graph_score
+"""
+import os
+import asyncio
+import json
+import math
+from datetime import date, datetime, timedelta
+from typing import Optional, List, Tuple
+
+import asyncpg
+import structlog
+import numpy as np
+import pandas as pd
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from fastapi import FastAPI, Query, BackgroundTasks
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from apscheduler.triggers.cron import CronTrigger
+from pytz import timezone
+
+# ─────────────────────────────────────────────────────────────
+# 설정
+# ─────────────────────────────────────────────────────────────
+PG = {
+    "host": os.getenv("POSTGRES_HOST", "postgres"),
+    "port": int(os.getenv("POSTGRES_PORT", 5432)),
+    "database": os.getenv("POSTGRES_DB", "trading_ai"),
+    "user": os.getenv("POSTGRES_USER", "kyu"),
+    "password": os.getenv("POSTGRES_PASSWORD", ""),
+}
+KST = timezone("Asia/Seoul")
+MODEL_DIR = os.getenv("GRAPH_MODEL_DIR", "/mnt/nas/models/graph")
+os.makedirs(MODEL_DIR, exist_ok=True)
+MODEL_PATH = os.path.join(MODEL_DIR, "gat_latest.pt")
+
+FEATURE_DIM = 12
+HIDDEN_DIM = 32
+ATT_HEADS = 4
+DROPOUT = 0.3
+CORR_LOOKBACK = 60
+CORR_THRESHOLD = 0.65
+TOP_K_NEIGHBORS = 20      # 노드당 가격 상관 엣지 상한 (메모리/속도 캡)
+NEWS_LOOKBACK = 14
+NEWS_COOC_MIN = 3
+TRAIN_WINDOW_DAYS = 180
+SAMPLE_STRIDE_DAYS = 5    # 6개월 / 5일 = ~36 학습 샘플
+
+logger = structlog.get_logger()
+app = FastAPI(title="Graph Engine (GAT)")
+pg_pool: Optional[asyncpg.Pool] = None
+scheduler = AsyncIOScheduler(timezone=KST)
+device = torch.device("cpu")
+
+DDL = """
+ALTER TABLE stock_scores
+  ADD COLUMN IF NOT EXISTS graph_score DOUBLE PRECISION;
+CREATE TABLE IF NOT EXISTS graph_model_meta (
+  model_date    DATE PRIMARY KEY,
+  train_samples INT,
+  val_samples   INT,
+  val_loss      DOUBLE PRECISION,
+  edge_count    INT,
+  node_count    INT,
+  hidden_dim    INT,
+  notes         TEXT,
+  created_at    TIMESTAMP DEFAULT NOW()
+);
+CREATE TABLE IF NOT EXISTS graph_predictions (
+  stock_code   VARCHAR(10),
+  predict_date DATE,
+  pred_return  DOUBLE PRECISION,
+  created_at   TIMESTAMP DEFAULT NOW(),
+  PRIMARY KEY (stock_code, predict_date)
+);
+"""
+
+
+# ─────────────────────────────────────────────────────────────
+# GAT 모델 (torch-geometric 없이 순수 torch)
+# ─────────────────────────────────────────────────────────────
+class GATLayer(nn.Module):
+    def __init__(self, in_dim, out_dim, heads, dropout):
+        super().__init__()
+        assert out_dim % heads == 0
+        self.heads = heads
+        self.head_dim = out_dim // heads
+        self.W = nn.Linear(in_dim, out_dim, bias=False)
+        a_src = torch.empty(1, heads, self.head_dim)
+        a_dst = torch.empty(1, heads, self.head_dim)
+        nn.init.xavier_uniform_(a_src)
+        nn.init.xavier_uniform_(a_dst)
+        self.a_src = nn.Parameter(a_src.squeeze(0))
+        self.a_dst = nn.Parameter(a_dst.squeeze(0))
+        nn.init.xavier_uniform_(self.W.weight)
+        self.dropout = dropout
+
+    def forward(self, x, edge_index, edge_weight=None):
+        N = x.size(0)
+        h = self.W(x).view(N, self.heads, self.head_dim)
+        src, dst = edge_index[0], edge_index[1]
+        # 어텐션 logits per (edge, head)
+        e_src = (h[src] * self.a_src.unsqueeze(0)).sum(-1)
+        e_dst = (h[dst] * self.a_dst.unsqueeze(0)).sum(-1)
+        e = F.leaky_relu(e_src + e_dst, 0.2)
+        if edge_weight is not None:
+            e = e + edge_weight.unsqueeze(-1).log().clamp(min=-10)
+        # dst별 softmax: max-subtract for stability
+        e_max = torch.full((N, self.heads), -1e9, device=x.device)
+        e_max = e_max.scatter_reduce(0, dst.unsqueeze(-1).expand(-1, self.heads),
+                                      e, reduce="amax", include_self=True)
+        e_exp = torch.exp(e - e_max[dst])
+        denom = torch.zeros(N, self.heads, device=x.device).index_add_(0, dst, e_exp)
+        alpha = e_exp / (denom[dst] + 1e-12)
+        alpha = F.dropout(alpha, p=self.dropout, training=self.training)
+        # 메시지 집계
+        m = h[src] * alpha.unsqueeze(-1)
+        out = torch.zeros(N, self.heads, self.head_dim, device=x.device)
+        out = out.index_add_(0, dst, m)
+        return out.reshape(N, -1)
+
+
+class GraphNet(nn.Module):
+    def __init__(self, in_dim=FEATURE_DIM, hidden=HIDDEN_DIM,
+                 heads=ATT_HEADS, dropout=DROPOUT):
+        super().__init__()
+        self.gat1 = GATLayer(in_dim, hidden, heads, dropout)
+        self.gat2 = GATLayer(hidden, hidden, heads, dropout)
+        self.head = nn.Linear(hidden, 1)
+        self.dropout = dropout
+
+    def forward(self, x, edge_index, edge_weight=None):
+        x = F.elu(self.gat1(x, edge_index, edge_weight))
+        x = F.dropout(x, p=self.dropout, training=self.training)
+        x = F.elu(self.gat2(x, edge_index, edge_weight))
+        return self.head(x).squeeze(-1)
+
+
+# ─────────────────────────────────────────────────────────────
+# 데이터 로더
+# ─────────────────────────────────────────────────────────────
+async def load_active_codes(conn) -> List[str]:
+    rows = await conn.fetch(
+        "SELECT stock_code FROM dart_corps WHERE is_active=true ORDER BY stock_code")
+    return [r["stock_code"] for r in rows]
+
+
+async def load_features(conn, codes: List[str], target_date: date) -> np.ndarray:
+    """노드 피처 행렬 (N, 12) — 결측은 0으로."""
+    N = len(codes)
+    F_ = FEATURE_DIM
+    out = np.zeros((N, F_), dtype=np.float32)
+    idx = {c: i for i, c in enumerate(codes)}
+
+    # ── 가격 모멘텀 (1d/5d/20d 수익률), vol_ratio ──
+    rows = await conn.fetch("""
+        SELECT stock_code, dt, close_price, volume
+        FROM stock_ohlcv
+        WHERE dt BETWEEN $1 AND $2
+        ORDER BY stock_code, dt
+    """, target_date - timedelta(days=35), target_date)
+    df = pd.DataFrame(rows, columns=["stock_code", "dt", "close", "volume"])
+    if not df.empty:
+        df["close"] = df["close"].astype(float)
+        df["volume"] = df["volume"].astype(float)
+        for code, g in df.groupby("stock_code"):
+            if code not in idx:
+                continue
+            g = g.sort_values("dt")
+            closes = g["close"].values
+            vols = g["volume"].values
+            if len(closes) >= 2:
+                out[idx[code], 0] = (closes[-1] / closes[-2] - 1) * 100
+            if len(closes) >= 6:
+                out[idx[code], 1] = (closes[-1] / closes[-6] - 1) * 100
+            if len(closes) >= 21:
+                out[idx[code], 2] = (closes[-1] / closes[-21] - 1) * 100
+            if len(vols) >= 20:
+                v20 = vols[-20:].mean()
+                out[idx[code], 3] = (vols[-1] / v20) if v20 > 0 else 1.0
+
+    # ── 기술적 (RSI, tech_score) ──
+    rows = await conn.fetch("""
+        SELECT DISTINCT ON (stock_code) stock_code, rsi, tech_score
+        FROM stock_technical
+        WHERE analyzed_at::date <= $1
+        ORDER BY stock_code, analyzed_at DESC
+    """, target_date)
+    for r in rows:
+        if r["stock_code"] in idx:
+            out[idx[r["stock_code"]], 4] = float(r["rsi"] or 50)
+            out[idx[r["stock_code"]], 5] = float(r["tech_score"] or 0)
+
+    # ── 펀더멘털 (ROE, op_margin, debt_ratio) ──
+    rows = await conn.fetch("""
+        SELECT DISTINCT ON (stock_code) stock_code, roe, operating_margin, debt_ratio
+        FROM dart_financials
+        WHERE reprt_code='11011'
+        ORDER BY stock_code, bsns_year DESC
+    """)
+    for r in rows:
+        if r["stock_code"] in idx:
+            out[idx[r["stock_code"]], 6] = float(r["roe"] or 0)
+            out[idx[r["stock_code"]], 7] = float(r["operating_margin"] or 0)
+            out[idx[r["stock_code"]], 8] = float(r["debt_ratio"] or 0)
+
+    # ── 뉴스 감성 7일 ──
+    rows = await conn.fetch("""
+        SELECT primary_stock,
+               SUM(CASE sentiment WHEN '호재' THEN intensity
+                                  WHEN '악재' THEN -intensity ELSE 0 END)::float AS s
+        FROM news_analysis
+        WHERE analyzed_at BETWEEN $1 AND $2
+          AND primary_stock IS NOT NULL
+        GROUP BY primary_stock
+    """, target_date - timedelta(days=7), target_date + timedelta(days=1))
+    for r in rows:
+        if r["primary_stock"] in idx:
+            out[idx[r["primary_stock"]], 9] = float(r["s"])
+
+    # ── us_overnight_adj (latest) ──
+    rows = await conn.fetch("""
+        SELECT DISTINCT ON (stock_code) stock_code, us_overnight_adj
+        FROM stock_scores
+        WHERE score_date <= $1 AND us_overnight_adj IS NOT NULL
+        ORDER BY stock_code, score_date DESC
+    """, target_date)
+    for r in rows:
+        if r["stock_code"] in idx:
+            out[idx[r["stock_code"]], 10] = float(r["us_overnight_adj"])
+
+    # ── log market cap ──
+    rows = await conn.fetch("""
+        SELECT DISTINCT ON (stock_code) stock_code, market_cap
+        FROM stock_prices
+        WHERE market_cap IS NOT NULL
+        ORDER BY stock_code, collected_at DESC
+    """)
+    for r in rows:
+        if r["stock_code"] in idx and r["market_cap"]:
+            out[idx[r["stock_code"]], 11] = math.log10(float(r["market_cap"]) + 1)
+
+    # 표준화 (피처별 z-score, 클리핑)
+    mu = out.mean(axis=0)
+    sd = out.std(axis=0) + 1e-6
+    out = np.clip((out - mu) / sd, -5, 5)
+    return out.astype(np.float32)
+
+
+async def load_price_corr_edges(conn, codes: List[str], target_date: date,
+                                 lookback: int = CORR_LOOKBACK,
+                                 threshold: float = CORR_THRESHOLD
+                                 ) -> Tuple[np.ndarray, np.ndarray]:
+    """가격 상관 엣지. 반환: edge_index (2, E), edge_weight (E,)."""
+    idx = {c: i for i, c in enumerate(codes)}
+    rows = await conn.fetch("""
+        SELECT stock_code, dt, close_price FROM stock_ohlcv
+        WHERE dt BETWEEN $1 AND $2 AND stock_code = ANY($3)
+        ORDER BY stock_code, dt
+    """, target_date - timedelta(days=int(lookback * 1.6)),
+        target_date, codes)
+    if not rows:
+        return np.zeros((2, 0), dtype=np.int64), np.zeros(0, dtype=np.float32)
+    df = pd.DataFrame(rows, columns=["code", "dt", "close"])
+    pivot = df.pivot(index="dt", columns="code", values="close").astype(float)
+    pivot = pivot.tail(lookback)
+    rets = pivot.pct_change().dropna(how="all")
+    # 결측치 많은 종목 제거 (절반 이상 결측)
+    rets = rets.dropna(axis=1, thresh=int(len(rets) * 0.5))
+    if rets.shape[1] < 2:
+        return np.zeros((2, 0), dtype=np.int64), np.zeros(0, dtype=np.float32)
+    rets = rets.fillna(0)
+    M = rets.values  # (T, K)
+    M = (M - M.mean(axis=0)) / (M.std(axis=0) + 1e-9)
+    K = M.shape[1]
+    corr = (M.T @ M) / max(M.shape[0] - 1, 1)
+    np.fill_diagonal(corr, 0)
+    abs_corr = np.abs(corr)
+    # 임계값 필터 + 노드당 top-K 이웃만 유지 (메모리 캡)
+    mask = abs_corr > threshold
+    src_set: List[int] = []
+    dst_set: List[int] = []
+    w_set: List[float] = []
+    code_arr = list(rets.columns)
+    for i in range(K):
+        cands = np.where(mask[i])[0]
+        if len(cands) == 0:
+            continue
+        if len(cands) > TOP_K_NEIGHBORS:
+            top_idx = np.argpartition(-abs_corr[i, cands], TOP_K_NEIGHBORS)[:TOP_K_NEIGHBORS]
+            cands = cands[top_idx]
+        for j in cands:
+            src_set.append(idx[code_arr[i]])
+            dst_set.append(idx[code_arr[j]])
+            w_set.append(float(abs_corr[i, j]))
+    if not src_set:
+        return np.zeros((2, 0), dtype=np.int64), np.zeros(0, dtype=np.float32)
+    return (np.stack([np.array(src_set, dtype=np.int64),
+                       np.array(dst_set, dtype=np.int64)]),
+            np.array(w_set, dtype=np.float32))
+
+
+async def load_sector_edges(conn, codes: List[str]) -> np.ndarray:
+    """같은 sector 노드 간 양방향 엣지."""
+    idx = {c: i for i, c in enumerate(codes)}
+    rows = await conn.fetch("""
+        SELECT stock_code, sector FROM dart_corps
+        WHERE is_active=true AND sector IS NOT NULL AND stock_code = ANY($1)
+    """, codes)
+    by_sec = {}
+    for r in rows:
+        by_sec.setdefault(r["sector"], []).append(r["stock_code"])
+    src, dst = [], []
+    for sec, lst in by_sec.items():
+        # 섹터당 종목 수가 100개 초과면 비대해지므로 무작위 샘플
+        if len(lst) > 80:
+            import random
+            random.seed(hash(sec) & 0xffff)
+            lst = random.sample(lst, 80)
+        for i, a in enumerate(lst):
+            for b in lst[i+1:]:
+                if a in idx and b in idx:
+                    src += [idx[a], idx[b]]
+                    dst += [idx[b], idx[a]]
+    if not src:
+        return np.zeros((2, 0), dtype=np.int64)
+    return np.stack([np.array(src, dtype=np.int64), np.array(dst, dtype=np.int64)])
+
+
+async def load_news_edges(conn, codes: List[str], target_date: date,
+                           lookback_days: int = NEWS_LOOKBACK,
+                           min_count: int = NEWS_COOC_MIN) -> np.ndarray:
+    """뉴스 공기관계: 같은 뉴스의 affected_stocks 페어 카운트 ≥ min_count."""
+    idx = {c: i for i, c in enumerate(codes)}
+    rows = await conn.fetch("""
+        SELECT affected_stocks FROM news_analysis
+        WHERE analyzed_at BETWEEN $1 AND $2
+          AND jsonb_array_length(affected_stocks) > 1
+    """, target_date - timedelta(days=lookback_days),
+        target_date + timedelta(days=1))
+    from collections import Counter
+    counter: Counter = Counter()
+    for r in rows:
+        codes_in_news = r["affected_stocks"]
+        if isinstance(codes_in_news, str):
+            codes_in_news = json.loads(codes_in_news)
+        codes_in_news = [c for c in codes_in_news if c in idx]
+        for i, a in enumerate(codes_in_news):
+            for b in codes_in_news[i+1:]:
+                key = (a, b) if a < b else (b, a)
+                counter[key] += 1
+    src, dst = [], []
+    for (a, b), cnt in counter.items():
+        if cnt >= min_count:
+            src += [idx[a], idx[b]]
+            dst += [idx[b], idx[a]]
+    if not src:
+        return np.zeros((2, 0), dtype=np.int64)
+    return np.stack([np.array(src, dtype=np.int64), np.array(dst, dtype=np.int64)])
+
+
+async def build_graph(conn, target_date: date):
+    """전체 그래프 구성. 반환: (codes, x, edge_index, edge_weight)."""
+    codes = await load_active_codes(conn)
+    x = await load_features(conn, codes, target_date)
+    ei_corr, ew_corr = await load_price_corr_edges(conn, codes, target_date)
+    ei_sec = await load_sector_edges(conn, codes)
+    ei_news = await load_news_edges(conn, codes, target_date)
+    # 결합 (sector/news는 weight=1.0)
+    ei_all = np.concatenate([ei_corr, ei_sec, ei_news], axis=1)
+    ew_all = np.concatenate([
+        ew_corr,
+        np.full(ei_sec.shape[1], 0.5, dtype=np.float32),
+        np.full(ei_news.shape[1], 0.7, dtype=np.float32),
+    ])
+    logger.info("graph.built", nodes=len(codes), edges=int(ei_all.shape[1]),
+                price=int(ei_corr.shape[1]), sector=int(ei_sec.shape[1]),
+                news=int(ei_news.shape[1]))
+    return codes, x, ei_all.astype(np.int64), ew_all.astype(np.float32)
+
+
+async def load_labels(conn, codes: List[str], target_date: date) -> np.ndarray:
+    """target_date 종가 → target_date+1 영업일 종가의 변화율 (%)."""
+    idx = {c: i for i, c in enumerate(codes)}
+    labels = np.zeros(len(codes), dtype=np.float32)
+    mask = np.zeros(len(codes), dtype=bool)
+    rows = await conn.fetch("""
+        SELECT stock_code, dt, close_price FROM stock_ohlcv
+        WHERE dt BETWEEN $1 AND $2 AND stock_code = ANY($3)
+        ORDER BY stock_code, dt
+    """, target_date, target_date + timedelta(days=7), codes)
+    by_code = {}
+    for r in rows:
+        by_code.setdefault(r["stock_code"], []).append(
+            (r["dt"], float(r["close_price"])))
+    for code, lst in by_code.items():
+        lst.sort()
+        if code in idx and len(lst) >= 2:
+            t0, c0 = lst[0]
+            t1, c1 = lst[1]
+            if c0 > 0:
+                labels[idx[code]] = (c1 / c0 - 1) * 100
+                mask[idx[code]] = True
+    return labels, mask
+
+
+# ─────────────────────────────────────────────────────────────
+# 학습
+# ─────────────────────────────────────────────────────────────
+async def train_model(window_days: int = TRAIN_WINDOW_DAYS,
+                       stride: int = SAMPLE_STRIDE_DAYS,
+                       epochs: int = 30, lr: float = 1e-3):
+    """rolling 윈도우 학습. 일별이 아닌 stride 간격으로 샘플링."""
+    today = date.today()
+    sample_dates = [today - timedelta(days=window_days - i)
+                    for i in range(0, window_days, stride)]
+    sample_dates = [d for d in sample_dates if d.weekday() < 5]
+    if len(sample_dates) < 8:
+        return {"err": "too few samples", "samples": len(sample_dates)}
+
+    val_split = max(1, len(sample_dates) // 5)
+    train_dates = sample_dates[:-val_split]
+    val_dates = sample_dates[-val_split:]
+    logger.info("train.split", train=len(train_dates), val=len(val_dates))
+
+    async with pg_pool.acquire() as conn:
+        # 모든 샘플 그래프 미리 구성
+        samples = []
+        for d in sample_dates:
+            try:
+                codes, x, ei, ew = await build_graph(conn, d)
+                y, m = await load_labels(conn, codes, d)
+                if m.sum() < 10:
+                    continue
+                samples.append({
+                    "date": d,
+                    "x": torch.tensor(x),
+                    "ei": torch.tensor(ei),
+                    "ew": torch.tensor(ew),
+                    "y": torch.tensor(y),
+                    "m": torch.tensor(m),
+                })
+            except Exception as e:
+                logger.warning("train.sample_err", date=str(d), err=str(e))
+
+    if len(samples) < 4:
+        return {"err": "no valid samples", "got": len(samples)}
+
+    val_count = max(1, len(samples) // 5)
+    train_set = samples[:-val_count]
+    val_set = samples[-val_count:]
+
+    model = GraphNet().to(device)
+    opt = torch.optim.Adam(model.parameters(), lr=lr, weight_decay=1e-5)
+    best_val = float("inf")
+
+    for epoch in range(epochs):
+        model.train()
+        tot = 0.0
+        for s in train_set:
+            opt.zero_grad()
+            pred = model(s["x"].to(device), s["ei"].to(device),
+                         s["ew"].to(device))
+            mask = s["m"].to(device)
+            loss = F.huber_loss(pred[mask], s["y"].to(device)[mask], delta=2.0)
+            loss.backward()
+            torch.nn.utils.clip_grad_norm_(model.parameters(), 5.0)
+            opt.step()
+            tot += float(loss)
+        tot /= len(train_set)
+
+        model.eval()
+        v = 0.0
+        with torch.no_grad():
+            for s in val_set:
+                pred = model(s["x"].to(device), s["ei"].to(device),
+                             s["ew"].to(device))
+                mask = s["m"].to(device)
+                v += float(F.huber_loss(pred[mask], s["y"].to(device)[mask],
+                                         delta=2.0))
+        v /= len(val_set)
+        logger.info("train.epoch", epoch=epoch, train_loss=tot, val_loss=v)
+        if v < best_val:
+            best_val = v
+            torch.save({"state": model.state_dict(),
+                        "feature_dim": FEATURE_DIM,
+                        "hidden": HIDDEN_DIM,
+                        "heads": ATT_HEADS}, MODEL_PATH)
+
+    async with pg_pool.acquire() as conn:
+        await conn.execute("""
+            INSERT INTO graph_model_meta
+              (model_date, train_samples, val_samples, val_loss,
+               edge_count, node_count, hidden_dim, notes)
+            VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
+            ON CONFLICT (model_date) DO UPDATE
+              SET train_samples=$2, val_samples=$3, val_loss=$4,
+                  edge_count=$5, node_count=$6, hidden_dim=$7, notes=$8
+        """, today, len(train_set), len(val_set), best_val,
+            int(samples[-1]["ei"].shape[1]), int(samples[-1]["x"].shape[0]),
+            HIDDEN_DIM, f"epochs={epochs} lr={lr} stride={stride}")
+    return {"trained": True, "train_samples": len(train_set),
+            "val_samples": len(val_set), "best_val_loss": best_val}
+
+
+# ─────────────────────────────────────────────────────────────
+# 추론
+# ─────────────────────────────────────────────────────────────
+async def predict_today() -> dict:
+    if not os.path.exists(MODEL_PATH):
+        return {"err": "no trained model"}
+    today = date.today()
+    async with pg_pool.acquire() as conn:
+        codes, x, ei, ew = await build_graph(conn, today)
+    model = GraphNet().to(device)
+    ckpt = torch.load(MODEL_PATH, map_location=device, weights_only=True)
+    model.load_state_dict(ckpt["state"])
+    model.eval()
+    with torch.no_grad():
+        pred = model(torch.tensor(x).to(device),
+                     torch.tensor(ei).to(device),
+                     torch.tensor(ew).to(device))
+    pred_np = pred.cpu().numpy()
+    # graph_predictions만 저장 — score-engine이 calculate_daily_scores 시 읽어 ensemble에 적용.
+    async with pg_pool.acquire() as conn:
+        async with conn.transaction():
+            for i, code in enumerate(codes):
+                p = float(pred_np[i])
+                await conn.execute("""
+                    INSERT INTO graph_predictions (stock_code, predict_date, pred_return)
+                    VALUES ($1, $2, $3)
+                    ON CONFLICT (stock_code, predict_date) DO UPDATE
+                      SET pred_return=$3, created_at=NOW()
+                """, code, today, p)
+    return {"predicted": len(codes), "date": str(today)}
+
+
+# ─────────────────────────────────────────────────────────────
+# FastAPI
+# ─────────────────────────────────────────────────────────────
+@app.on_event("startup")
+async def on_start():
+    global pg_pool
+    pg_pool = await asyncpg.create_pool(**PG, min_size=2, max_size=5)
+    async with pg_pool.acquire() as conn:
+        await conn.execute(DDL)
+    # 16:25 KST: score-engine 16:30 calculate_daily_scores 직전 추론
+    scheduler.add_job(predict_today, CronTrigger(hour=16, minute=25,
+                                                  day_of_week="mon-fri"),
+                      id="graph_predict", replace_existing=True)
+    scheduler.add_job(train_model, CronTrigger(day_of_week="sun", hour=6,
+                                                minute=0),
+                      id="graph_train", replace_existing=True)
+    scheduler.start()
+    logger.info("graph-engine.started")
+
+
+@app.on_event("shutdown")
+async def on_stop():
+    if scheduler.running:
+        scheduler.shutdown()
+    if pg_pool:
+        await pg_pool.close()
+
+
+@app.get("/health")
+async def health():
+    return {"status": "ok",
+            "model_exists": os.path.exists(MODEL_PATH)}
+
+
+@app.post("/graph/build")
+async def manual_build(target: Optional[str] = None):
+    d = date.fromisoformat(target) if target else date.today()
+    async with pg_pool.acquire() as conn:
+        codes, x, ei, ew = await build_graph(conn, d)
+    return {"nodes": len(codes), "edges": int(ei.shape[1]),
+            "feature_shape": list(x.shape), "date": str(d)}
+
+
+@app.post("/train")
+async def manual_train(epochs: int = Query(default=30, ge=1, le=200),
+                        window_days: int = Query(default=180, ge=30, le=365),
+                        stride: int = Query(default=5, ge=1, le=30),
+                        bg: BackgroundTasks = None):
+    if bg:
+        bg.add_task(train_model, window_days, stride, epochs)
+        return {"status": "queued", "epochs": epochs}
+    return await train_model(window_days, stride, epochs)
+
+
+@app.post("/predict")
+async def manual_predict():
+    return await predict_today()
+
+
+@app.get("/predict/{code}")
+async def predict_stock(code: str, limit: int = Query(default=30, ge=1, le=180)):
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT predict_date, pred_return
+            FROM graph_predictions
+            WHERE stock_code=$1
+            ORDER BY predict_date DESC LIMIT $2
+        """, code, limit)
+    return {"code": code, "history": [dict(r) for r in rows]}
+
+
+@app.get("/status")
+async def status():
+    async with pg_pool.acquire() as conn:
+        meta = await conn.fetchrow("""
+            SELECT * FROM graph_model_meta ORDER BY model_date DESC LIMIT 1
+        """)
+        latest_pred = await conn.fetchrow("""
+            SELECT predict_date, COUNT(*) AS n
+            FROM graph_predictions
+            GROUP BY predict_date
+            ORDER BY predict_date DESC LIMIT 1
+        """)
+    return {
+        "model": dict(meta) if meta else None,
+        "latest_prediction": dict(latest_pred) if latest_pred else None,
+        "model_exists": os.path.exists(MODEL_PATH),
+    }
@@ -0,0 +1,12 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+asyncpg==0.29.0
+apscheduler==3.10.4
+structlog==24.2.0
+orjson==3.10.3
+httpx==0.27.0
+pandas==2.2.2
+numpy==1.26.4
+scipy==1.13.1
+pytz==2024.1
+torch==2.4.1
@@ -0,0 +1,150 @@
+-- ============================================================
+-- Trading AI - PostgreSQL 테이블 초기화
+-- NAS (192.168.0.36:32770) 에서 실행
+-- ============================================================
+
+-- 뉴스 분석 결과 테이블
+CREATE TABLE IF NOT EXISTS news_analysis (
+    id              SERIAL PRIMARY KEY,
+    title           TEXT NOT NULL,
+    url             TEXT DEFAULT '',
+    source          VARCHAR(100) DEFAULT '',
+    published_at    TIMESTAMP WITH TIME ZONE,
+    hash            VARCHAR(16) UNIQUE NOT NULL,
+    
+    -- AI 분석 결과
+    sentiment       VARCHAR(10) DEFAULT '중립',     -- 호재/악재/중립
+    intensity       INTEGER DEFAULT 0,              -- 1~5
+    primary_stock   VARCHAR(10) DEFAULT '',          -- 주요 종목코드
+    affected_stocks JSONB DEFAULT '[]'::jsonb,       -- 영향 종목 목록
+    reason          TEXT DEFAULT '',                  -- 판단 근거
+    investment_action VARCHAR(20) DEFAULT '관망',     -- 매수관심/매도관심/관망
+    
+    -- 형태소 분석 결과
+    keywords        JSONB DEFAULT '[]'::jsonb,
+    stock_names     JSONB DEFAULT '[]'::jsonb,
+    stock_codes     JSONB DEFAULT '[]'::jsonb,
+    
+    -- 메타데이터
+    similar_count   INTEGER DEFAULT 0,
+    analyzed_at     TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
+    created_at      TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+);
+
+-- 인덱스
+CREATE INDEX IF NOT EXISTS idx_news_hash ON news_analysis(hash);
+CREATE INDEX IF NOT EXISTS idx_news_sentiment ON news_analysis(sentiment);
+CREATE INDEX IF NOT EXISTS idx_news_primary_stock ON news_analysis(primary_stock);
+CREATE INDEX IF NOT EXISTS idx_news_published_at ON news_analysis(published_at DESC);
+CREATE INDEX IF NOT EXISTS idx_news_analyzed_at ON news_analysis(analyzed_at DESC);
+CREATE INDEX IF NOT EXISTS idx_news_intensity ON news_analysis(intensity DESC);
+CREATE INDEX IF NOT EXISTS idx_news_investment_action ON news_analysis(investment_action);
+
+-- 종목별 빠른 조회용 GIN 인덱스
+CREATE INDEX IF NOT EXISTS idx_news_affected_stocks ON news_analysis USING GIN (affected_stocks);
+CREATE INDEX IF NOT EXISTS idx_news_stock_codes ON news_analysis USING GIN (stock_codes);
+
+-- 유용한 뷰: 최근 호재/악재 요약
+CREATE OR REPLACE VIEW v_recent_signals AS
+SELECT 
+    sentiment,
+    intensity,
+    title,
+    primary_stock,
+    affected_stocks,
+    investment_action,
+    reason,
+    published_at,
+    analyzed_at
+FROM news_analysis
+WHERE sentiment IN ('호재', '악재')
+  AND analyzed_at >= NOW() - INTERVAL '24 hours'
+ORDER BY intensity DESC, analyzed_at DESC;
+
+-- 기술적 분석 결과 테이블
+CREATE TABLE IF NOT EXISTS stock_technical (
+    id          SERIAL PRIMARY KEY,
+    stock_code  VARCHAR(10) UNIQUE NOT NULL,
+    stock_name  VARCHAR(100) DEFAULT '',
+    price       INTEGER DEFAULT 0,
+    ma5         FLOAT DEFAULT 0,
+    ma20        FLOAT DEFAULT 0,
+    ma60        FLOAT DEFAULT 0,
+    ma120       FLOAT DEFAULT 0,
+    rsi         FLOAT DEFAULT 50,
+    macd        FLOAT DEFAULT 0,
+    macd_signal FLOAT DEFAULT 0,
+    macd_hist   FLOAT DEFAULT 0,
+    bb_upper    FLOAT DEFAULT 0,
+    bb_mid      FLOAT DEFAULT 0,
+    bb_lower    FLOAT DEFAULT 0,
+    pct_b       FLOAT DEFAULT 0.5,
+    stoch_k     FLOAT DEFAULT 50,
+    stoch_d     FLOAT DEFAULT 50,
+    vol_ratio   FLOAT DEFAULT 1,
+    tech_score  FLOAT DEFAULT 0,
+    signal      VARCHAR(10) DEFAULT '관망',
+    signals     JSONB DEFAULT '[]'::jsonb,
+    targets     JSONB DEFAULT '{}'::jsonb,
+    analyzed_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
+);
+
+CREATE INDEX IF NOT EXISTS idx_ta_score  ON stock_technical(tech_score DESC);
+CREATE INDEX IF NOT EXISTS idx_ta_signal ON stock_technical(signal);
+CREATE INDEX IF NOT EXISTS idx_ta_code   ON stock_technical(stock_code);
+
+-- 유용한 뷰: 종목별 뉴스 카운트
+CREATE OR REPLACE VIEW v_stock_news_count AS
+SELECT 
+    primary_stock,
+    sentiment,
+    COUNT(*) as count,
+    AVG(intensity) as avg_intensity,
+    MAX(analyzed_at) as latest_at
+FROM news_analysis
+WHERE primary_stock != ''
+  AND analyzed_at >= NOW() - INTERVAL '7 days'
+GROUP BY primary_stock, sentiment
+ORDER BY count DESC;
+
+-- DART 기업 정보 테이블
+CREATE TABLE IF NOT EXISTS dart_corps (
+    stock_code  VARCHAR(10)  PRIMARY KEY,
+    corp_code   VARCHAR(20)  NOT NULL,
+    corp_name   VARCHAR(200) NOT NULL,
+    modify_date VARCHAR(20)  DEFAULT '',
+    is_active   BOOLEAN      DEFAULT false
+);
+
+CREATE INDEX IF NOT EXISTS idx_dart_corps_active   ON dart_corps(is_active);
+CREATE INDEX IF NOT EXISTS idx_dart_corps_corp_code ON dart_corps(corp_code);
+
+-- DART 재무제표 테이블 (버핏 가치지표 포함)
+CREATE TABLE IF NOT EXISTS dart_financials (
+    id                 SERIAL PRIMARY KEY,
+    stock_code         VARCHAR(10)  NOT NULL,
+    corp_code          VARCHAR(20),
+    corp_name          VARCHAR(200),
+    bsns_year          VARCHAR(4)   NOT NULL,
+    reprt_code         VARCHAR(10)  NOT NULL,
+    reprt_name         VARCHAR(50),
+    revenue            BIGINT       DEFAULT 0,
+    operating_profit   BIGINT       DEFAULT 0,
+    net_income         BIGINT       DEFAULT 0,
+    total_assets       BIGINT       DEFAULT 0,
+    total_liabilities  BIGINT       DEFAULT 0,
+    total_equity       BIGINT       DEFAULT 0,
+    operating_cashflow BIGINT       DEFAULT 0,
+    roe                FLOAT        DEFAULT 0,
+    operating_margin   FLOAT        DEFAULT 0,
+    net_margin         FLOAT        DEFAULT 0,
+    debt_ratio         FLOAT        DEFAULT 0,
+    revenue_growth     FLOAT        DEFAULT 0,
+    fcf_ratio          FLOAT        DEFAULT 0,
+    collected_at       TIMESTAMP    DEFAULT NOW(),
+    UNIQUE(stock_code, bsns_year, reprt_code)
+);
+
+CREATE INDEX IF NOT EXISTS idx_fin_stock ON dart_financials(stock_code);
+CREATE INDEX IF NOT EXISTS idx_fin_year  ON dart_financials(bsns_year DESC);
+CREATE INDEX IF NOT EXISTS idx_fin_roe   ON dart_financials(roe DESC);
@@ -0,0 +1,235 @@
+"""
+주가 수집 + 매매 시그널 서비스 (네이버 금융 기반)
+- API 키 불필요
+- 시총 상위 200개 종목 현재가
+- 매수/매도 목표가 + 손절가 자동 계산
+"""
+import asyncio,json,os,re
+from datetime import datetime,timedelta
+from typing import Optional
+import asyncpg,httpx,redis.asyncio as aioredis,structlog
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from fastapi import FastAPI,Query
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+
+structlog.configure(processors=[structlog.processors.TimeStamper(fmt="iso"),structlog.processors.add_log_level,structlog.processors.JSONRenderer()])
+logger=structlog.get_logger()
+REDIS_HOST=os.getenv("REDIS_HOST","redis")
+REDIS_PASSWORD=os.getenv("REDIS_PASSWORD","")
+PG_HOST=os.getenv("POSTGRES_HOST","postgres")
+PG_PORT=int(os.getenv("POSTGRES_PORT","5432"))
+PG_DB=os.getenv("POSTGRES_DB","trading_ai")
+PG_USER=os.getenv("POSTGRES_USER","kyu")
+PG_PASS=os.getenv("POSTGRES_PASSWORD","7895123")
+HEADERS={"User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"}
+
+pg_pool:Optional[asyncpg.Pool]=None
+redis_cl:Optional[aioredis.Redis]=None
+scheduler=AsyncIOScheduler(timezone="Asia/Seoul")
+
+class Stats:
+    collected=0;last_run="";errors=0
+stats=Stats()
+
+async def init_db():
+    async with pg_pool.acquire() as c:
+        await c.execute("""CREATE TABLE IF NOT EXISTS stock_prices(
+            id SERIAL PRIMARY KEY,stock_code VARCHAR(10) NOT NULL,stock_name VARCHAR(100) DEFAULT '',
+            price INTEGER DEFAULT 0,change_pct FLOAT DEFAULT 0,change_amount INTEGER DEFAULT 0,
+            volume BIGINT DEFAULT 0,high INTEGER DEFAULT 0,low INTEGER DEFAULT 0,open_price INTEGER DEFAULT 0,
+            market_cap BIGINT DEFAULT 0,per FLOAT DEFAULT 0,pbr FLOAT DEFAULT 0,
+            high_52w INTEGER DEFAULT 0,low_52w INTEGER DEFAULT 0,collected_at TIMESTAMP DEFAULT NOW())""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_sp_code ON stock_prices(stock_code)")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_sp_time ON stock_prices(collected_at DESC)")
+        await c.execute("""CREATE TABLE IF NOT EXISTS trade_signals(
+            id SERIAL PRIMARY KEY,stock_code VARCHAR(10) NOT NULL,stock_name VARCHAR(100) DEFAULT '',
+            signal_type VARCHAR(10) NOT NULL,current_price INTEGER DEFAULT 0,target_price INTEGER DEFAULT 0,
+            stop_loss INTEGER DEFAULT 0,expected_return_pct FLOAT DEFAULT 0,risk_reward_ratio FLOAT DEFAULT 0,
+            confidence FLOAT DEFAULT 0,reason TEXT DEFAULT '',news_score FLOAT DEFAULT 0,
+            dart_score FLOAT DEFAULT 0,price_momentum FLOAT DEFAULT 0,created_at TIMESTAMP DEFAULT NOW())""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ts_code ON trade_signals(stock_code)")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ts_time ON trade_signals(created_at DESC)")
+    logger.info("db.ok")
+
+async def naver_price(client,code):
+    try:
+        r=await client.get(f"https://finance.naver.com/item/main.naver?code={code}",headers=HEADERS,timeout=15)
+        r.encoding="euc-kr";h=r.text
+        def ex(p,d="0"):
+            m=re.search(p,h);return m.group(1).replace(",","").strip() if m else d
+        name=ex(r'class="wrap_company".*?<h2><a[^>]*>([^<]+)</a>',"")
+        price=int(ex(r'<p class="no_today">.*?<em.*?<span class="blind">([0-9,]+)</span>'))
+        cpct=ex(r'class="blind">([0-9.\-+]+)%',"0")
+        vol=ex(r'<td class="first">.*?<span class="blind">([0-9,]+)</span>')
+        hi=ex(r'최고.*?<em.*?<span class="blind">([0-9,]+)</span>')
+        lo=ex(r'최저.*?<em.*?<span class="blind">([0-9,]+)</span>')
+        cap=ex(r'시가총액.*?<em>([0-9,]+)</em>',"0")
+        per=ex(r'PER.*?<em>([0-9.]+)</em>',"0")
+        pbr=ex(r'PBR.*?<em>([0-9.]+)</em>',"0")
+        h52=ex(r'52주.*?최고.*?([0-9,]+)',"0")
+        l52=ex(r'52주.*?최저.*?([0-9,]+)',"0")
+        return {"code":code,"name":name,"price":price,"change_pct":float(cpct),
+            "volume":int(vol),"high":int(hi),"low":int(lo),"market_cap":int(cap),
+            "per":float(per),"pbr":float(pbr),"high_52w":int(h52),"low_52w":int(l52),
+            "timestamp":datetime.now().isoformat()}
+    except Exception as e:
+        return None
+
+async def get_stock_codes(client,count=200):
+    if pg_pool:
+        try:
+            rows=await pg_pool.fetch("SELECT stock_code FROM dart_corps LIMIT $1",count)
+            codes=[r["stock_code"] for r in rows if r["stock_code"]];
+            if len(codes)>50:return codes[:count]
+        except:pass
+    codes=[]
+    for sosok in [0,1]:
+        for page in range(1,20):
+            try:
+                r=await client.get(f"https://finance.naver.com/sise/sise_market_sum.naver?sosok={sosok}&page={page}",headers=HEADERS,timeout=15)
+                r.encoding="euc-kr"
+                found=re.findall(r'main\.naver\?code=(\d{6})',r.text)
+                if not found:break
+                codes.extend(found);await asyncio.sleep(0.2)
+            except:break
+            if len(codes)>=count:break
+    return list(dict.fromkeys(codes))[:count]
+
+def calc_signal(p,news_sc,dart_sc):
+    price=p["price"]
+    if price<=0:return None
+    h52=p.get("high_52w",0);l52=p.get("low_52w",0);cpct=p.get("change_pct",0)
+    total=news_sc*0.4+dart_sc*0.3+(cpct*10)*0.3
+    total=max(-100,min(100,total))
+    pos=(price-l52)/(h52-l52) if h52>l52>0 else 0.5
+
+    if total>=30:
+        sig="매수"
+        tp=int(price+(h52-price)*(0.3+min(total,100)/200)) if h52>price else int(price*1.1)
+        sl=int(price*(0.92+pos*0.03))
+        conf=min(95,50+total*0.3+(1-pos)*20)
+    elif total<=-30:
+        sig="매도"
+        tp=int(price-(price-l52)*(0.3+min(abs(total),100)/200)) if l52>0 and l52<price else int(price*0.9)
+        sl=int(price*1.05)
+        conf=min(95,50+abs(total)*0.3+pos*20)
+    else:return None
+
+    er=((tp-price)/price)*100 if sig=="매수" else ((price-tp)/price)*100
+    risk=abs(sl-price);reward=abs(tp-price)
+    rr=reward/risk if risk>0 else 0
+    return {"signal_type":sig,"current_price":price,"target_price":tp,"stop_loss":sl,
+        "expected_return_pct":round(er,2),"risk_reward_ratio":round(rr,2),
+        "confidence":round(conf,1),"price_momentum":round(cpct,2),
+        "news_score":round(news_sc,1),"dart_score":round(dart_sc,1)}
+
+async def collect_prices():
+    logger.info("prices.start")
+    async with httpx.AsyncClient() as client:
+        codes=await get_stock_codes(client,200)
+        ok=0
+        for code in codes:
+            p=await naver_price(client,code)
+            if not p or p["price"]<=0:continue
+            if redis_cl:await redis_cl.set(f"price:{code}",json.dumps(p,ensure_ascii=False),ex=600)
+            if pg_pool:
+                try:
+                    await pg_pool.execute("""INSERT INTO stock_prices(stock_code,stock_name,price,change_pct,
+                        volume,high,low,open_price,market_cap,per,pbr,high_52w,low_52w,collected_at)
+                        VALUES($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14)""",
+                        code,p["name"],p["price"],p["change_pct"],p["volume"],p["high"],p["low"],
+                        0,p["market_cap"],p["per"],p["pbr"],p["high_52w"],p["low_52w"],datetime.now())
+                except:pass
+            ok+=1;await asyncio.sleep(0.3)
+    await gen_signals()
+    stats.collected+=ok;stats.last_run=datetime.now().isoformat()
+    if redis_cl:await redis_cl.set("prices:last_update",datetime.now().isoformat())
+    logger.info("prices.done",count=ok)
+
+async def gen_signals():
+    if not pg_pool:return
+    async with pg_pool.acquire() as c:
+        scores=await c.fetch("""SELECT stock_code,stock_name,news_score,dart_score,total_score
+            FROM stock_scores WHERE score_date=(SELECT MAX(score_date) FROM stock_scores)
+            AND (total_score>=30 OR total_score<=-30)""")
+        n=0
+        for row in scores:
+            code=row["stock_code"]
+            if not code or len(code)!=6:continue
+            if redis_cl:
+                cached=await redis_cl.get(f"price:{code}")
+                if not cached:continue
+                pd=json.loads(cached)
+            else:continue
+            sig=calc_signal(pd,row["news_score"],row["dart_score"])
+            if not sig:continue
+            reasons=await c.fetch("SELECT reason FROM news_analysis WHERE primary_stock=$1 AND intensity>=2 ORDER BY analyzed_at DESC LIMIT 2",code)
+            reason=" | ".join([r["reason"][:100] for r in reasons])
+            await c.execute("""INSERT INTO trade_signals(stock_code,stock_name,signal_type,current_price,
+                target_price,stop_loss,expected_return_pct,risk_reward_ratio,confidence,reason,
+                news_score,dart_score,price_momentum) VALUES($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13)""",
+                code,row["stock_name"],sig["signal_type"],sig["current_price"],sig["target_price"],
+                sig["stop_loss"],sig["expected_return_pct"],sig["risk_reward_ratio"],sig["confidence"],
+                reason,sig["news_score"],sig["dart_score"],sig["price_momentum"])
+            n+=1
+    logger.info("signals.done",count=n)
+
+app=FastAPI(title="주가+시그널")
+app.add_middleware(CORSMiddleware,allow_origins=["*"],allow_methods=["*"],allow_headers=["*"])
+
+@app.on_event("startup")
+async def startup():
+    global pg_pool,redis_cl
+    pg_pool=await asyncpg.create_pool(host=PG_HOST,port=PG_PORT,database=PG_DB,user=PG_USER,password=PG_PASS,min_size=2,max_size=5)
+    redis_cl=aioredis.Redis(host=REDIS_HOST,port=6379,password=REDIS_PASSWORD,db=3,decode_responses=True)
+    await init_db()
+    scheduler.add_job(collect_prices,"cron",day_of_week="mon-fri",hour="9-16",minute="*/5",id="prices",replace_existing=True)
+    scheduler.start();logger.info("started")
+
+@app.on_event("shutdown")
+async def shutdown():
+    scheduler.shutdown()
+    if pg_pool:await pg_pool.close()
+    if redis_cl:await redis_cl.aclose()
+
+@app.get("/health")
+async def health():
+    lu=await redis_cl.get("prices:last_update") if redis_cl else ""
+    return JSONResponse(content={"status":"ok","collected":stats.collected,"last_run":stats.last_run,"last_update":lu or ""})
+
+@app.get("/price/{code}")
+async def price(code:str):
+    if redis_cl:
+        c=await redis_cl.get(f"price:{code}")
+        if c:return JSONResponse(content=json.loads(c))
+    async with httpx.AsyncClient() as cl:
+        p=await naver_price(cl,code)
+        if p:return JSONResponse(content=p)
+    return JSONResponse(content={"error":"not found"},status_code=404)
+
+@app.get("/prices")
+async def prices(limit:int=Query(default=50)):
+    if not redis_cl:return JSONResponse(content={"data":{}})
+    keys=await redis_cl.keys("price:*")
+    r={};
+    for k in keys[:limit]:
+        c=await redis_cl.get(k)
+        if c:d=json.loads(c);r[d["code"]]=d
+    return JSONResponse(content={"count":len(r),"data":r})
+
+@app.get("/signals")
+async def signals(days:int=Query(default=7)):
+    async with pg_pool.acquire() as c:
+        rows=await c.fetch("SELECT * FROM trade_signals WHERE created_at>=NOW()-INTERVAL '%s days' ORDER BY confidence DESC LIMIT 30"%days)
+    return [dict(r) for r in rows]
+
+@app.get("/signals/{code}")
+async def stock_signals(code:str):
+    async with pg_pool.acquire() as c:
+        rows=await c.fetch("SELECT * FROM trade_signals WHERE stock_code=$1 ORDER BY created_at DESC LIMIT 10",code)
+    return [dict(r) for r in rows]
+
+@app.post("/collect")
+async def manual():
+    asyncio.create_task(collect_prices());return {"status":"started"}
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8585
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8585", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,902 @@
+"""
+키움증권 REST API 기반 주가·수급·공매도 수집 서비스
+- ka10001: 현재가 + 재무지표 (PER/PBR/ROE/EPS/BPS/외국인비중/시가총액)
+- ka10005: 일봉 OHLCV + 외국인·기관 순매수
+- ka10008: 외국인 종목별 매매동향 (일자별 보유비중 변화)
+- ka10014: 공매도 추이 (잔고수량·거래비중)
+- Redis db=3 price:{code} 형식 유지 (score-engine·ta-engine 호환)
+"""
+import asyncio, json, os, re
+from datetime import datetime, timedelta
+from typing import Optional
+import asyncpg, httpx, redis.asyncio as aioredis, structlog
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from fastapi import FastAPI, Query
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+REDIS_HOST     = os.getenv("REDIS_HOST", "redis")
+REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", "")
+PG_HOST        = os.getenv("POSTGRES_HOST", "postgres")
+PG_PORT        = int(os.getenv("POSTGRES_PORT", "5432"))
+PG_DB          = os.getenv("POSTGRES_DB", "trading_ai")
+PG_USER        = os.getenv("POSTGRES_USER", "kyu")
+PG_PASS        = os.getenv("POSTGRES_PASSWORD", "")
+KIWOOM_APP_KEY    = os.getenv("KIWOOM_APP_KEY", "")
+KIWOOM_SECRET_KEY = os.getenv("KIWOOM_SECRET_KEY", "")
+KIWOOM_BASE_URL   = os.getenv("KIWOOM_BASE_URL", "https://api.kiwoom.com")
+
+pg_pool: Optional[asyncpg.Pool] = None
+redis_cl: Optional[aioredis.Redis] = None
+scheduler = AsyncIOScheduler(timezone="Asia/Seoul")
+
+# ── 키움 토큰 관리 ────────────────────────────────────────
+
+class KiwoomToken:
+    token: str = ""
+    expires_at: datetime = datetime.min
+
+    async def get(self, client: httpx.AsyncClient) -> str:
+        if self.token and datetime.now() < self.expires_at:
+            return self.token
+        resp = await client.post(
+            f"{KIWOOM_BASE_URL}/oauth2/token",
+            json={"grant_type": "client_credentials",
+                  "appkey": KIWOOM_APP_KEY,
+                  "secretkey": KIWOOM_SECRET_KEY},
+            headers={"Content-Type": "application/json;charset=UTF-8"},
+            timeout=10)
+        data = resp.json()
+        if data.get("return_code", -1) != 0:
+            raise RuntimeError(f"토큰 발급 실패: {data.get('return_msg')}")
+        self.token = data["token"]
+        # expires_dt: "20260508151325" 형식
+        exp_str = data.get("expires_dt", "")
+        try:
+            self.expires_at = datetime.strptime(exp_str, "%Y%m%d%H%M%S") - timedelta(minutes=5)
+        except Exception:
+            self.expires_at = datetime.now() + timedelta(hours=23)
+        logger.info("kiwoom.token.refreshed", expires=exp_str)
+        return self.token
+
+kiwoom_token = KiwoomToken()
+
+# ── 키움 API 호출 헬퍼 ────────────────────────────────────
+
+async def kiwoom_post(client: httpx.AsyncClient, endpoint: str, api_id: str,
+                      body: dict, cont_yn: str = "N", next_key: str = "",
+                      return_headers: bool = False):
+    token = await kiwoom_token.get(client)
+    headers = {
+        "Content-Type": "application/json;charset=UTF-8",
+        "Authorization": f"Bearer {token}",
+        "api-id": api_id,
+        "cont-yn": cont_yn,
+    }
+    if next_key:
+        headers["next-key"] = next_key
+    r = await client.post(f"{KIWOOM_BASE_URL}{endpoint}",
+                          headers=headers, json=body, timeout=15)
+    if return_headers:                       # 연속조회용 cont-yn/next-key 응답헤더 필요
+        return r.json(), r.headers
+    return r.json()
+
+def to_int(v: str) -> int:
+    try:
+        return int(str(v).replace(",", "").lstrip("+").replace(" ", "") or 0)
+    except:
+        return 0
+
+def to_float(v: str) -> float:
+    try:
+        return float(str(v).replace(",", "").lstrip("+").replace(" ", "") or 0)
+    except:
+        return 0.0
+
+# ── DB 초기화 ─────────────────────────────────────────────
+
+async def init_db():
+    async with pg_pool.acquire() as c:
+        await c.execute("""
+            CREATE TABLE IF NOT EXISTS stock_prices (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10) NOT NULL,
+                stock_name VARCHAR(100) DEFAULT '',
+                price INTEGER DEFAULT 0,
+                change_pct FLOAT DEFAULT 0,
+                change_amount INTEGER DEFAULT 0,
+                volume BIGINT DEFAULT 0,
+                high INTEGER DEFAULT 0,
+                low INTEGER DEFAULT 0,
+                open_price INTEGER DEFAULT 0,
+                market_cap BIGINT DEFAULT 0,
+                per FLOAT DEFAULT 0,
+                pbr FLOAT DEFAULT 0,
+                eps FLOAT DEFAULT 0,
+                bps FLOAT DEFAULT 0,
+                roe FLOAT DEFAULT 0,
+                ev FLOAT DEFAULT 0,
+                high_52w INTEGER DEFAULT 0,
+                low_52w INTEGER DEFAULT 0,
+                foreign_ratio FLOAT DEFAULT 0,
+                credit_ratio FLOAT DEFAULT 0,
+                float_shares BIGINT DEFAULT 0,
+                collected_at TIMESTAMP DEFAULT NOW()
+            )""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_sp_code ON stock_prices(stock_code)")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_sp_time ON stock_prices(collected_at DESC)")
+
+        await c.execute("""
+            CREATE TABLE IF NOT EXISTS stock_ohlcv (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10) NOT NULL,
+                dt DATE NOT NULL,
+                open_price INTEGER DEFAULT 0,
+                high_price INTEGER DEFAULT 0,
+                low_price INTEGER DEFAULT 0,
+                close_price INTEGER DEFAULT 0,
+                volume BIGINT DEFAULT 0,
+                trade_amount BIGINT DEFAULT 0,
+                foreign_ratio FLOAT DEFAULT 0,
+                foreign_net BIGINT DEFAULT 0,
+                institution_net BIGINT DEFAULT 0,
+                individual_net BIGINT DEFAULT 0,
+                created_at TIMESTAMP DEFAULT NOW(),
+                UNIQUE(stock_code, dt)
+            )""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ohlcv_code_dt ON stock_ohlcv(stock_code, dt DESC)")
+
+        await c.execute("""
+            CREATE TABLE IF NOT EXISTS stock_foreign_flow (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10) NOT NULL,
+                dt DATE NOT NULL,
+                close_price INTEGER DEFAULT 0,
+                change_qty BIGINT DEFAULT 0,
+                hold_qty BIGINT DEFAULT 0,
+                hold_ratio FLOAT DEFAULT 0,
+                limit_ratio FLOAT DEFAULT 0,
+                created_at TIMESTAMP DEFAULT NOW(),
+                UNIQUE(stock_code, dt)
+            )""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ff_code_dt ON stock_foreign_flow(stock_code, dt DESC)")
+
+        await c.execute("""
+            CREATE TABLE IF NOT EXISTS stock_short_sale (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10) NOT NULL,
+                dt DATE NOT NULL,
+                close_price INTEGER DEFAULT 0,
+                short_qty BIGINT DEFAULT 0,
+                short_balance_qty BIGINT DEFAULT 0,
+                trade_weight FLOAT DEFAULT 0,
+                short_avg_price INTEGER DEFAULT 0,
+                created_at TIMESTAMP DEFAULT NOW(),
+                UNIQUE(stock_code, dt)
+            )""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ss_code_dt ON stock_short_sale(stock_code, dt DESC)")
+
+        await c.execute("""
+            CREATE TABLE IF NOT EXISTS trade_signals (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10) NOT NULL,
+                stock_name VARCHAR(100) DEFAULT '',
+                signal_type VARCHAR(10) NOT NULL,
+                current_price INTEGER DEFAULT 0,
+                target_price INTEGER DEFAULT 0,
+                stop_loss INTEGER DEFAULT 0,
+                expected_return_pct FLOAT DEFAULT 0,
+                risk_reward_ratio FLOAT DEFAULT 0,
+                confidence FLOAT DEFAULT 0,
+                reason TEXT DEFAULT '',
+                news_score FLOAT DEFAULT 0,
+                dart_score FLOAT DEFAULT 0,
+                price_momentum FLOAT DEFAULT 0,
+                foreign_net_5d BIGINT DEFAULT 0,
+                short_weight FLOAT DEFAULT 0,
+                created_at TIMESTAMP DEFAULT NOW()
+            )""")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ts_code ON trade_signals(stock_code)")
+        await c.execute("CREATE INDEX IF NOT EXISTS idx_ts_time ON trade_signals(created_at DESC)")
+    logger.info("kiwoom.db.initialized")
+
+# ── 수집 함수 ─────────────────────────────────────────────
+
+async def fetch_basic_info(client: httpx.AsyncClient, code: str, sem: asyncio.Semaphore) -> Optional[dict]:
+    """ka10001: 현재가 + 재무지표 (PER/PBR/ROE/EPS/BPS/외국인비중/시가총액)"""
+    async with sem:
+        try:
+            d = await kiwoom_post(client, "/api/dostk/stkinfo", "ka10001", {"stk_cd": code})
+            if d.get("return_code", -1) != 0:
+                return None
+            return {
+                "code": code,
+                "name": d.get("stk_nm", ""),
+                "price": abs(to_int(d.get("cur_prc", "0"))),
+                "change_amount": to_int(d.get("pred_pre", "0")),
+                "change_pct": to_float(d.get("flu_rt", "0")),
+                "volume": to_int(d.get("trde_qty", "0")),
+                "open_price": abs(to_int(d.get("open_pric", "0"))),
+                "high": abs(to_int(d.get("high_pric", "0"))),
+                "low": abs(to_int(d.get("low_pric", "0"))),
+                "market_cap": to_int(d.get("mac", "0")) * 100_000_000,  # 억 → 원
+                "per": to_float(d.get("per", "0")),
+                "pbr": to_float(d.get("pbr", "0")),
+                "eps": to_float(d.get("eps", "0")),
+                "bps": to_float(d.get("bps", "0")),
+                "roe": to_float(d.get("roe", "0")),
+                "ev": to_float(d.get("ev", "0")),
+                "high_52w": abs(to_int(d.get("250hgst", "0"))),
+                "low_52w": abs(to_int(d.get("250lwst", "0"))),
+                "foreign_ratio": to_float(d.get("for_exh_rt", "0")),
+                "credit_ratio": to_float(d.get("crd_rt", "0")),
+                "float_shares": to_int(d.get("flo_stk", "0")),
+                "sale_amt": to_int(d.get("sale_amt", "0")),
+                "operating_profit": to_int(d.get("bus_pro", "0")),
+                "net_income": to_int(d.get("cup_nga", "0")),
+                "timestamp": datetime.now().isoformat(),
+            }
+        except Exception as e:
+            logger.debug("ka10001.err", code=code, error=str(e))
+            return None
+
+async def fetch_ohlcv(client: httpx.AsyncClient, code: str, sem: asyncio.Semaphore, days: int = 365) -> list:
+    """ka10005: 일봉 OHLCV + 외국인·기관 순매수"""
+    async with sem:
+        try:
+            today = datetime.now().strftime("%Y%m%d")
+            result = []
+            cont_yn, next_key = "N", ""
+            for _ in range(20):                       # 안전 상한(20p ≈ 365봉 충분)
+                d, hdr = await kiwoom_post(
+                    client, "/api/dostk/mrkcond", "ka10005",
+                    {"stk_cd": code, "dt": today, "req_cnt": days},
+                    cont_yn=cont_yn, next_key=next_key, return_headers=True)
+                for r in d.get("stk_ddwkmm", []):
+                    cp = abs(to_int(r.get("close_pric", "0")))
+                    if cp <= 0:
+                        continue
+                    result.append({
+                        "dt": r.get("date", ""),
+                        "open": abs(to_int(r.get("open_pric", "0"))),
+                        "high": abs(to_int(r.get("high_pric", "0"))),
+                        "low": abs(to_int(r.get("low_pric", "0"))),
+                        "close": cp,
+                        "volume": to_int(r.get("trde_qty", "0")),
+                        "trade_amount": to_int(r.get("trde_prica", "0")),
+                        "foreign_ratio": to_float(r.get("for_poss", "0")),
+                        "foreign_net": to_int(r.get("for_netprps", "0")),
+                        "institution_net": to_int(r.get("orgn_netprps", "0")),
+                        "individual_net": to_int(r.get("ind_netprps", "0")),
+                    })
+                # 키움 연속조회: 응답 cont-yn=Y + next-key 있으면 다음 페이지
+                if (len(result) >= days or hdr.get("cont-yn") != "Y"
+                        or not hdr.get("next-key")):
+                    break
+                cont_yn, next_key = "Y", hdr.get("next-key", "")
+            return result
+        except Exception as e:
+            logger.debug("ka10005.err", code=code, error=str(e))
+            return []
+
+async def fetch_foreign_flow(client: httpx.AsyncClient, code: str, sem: asyncio.Semaphore) -> list:
+    """ka10008: 외국인 종목별 매매동향"""
+    async with sem:
+        try:
+            d = await kiwoom_post(client, "/api/dostk/frgnistt", "ka10008", {"stk_cd": code})
+            rows = d.get("stk_frgnr", [])
+            result = []
+            for r in rows:
+                result.append({
+                    "dt": r.get("dt", ""),
+                    "close_price": abs(to_int(r.get("close_pric", "0"))),
+                    "change_qty": to_int(r.get("chg_qty", "0")),
+                    "hold_qty": to_int(r.get("poss_stkcnt", "0")),
+                    "hold_ratio": to_float(r.get("wght", "0")),
+                    "limit_ratio": to_float(r.get("limit_exh_rt", "0")),
+                })
+            return result
+        except Exception as e:
+            logger.debug("ka10008.err", code=code, error=str(e))
+            return []
+
+async def fetch_minute_chart(client: httpx.AsyncClient, code: str, tic_scope: str = "1") -> list:
+    """ka10080: 주식분봉차트조회. tic_scope: 1, 3, 5, 10, 15, 30, 45, 60 분"""
+    try:
+        d = await kiwoom_post(client, "/api/dostk/chart", "ka10080",
+            {"stk_cd": code, "tic_scope": tic_scope, "upd_stkpc_tp": "1"})
+        if d.get("return_code", -1) != 0:
+            return []
+        rows = d.get("stk_min_pole_chart_qry", []) or d.get("stk_min_pole", []) or []
+        result = []
+        for r in rows:
+            result.append({
+                "dt":     r.get("cntr_tm",  r.get("dt", "")),
+                "open":   abs(to_int(r.get("open_pric", "0"))),
+                "high":   abs(to_int(r.get("high_pric", "0"))),
+                "low":    abs(to_int(r.get("low_pric",  "0"))),
+                "close":  abs(to_int(r.get("cur_prc",   r.get("close_pric", "0")))),
+                "volume": to_int(r.get("trde_qty",  "0")),
+            })
+        return result
+    except Exception as e:
+        logger.debug("ka10080.err", code=code, error=str(e))
+        return []
+
+async def fetch_orderbook(client: httpx.AsyncClient, code: str) -> dict:
+    """ka10004: 호가잔량 (10단계 매수/매도). path=/api/dostk/mrkcond"""
+    try:
+        d = await kiwoom_post(client, "/api/dostk/mrkcond", "ka10004", {"stk_cd": code})
+        if d.get("return_code", -1) != 0:
+            return {}
+        bid, ask = [], []
+        # 매도 호가 1~10 (sel_1th_pre_bid = 가격, sel_1th_pre_req = 잔량)
+        for i in range(1, 11):
+            p = abs(to_int(d.get(f"sel_{i}th_pre_bid", "0")))
+            q = to_int(d.get(f"sel_{i}th_pre_req", "0"))
+            if p > 0:
+                ask.append({"price": p, "qty": q})
+        # 매수 호가 1~10
+        for i in range(1, 11):
+            p = abs(to_int(d.get(f"buy_{i}th_pre_bid", "0")))
+            q = to_int(d.get(f"buy_{i}th_pre_req", "0"))
+            if p > 0:
+                bid.append({"price": p, "qty": q})
+        return {
+            "code": code,
+            "ask": ask,            # 매도 (가격 낮은 것부터)
+            "bid": bid,            # 매수 (가격 높은 것부터)
+            "ask_total": to_int(d.get("tot_sel_req", "0")),
+            "bid_total": to_int(d.get("tot_buy_req", "0")),
+            "base_time": d.get("bid_req_base_tm", ""),
+            "timestamp": datetime.now().isoformat(),
+        }
+    except Exception as e:
+        logger.debug("ka10004.err", code=code, error=str(e))
+        return {}
+
+async def fetch_volume_surge() -> list:
+    """ka10023: 거래량급증 종목. path=/api/dostk/rkinfo"""
+    try:
+        async with httpx.AsyncClient() as c:
+            d = await kiwoom_post(c, "/api/dostk/rkinfo", "ka10023",
+                {"mrkt_tp": "000",     # 000=전체, 001=코스피, 101=코스닥
+                 "sort_tp": "1",        # 1=급증량, 2=급증률
+                 "tm_tp":   "2",        # 1=분, 2=전일
+                 "trde_qty_tp": "5",   # 1=5천주, 2=1만주, 5=5만주
+                 "tm": "",
+                 "stk_cnd": "0",        # 0=전체
+                 "pric_tp": "0",        # 0=전체
+                 "stex_tp": "3"})        # 3=통합
+            if d.get("return_code", -1) != 0:
+                return []
+            rows = d.get("trde_qty_sdnin", []) or []
+            result = []
+            for r in rows[:50]:
+                raw_code = r.get("stk_cd", "")
+                # NXT 등 접미사 제거 ('_AL', '_NX' 등)
+                clean_code = raw_code.split("_")[0] if raw_code else ""
+                result.append({
+                    "code":       clean_code,
+                    "name":       r.get("stk_nm", ""),
+                    "price":      abs(to_int(r.get("cur_prc", "0"))),
+                    "change_pct": to_float(r.get("flu_rt", "0")),
+                    "volume":     to_int(r.get("now_trde_qty", r.get("trde_qty", "0"))),
+                    "prev_volume":to_int(r.get("pred_trde_qty", "0")),
+                    "surge_rate": to_float(r.get("sdnin_rt", r.get("sdnin_qty", "0"))),
+                })
+            return result
+    except Exception as e:
+        logger.debug("ka10023.err", error=str(e))
+        return []
+
+async def fetch_short_sale(client: httpx.AsyncClient, code: str, sem: asyncio.Semaphore) -> list:
+    """ka10014: 공매도 추이"""
+    async with sem:
+        try:
+            today = datetime.now().strftime("%Y%m%d")
+            month_ago = (datetime.now() - timedelta(days=30)).strftime("%Y%m%d")
+            d = await kiwoom_post(client, "/api/dostk/shsa", "ka10014",
+                                  {"stk_cd": code, "strt_dt": month_ago, "end_dt": today})
+            rows = d.get("shrts_trnsn", [])
+            result = []
+            for r in rows:
+                result.append({
+                    "dt": r.get("dt", ""),
+                    "close_price": abs(to_int(r.get("close_pric", "0"))),
+                    "short_qty": to_int(r.get("shrts_qty", "0")),
+                    "short_balance_qty": to_int(r.get("ovr_shrts_qty", "0")),
+                    "trade_weight": to_float(r.get("trde_wght", "0")),
+                    "short_avg_price": to_int(r.get("shrts_avg_pric", "0")),
+                })
+            return result
+        except Exception as e:
+            logger.debug("ka10014.err", code=code, error=str(e))
+            return []
+
+# ── 종목 코드 로드 ────────────────────────────────────────
+
+async def get_stock_codes(limit: int = 0) -> list:
+    """is_active 종목 전체 (limit=0이면 제한 없음)"""
+    if pg_pool:
+        try:
+            if limit > 0:
+                rows = await pg_pool.fetch(
+                    "SELECT stock_code FROM dart_corps WHERE is_active=TRUE ORDER BY stock_code LIMIT $1", limit)
+            else:
+                rows = await pg_pool.fetch(
+                    "SELECT stock_code FROM dart_corps WHERE is_active=TRUE ORDER BY stock_code")
+            codes = [r["stock_code"] for r in rows if r["stock_code"]]
+            if len(codes) >= 50:
+                return codes
+        except Exception:
+            pass
+    return []
+
+# ── 저장 함수 ─────────────────────────────────────────────
+
+async def save_price(info: dict):
+    if redis_cl:
+        redis_data = {
+            "code": info["code"], "name": info["name"],
+            "price": info["price"], "change_pct": info["change_pct"],
+            "change_amount": info["change_amount"],
+            "volume": info["volume"], "high": info["high"], "low": info["low"],
+            "open_price": info["open_price"],
+            "market_cap": info["market_cap"],
+            "per": info["per"], "pbr": info["pbr"],
+            "eps": info["eps"], "bps": info["bps"],
+            "roe": info["roe"], "ev": info["ev"],
+            "high_52w": info["high_52w"], "low_52w": info["low_52w"],
+            "foreign_ratio": info["foreign_ratio"],
+            "credit_ratio": info["credit_ratio"],
+            "timestamp": info["timestamp"],
+        }
+        await redis_cl.set(f"price:{info['code']}", json.dumps(redis_data, ensure_ascii=False), ex=600)
+
+    if pg_pool:
+        try:
+            async with pg_pool.acquire() as c:
+                await c.execute("""
+                    INSERT INTO stock_prices (
+                        stock_code, stock_name, price, change_pct, change_amount,
+                        volume, high, low, open_price, market_cap,
+                        per, pbr, eps, bps, roe, ev,
+                        high_52w, low_52w, foreign_ratio, credit_ratio, float_shares, collected_at)
+                    VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14,$15,$16,$17,$18,$19,$20,$21,$22)
+                """, info["code"], info["name"], info["price"], info["change_pct"], info["change_amount"],
+                    info["volume"], info["high"], info["low"], info["open_price"], info["market_cap"],
+                    info["per"], info["pbr"], info["eps"], info["bps"], info["roe"], info["ev"],
+                    info["high_52w"], info["low_52w"], info["foreign_ratio"], info["credit_ratio"],
+                    info["float_shares"], datetime.now())
+        except Exception as e:
+            logger.debug("save_price.err", code=info["code"], error=str(e))
+
+async def save_ohlcv(code: str, rows: list):
+    if not rows or not pg_pool:
+        return
+    # Redis에 최근 60일 일봉 저장 (ta-engine 보조)
+    if redis_cl:
+        await redis_cl.set(f"ohlcv:{code}", json.dumps(rows[:60], ensure_ascii=False), ex=86400)
+    async with pg_pool.acquire() as c:
+        for r in rows:
+            try:
+                dt = datetime.strptime(r["dt"], "%Y%m%d").date()
+                await c.execute("""
+                    INSERT INTO stock_ohlcv (
+                        stock_code, dt, open_price, high_price, low_price, close_price,
+                        volume, trade_amount, foreign_ratio, foreign_net, institution_net, individual_net)
+                    VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12)
+                    ON CONFLICT (stock_code, dt) DO UPDATE SET
+                        close_price=$6, volume=$7, foreign_ratio=$9,
+                        foreign_net=$10, institution_net=$11, individual_net=$12
+                """, code, dt, r["open"], r["high"], r["low"], r["close"],
+                    r["volume"], r["trade_amount"], r["foreign_ratio"],
+                    r["foreign_net"], r["institution_net"], r["individual_net"])
+            except Exception:
+                pass
+
+async def save_foreign_flow(code: str, rows: list):
+    if not rows or not pg_pool:
+        return
+    if redis_cl and rows:
+        await redis_cl.set(f"foreign:{code}", json.dumps(rows[:20], ensure_ascii=False), ex=86400)
+    async with pg_pool.acquire() as c:
+        for r in rows:
+            try:
+                dt = datetime.strptime(r["dt"], "%Y%m%d").date()
+                await c.execute("""
+                    INSERT INTO stock_foreign_flow (
+                        stock_code, dt, close_price, change_qty, hold_qty, hold_ratio, limit_ratio)
+                    VALUES ($1,$2,$3,$4,$5,$6,$7)
+                    ON CONFLICT (stock_code, dt) DO UPDATE SET
+                        change_qty=$4, hold_qty=$5, hold_ratio=$6, limit_ratio=$7
+                """, code, dt, r["close_price"], r["change_qty"],
+                    r["hold_qty"], r["hold_ratio"], r["limit_ratio"])
+            except Exception:
+                pass
+
+async def save_short_sale(code: str, rows: list):
+    if not rows or not pg_pool:
+        return
+    if redis_cl and rows:
+        await redis_cl.set(f"short:{code}", json.dumps(rows[:20], ensure_ascii=False), ex=86400)
+    async with pg_pool.acquire() as c:
+        for r in rows:
+            try:
+                dt = datetime.strptime(r["dt"], "%Y%m%d").date()
+                await c.execute("""
+                    INSERT INTO stock_short_sale (
+                        stock_code, dt, close_price, short_qty, short_balance_qty,
+                        trade_weight, short_avg_price)
+                    VALUES ($1,$2,$3,$4,$5,$6,$7)
+                    ON CONFLICT (stock_code, dt) DO UPDATE SET
+                        short_qty=$4, short_balance_qty=$5, trade_weight=$6, short_avg_price=$7
+                """, code, dt, r["close_price"], r["short_qty"],
+                    r["short_balance_qty"], r["trade_weight"], r["short_avg_price"])
+            except Exception:
+                pass
+
+# ── 수집 작업 ─────────────────────────────────────────────
+
+class Stats:
+    collected = 0; errors = 0; last_run = ""; last_full_run = ""
+
+stats = Stats()
+
+async def job_price():
+    """평일 9~16시 5분마다: 현재가·재무지표 수집 (ka10001)"""
+    codes = await get_stock_codes(0)
+    if not codes:
+        logger.warning("job_price.no_codes")
+        return
+    sem = asyncio.Semaphore(10)  # 동시 10개 제한
+    ok = 0
+    async with httpx.AsyncClient() as client:
+        tasks = [fetch_basic_info(client, code, sem) for code in codes]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+        for info in results:
+            if isinstance(info, dict) and info.get("price", 0) > 0:
+                await save_price(info)
+                ok += 1
+            else:
+                stats.errors += 1
+    await gen_signals()
+    stats.collected += ok
+    stats.last_run = datetime.now().isoformat()
+    if redis_cl:
+        await redis_cl.set("prices:last_update", datetime.now().isoformat())
+    logger.info("job_price.done", ok=ok, total=len(codes))
+
+async def job_full(days: int = 10):
+    """평일 17:00: 일봉·외국인·공매도 전체 수집 (ka10005·ka10008·ka10014).
+    일별은 days=10(경량, ON CONFLICT 갱신). 백필은 /collect/full?days=400."""
+    codes = await get_stock_codes(0)
+    if not codes:
+        return
+    sem = asyncio.Semaphore(5)  # 야간 수집 느리게
+    ok = 0
+    async with httpx.AsyncClient() as client:
+        for code in codes:
+            try:
+                ohlcv, foreign, short = await asyncio.gather(
+                    fetch_ohlcv(client, code, sem, days),
+                    fetch_foreign_flow(client, code, sem),
+                    fetch_short_sale(client, code, sem),
+                    return_exceptions=True
+                )
+                if isinstance(ohlcv, list):
+                    await save_ohlcv(code, ohlcv)
+                if isinstance(foreign, list):
+                    await save_foreign_flow(code, foreign)
+                if isinstance(short, list):
+                    await save_short_sale(code, short)
+                ok += 1
+            except Exception as e:
+                logger.debug("job_full.err", code=code, error=str(e))
+    stats.last_full_run = datetime.now().isoformat()
+    logger.info("job_full.done", ok=ok, total=len(codes))
+
+# ── 시그널 생성 ───────────────────────────────────────────
+
+def calc_signal(info: dict, news_sc: float, dart_sc: float,
+                foreign_net_5d: int, short_weight: float) -> Optional[dict]:
+    price = info["price"]
+    if price <= 0:
+        return None
+    h52 = info.get("high_52w", 0)
+    l52 = info.get("low_52w", 0)
+    cpct = info.get("change_pct", 0)
+    foreign_ratio = info.get("foreign_ratio", 0)
+
+    # 외국인 수급 보너스 (5일 누적 순매수 양수면 +10, 음수면 -10)
+    foreign_bonus = min(10, max(-10, foreign_net_5d / 500_000)) if foreign_net_5d else 0
+    # 공매도 패널티 (비중 5% 이상이면 -5)
+    short_penalty = -5 if short_weight >= 5 else 0
+
+    total = (news_sc * 0.35 + dart_sc * 0.25
+             + (cpct * 10) * 0.25 + foreign_bonus * 0.10 + short_penalty * 0.05)
+    total = max(-100, min(100, total))
+    pos = (price - l52) / (h52 - l52) if h52 > l52 > 0 else 0.5
+
+    if total >= 30:
+        sig = "매수"
+        tp = int(price + (h52 - price) * (0.3 + min(total, 100) / 200)) if h52 > price else int(price * 1.10)
+        sl = int(price * (0.92 + pos * 0.03))
+        conf = min(95, 50 + total * 0.3 + (1 - pos) * 20)
+    elif total <= -30:
+        sig = "매도"
+        tp = int(price - (price - l52) * (0.3 + min(abs(total), 100) / 200)) if l52 > 0 and l52 < price else int(price * 0.90)
+        sl = int(price * 1.05)
+        conf = min(95, 50 + abs(total) * 0.3 + pos * 20)
+    else:
+        return None
+
+    er = ((tp - price) / price) * 100 if sig == "매수" else ((price - tp) / price) * 100
+    risk = abs(sl - price)
+    reward = abs(tp - price)
+    rr = reward / risk if risk > 0 else 0
+    return {
+        "signal_type": sig, "current_price": price, "target_price": tp, "stop_loss": sl,
+        "expected_return_pct": round(er, 2), "risk_reward_ratio": round(rr, 2),
+        "confidence": round(conf, 1), "price_momentum": round(cpct, 2),
+        "news_score": round(news_sc, 1), "dart_score": round(dart_sc, 1),
+        "foreign_net_5d": foreign_net_5d, "short_weight": short_weight,
+    }
+
+async def gen_signals():
+    if not pg_pool:
+        return
+    async with pg_pool.acquire() as c:
+        scores = await c.fetch("""
+            SELECT stock_code, stock_name, news_score, dart_score, total_score
+            FROM stock_scores
+            WHERE score_date = (SELECT MAX(score_date) FROM stock_scores)
+              AND (total_score >= 30 OR total_score <= -30)
+        """)
+        n = 0
+        for row in scores:
+            code = row["stock_code"]
+            if not code or len(code) != 6:
+                continue
+            if not redis_cl:
+                continue
+            cached = await redis_cl.get(f"price:{code}")
+            if not cached:
+                continue
+            info = json.loads(cached)
+
+            # 외국인 5일 순매수 합산
+            foreign_net_5d = 0
+            f_cached = await redis_cl.get(f"foreign:{code}")
+            if f_cached:
+                f_rows = json.loads(f_cached)
+                foreign_net_5d = sum(r.get("change_qty", 0) for r in f_rows[:5])
+
+            # 공매도 최근 비중
+            short_weight = 0.0
+            s_cached = await redis_cl.get(f"short:{code}")
+            if s_cached:
+                s_rows = json.loads(s_cached)
+                if s_rows:
+                    short_weight = s_rows[0].get("trade_weight", 0.0)
+
+            sig = calc_signal(info, row["news_score"], row["dart_score"],
+                              foreign_net_5d, short_weight)
+            if not sig:
+                continue
+            reasons = await c.fetch("""
+                SELECT reason FROM news_analysis
+                WHERE primary_stock=$1 AND intensity>=2
+                ORDER BY analyzed_at DESC LIMIT 2
+            """, code)
+            reason = " | ".join([r["reason"][:100] for r in reasons])
+            await c.execute("""
+                INSERT INTO trade_signals (
+                    stock_code, stock_name, signal_type, current_price,
+                    target_price, stop_loss, expected_return_pct, risk_reward_ratio,
+                    confidence, reason, news_score, dart_score, price_momentum,
+                    foreign_net_5d, short_weight)
+                VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14,$15)
+                ON CONFLICT (stock_code, (created_at::date)) DO UPDATE SET
+                    signal_type=EXCLUDED.signal_type, current_price=EXCLUDED.current_price,
+                    target_price=EXCLUDED.target_price, stop_loss=EXCLUDED.stop_loss,
+                    expected_return_pct=EXCLUDED.expected_return_pct,
+                    risk_reward_ratio=EXCLUDED.risk_reward_ratio,
+                    confidence=EXCLUDED.confidence, reason=EXCLUDED.reason,
+                    news_score=EXCLUDED.news_score, dart_score=EXCLUDED.dart_score,
+                    price_momentum=EXCLUDED.price_momentum,
+                    foreign_net_5d=EXCLUDED.foreign_net_5d, short_weight=EXCLUDED.short_weight
+            """, code, row["stock_name"], sig["signal_type"], sig["current_price"],
+                sig["target_price"], sig["stop_loss"], sig["expected_return_pct"],
+                sig["risk_reward_ratio"], sig["confidence"], reason,
+                sig["news_score"], sig["dart_score"], sig["price_momentum"],
+                sig["foreign_net_5d"], sig["short_weight"])
+            n += 1
+    logger.info("signals.done", count=n)
+
+# ── FastAPI ───────────────────────────────────────────────
+
+app = FastAPI(title="키움증권 주가·수급·공매도")
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+@app.on_event("startup")
+async def startup():
+    global pg_pool, redis_cl
+    pg_pool = await asyncpg.create_pool(
+        host=PG_HOST, port=PG_PORT, database=PG_DB,
+        user=PG_USER, password=PG_PASS, min_size=2, max_size=5)
+    redis_cl = aioredis.Redis(
+        host=REDIS_HOST, port=6379, password=REDIS_PASSWORD, db=3, decode_responses=True)
+    await init_db()
+    # 현재가: 평일 9~16시 5분마다
+    scheduler.add_job(job_price, "cron", day_of_week="mon-fri",
+                      hour="9-15", minute="*/5", id="price", replace_existing=True)
+    # 장 마감 후 전체 수집: 평일 17:00
+    scheduler.add_job(job_full, "cron", day_of_week="mon-fri",
+                      hour="17", minute="0", id="full", replace_existing=True)
+    scheduler.start()
+    logger.info("kiwoom-api.started")
+
+@app.on_event("shutdown")
+async def shutdown():
+    scheduler.shutdown()
+    if pg_pool:
+        await pg_pool.close()
+    if redis_cl:
+        await redis_cl.aclose()
+
+@app.get("/health")
+async def health():
+    lu = await redis_cl.get("prices:last_update") if redis_cl else ""
+    return JSONResponse(content={
+        "status": "ok", "collected": stats.collected, "errors": stats.errors,
+        "last_run": stats.last_run, "last_full_run": stats.last_full_run,
+        "last_update": lu or ""
+    })
+
+@app.get("/price/{code}")
+async def price(code: str):
+    if redis_cl:
+        c = await redis_cl.get(f"price:{code}")
+        if c:
+            return JSONResponse(content=json.loads(c))
+    # 실시간 단건 조회
+    sem = asyncio.Semaphore(1)
+    async with httpx.AsyncClient() as client:
+        info = await fetch_basic_info(client, code, sem)
+        if info:
+            await save_price(info)
+            return JSONResponse(content=info)
+    return JSONResponse(content={"error": "not found"}, status_code=404)
+
+@app.get("/prices")
+async def prices(limit: int = Query(default=50)):
+    if not redis_cl:
+        return JSONResponse(content={"data": {}})
+    keys = await redis_cl.keys("price:*")
+    result = {}
+    for k in keys[:limit]:
+        c = await redis_cl.get(k)
+        if c:
+            d = json.loads(c)
+            result[d["code"]] = d
+    return JSONResponse(content={"count": len(result), "data": result})
+
+@app.get("/foreign/{code}")
+async def foreign(code: str):
+    if redis_cl:
+        c = await redis_cl.get(f"foreign:{code}")
+        if c:
+            return JSONResponse(content={"code": code, "data": json.loads(c)})
+    sem = asyncio.Semaphore(1)
+    async with httpx.AsyncClient() as client:
+        rows = await fetch_foreign_flow(client, code, sem)
+        await save_foreign_flow(code, rows)
+        return JSONResponse(content={"code": code, "data": rows})
+
+@app.get("/short/{code}")
+async def short_sale(code: str):
+    if redis_cl:
+        c = await redis_cl.get(f"short:{code}")
+        if c:
+            return JSONResponse(content={"code": code, "data": json.loads(c)})
+    sem = asyncio.Semaphore(1)
+    async with httpx.AsyncClient() as client:
+        rows = await fetch_short_sale(client, code, sem)
+        await save_short_sale(code, rows)
+        return JSONResponse(content={"code": code, "data": rows})
+
+@app.get("/ohlcv/{code}")
+async def ohlcv(code: str, days: int = Query(default=60)):
+    if redis_cl:
+        c = await redis_cl.get(f"ohlcv:{code}")
+        if c:
+            data = json.loads(c)
+            return JSONResponse(content={"code": code, "data": data[:days]})
+    sem = asyncio.Semaphore(1)
+    async with httpx.AsyncClient() as client:
+        rows = await fetch_ohlcv(client, code, sem, days)
+        await save_ohlcv(code, rows)
+        return JSONResponse(content={"code": code, "data": rows})
+
+@app.get("/signals")
+async def signals(days: int = Query(default=7)):
+    async with pg_pool.acquire() as c:
+        rows = await c.fetch("""
+            SELECT * FROM trade_signals
+            WHERE created_at >= NOW() - INTERVAL '%s days'
+            ORDER BY confidence DESC LIMIT 50
+        """ % days)
+    return [dict(r) for r in rows]
+
+@app.get("/signals/{code}")
+async def stock_signals(code: str):
+    async with pg_pool.acquire() as c:
+        rows = await c.fetch("""
+            SELECT * FROM trade_signals WHERE stock_code=$1
+            ORDER BY created_at DESC LIMIT 10
+        """, code)
+    return [dict(r) for r in rows]
+
+@app.get("/summary/{code}")
+async def summary(code: str):
+    """종목 종합 요약 (현재가 + 외국인 + 공매도 + 최근 신호)"""
+    result = {"code": code}
+    if redis_cl:
+        for key in [f"price:{code}", f"foreign:{code}", f"short:{code}", f"ohlcv:{code}"]:
+            c = await redis_cl.get(key)
+            if c:
+                field = key.split(":")[0]
+                result[field] = json.loads(c) if field != "price" else json.loads(c)
+    return JSONResponse(content=result)
+
+@app.post("/collect/price")
+async def collect_price():
+    asyncio.create_task(job_price())
+    return {"status": "started", "job": "price"}
+
+@app.post("/collect/full")
+async def collect_full(days: int = Query(default=10, ge=1, le=600)):
+    asyncio.create_task(job_full(days))
+    return {"status": "started", "job": "full", "days": days}
+
+# ── 추가: 분봉 / 호가 / 거래량급증 ─────────────────────────
+
+@app.get("/minute/{code}")
+async def minute_chart(code: str, scope: str = Query(default="5", description="분봉 단위: 1,3,5,10,15,30,60")):
+    """ka10080: 분봉차트 (실시간 호출)"""
+    async with httpx.AsyncClient() as c:
+        data = await fetch_minute_chart(c, code, scope)
+    return JSONResponse(content={"code": code, "scope": scope, "data": data})
+
+@app.get("/orderbook/{code}")
+async def orderbook(code: str):
+    """ka10004: 10단계 호가 잔량"""
+    async with httpx.AsyncClient() as c:
+        data = await fetch_orderbook(c, code)
+    return JSONResponse(content=data or {"code": code, "ask": [], "bid": []})
+
+@app.get("/volume-surge")
+async def volume_surge():
+    """ka10023: 거래량 급증 종목 TOP50 (Redis 5분 캐시)"""
+    cached = None
+    if redis_cl:
+        try:
+            v = await redis_cl.get("vol_surge:list")
+            if v: cached = json.loads(v)
+        except: pass
+    if cached:
+        return JSONResponse(content={"data": cached, "cached": True})
+    data = await fetch_volume_surge()
+    if redis_cl and data:
+        try: await redis_cl.setex("vol_surge:list", 300, json.dumps(data))
+        except: pass
+    return JSONResponse(content={"data": data, "cached": False})
@@ -0,0 +1,8 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+redis==5.0.4
+asyncpg==0.29.0
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
@@ -0,0 +1,269 @@
+"""
+네이버 금융 뉴스 수집기 + AI 분석 파이프라인
+- 시장 전체 뉴스 (5분마다)
+- 시총 상위 200개 종목별 뉴스 (30분마다)
+- 수집 즉시 바른API → Ollama → Qdrant → vLLM → PostgreSQL
+"""
+import asyncio, hashlib, json, os, re, random, time
+from datetime import datetime, timedelta
+from typing import Optional
+import asyncpg, httpx, redis.asyncio as aioredis, structlog
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from bs4 import BeautifulSoup
+from fastapi import FastAPI, Query
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+REDIS_HOST = os.getenv("REDIS_HOST", "redis")
+REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", "")
+PG_HOST = os.getenv("POSTGRES_HOST", "postgres")
+PG_PORT = int(os.getenv("POSTGRES_PORT", "5432"))
+PG_DB = os.getenv("POSTGRES_DB", "trading_ai")
+PG_USER = os.getenv("POSTGRES_USER", "kyu")
+PG_PASS = os.getenv("POSTGRES_PASSWORD", "7895123")
+BAREUN_URL = os.getenv("BAREUN_API_URL", "http://bareunaapi:5757")
+OLLAMA_URL = os.getenv("OLLAMA_URL", "http://ollama:11434")
+VLLM_URL = os.getenv("VLLM_URL", "http://vllm:8000")
+QDRANT_URL = os.getenv("QDRANT_URL", "http://qdrant:6333")
+
+HEADERS = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"}
+
+pg_pool: Optional[asyncpg.Pool] = None
+redis_cl: Optional[aioredis.Redis] = None
+scheduler = AsyncIOScheduler(timezone="Asia/Seoul")
+
+class S:
+    collected = 0; processed = 0; duplicates = 0; errors = 0
+    last_run = ""; running = False
+stats = S()
+
+def nhash(title, url=""): return hashlib.sha256(f"{title.strip()}{url.strip()}".encode()).hexdigest()[:16]
+
+async def is_dup(h):
+    if not redis_cl: return False
+    try:
+        r = await redis_cl.set(f"news:naver:{h}", "1", ex=86400, nx=True)
+        return r is None
+    except: return False
+
+# ── 크롤러 ────────────────────────────────────────────────
+async def crawl_market_news(client):
+    news = []
+    urls = [
+        "https://finance.naver.com/news/mainnews.naver",
+        "https://finance.naver.com/news/news_list.naver?mode=LSS2D&section_id=101&section_id2=258",
+        "https://finance.naver.com/news/news_list.naver?mode=LSS2D&section_id=101&section_id2=259",
+        "https://finance.naver.com/news/news_list.naver?mode=LSS2D&section_id=101&section_id2=261",
+    ]
+    for url in urls:
+        try:
+            r = await client.get(url, headers=HEADERS, timeout=15)
+            r.encoding = "euc-kr"
+            soup = BeautifulSoup(r.text, "lxml")
+            for a in soup.select("a[href*='article_id']"):
+                t = a.get_text(strip=True)
+                h = a.get("href", "")
+                if t and len(t) > 10:
+                    news.append({"title": t, "url": f"https://finance.naver.com{h}" if h.startswith("/") else h,
+                                 "source": "네이버금융", "content": "", "published_at": datetime.now().isoformat()})
+            await asyncio.sleep(0.3)
+        except Exception as e:
+            logger.warning("crawl.market.err", error=str(e))
+    return news
+
+async def crawl_stock_news(client, code, name):
+    news = []
+    try:
+        r = await client.get(f"https://finance.naver.com/item/news_news.naver?code={code}&page=1", headers=HEADERS, timeout=15)
+        r.encoding = "euc-kr"
+        soup = BeautifulSoup(r.text, "lxml")
+        for tr in soup.select("table.type5 tr"):
+            a = tr.select_one("td.title a")
+            dt = tr.select_one("td.date")
+            src = tr.select_one("td.info")
+            if a:
+                t = a.get_text(strip=True)
+                if t and len(t) > 5:
+                    news.append({"title": t, "url": f"https://finance.naver.com{a.get('href','')}",
+                                 "source": src.get_text(strip=True) if src else "네이버금융",
+                                 "content": f"[{name}({code})] {t}",
+                                 "published_at": dt.get_text(strip=True) if dt else datetime.now().isoformat(),
+                                 "stock_code": code, "stock_name": name})
+    except: pass
+    return news
+
+async def get_top_stocks(client, count=200):
+    stocks = []
+    for sosok in [0, 1]:
+        for page in range(1, 50):
+            try:
+                r = await client.get(f"https://finance.naver.com/sise/sise_market_sum.naver?sosok={sosok}&page={page}", headers=HEADERS, timeout=15)
+                r.encoding = "euc-kr"
+                rows = re.findall(r'main\.naver\?code=(\d{6})[^>]*>([^<]+)</a>', r.text)
+                if not rows: break
+                for c, n in rows: stocks.append({"code": c.strip(), "name": n.strip()})
+                await asyncio.sleep(0.2)
+            except: break
+            if len(stocks) >= count: break
+    return stocks[:count]
+
+# ── 파이프라인 ─────────────────────────────────────────────
+async def pipeline(item, client):
+    try:
+        # 1. 바른API
+        br = await client.post(f"{BAREUN_URL}/analyze", json={
+            "title": item["title"], "content": item.get("content",""),
+            "url": item.get("url",""), "source": item.get("source",""),
+            "published_at": item.get("published_at","")}, timeout=30)
+        bd = br.json()
+        if bd.get("is_duplicate"): return "dup"
+
+        data = {**item, "hash": bd.get("hash",""), "stocks": bd.get("stocks",[]),
+                "keywords": bd.get("keywords",[]), "filtered_text": bd.get("filtered_text","")}
+
+        # 2. Ollama 임베딩
+        er = await client.post(f"{OLLAMA_URL}/api/embeddings",
+            json={"model":"bge-m3","prompt": data["filtered_text"] or data["title"]}, timeout=60)
+        emb = er.json().get("embedding")
+        if not emb: return "no_embed"
+
+        # 3. Qdrant 유사도
+        try:
+            sr = await client.post(f"{QDRANT_URL}/collections/news_vectors/points/search",
+                json={"vector":emb,"limit":3,"score_threshold":0.85,"with_payload":True}, timeout=15)
+            hits = sr.json().get("result",[])
+            if any(h["score"]>=0.92 and h["score"]<0.99 for h in hits): return "sim_dup"
+        except: hits = []
+
+        # 4. vLLM 분석
+        stocks_str = ", ".join([f'{s["name"]}({s["code"]})' for s in data["stocks"][:5]])
+        prompt = (f"뉴스: {data['title'][:200]}\n키워드: {(data['filtered_text'] or '')[:300]}\n"
+                  f"감지된 종목: {stocks_str}\n\n"
+                  "JSON 응답. primary_stock에 6자리 종목코드, 시장전체면 KOSPI/KOSDAQ:\n"
+                  '{"sentiment":"호재/악재/중립","intensity":1~5,"primary_stock":"종목코드",'
+                  '"affected_stocks":["코드"],"reason":"근거","investment_action":"매수관심/매도관심/관망"}')
+        vr = await client.post(f"{VLLM_URL}/v1/chat/completions", json={
+            "model":"exaone","messages":[
+                {"role":"system","content":"한국 주식 전문 애널리스트. JSON만 응답."},
+                {"role":"user","content":prompt}],
+            "max_tokens":300,"temperature":0.1}, timeout=120)
+        try:
+            c = vr.json()["choices"][0]["message"]["content"]
+            a = json.loads(re.sub(r"```json\n?|```","",c).strip())
+        except:
+            a = {"sentiment":"중립","intensity":0,"primary_stock":"","affected_stocks":[],"reason":"파싱실패","investment_action":"관망"}
+
+        # 5. Qdrant 저장
+        try:
+            await client.put(f"{QDRANT_URL}/collections/news_vectors/points", json={
+                "points":[{"id":random.randint(1,999999999),"vector":emb,
+                    "payload":{"title":data["title"],"hash":data["hash"],
+                               "sentiment":a.get("sentiment",""),"intensity":a.get("intensity",0),
+                               "primary_stock":a.get("primary_stock","")}}]}, timeout=15)
+        except: pass
+
+        # 6. PostgreSQL 저장
+        esc = lambda s: (s or "").replace("'","''")
+        await pg_pool.execute(f"""
+            INSERT INTO news_analysis (title,url,source,published_at,hash,sentiment,intensity,
+                primary_stock,affected_stocks,reason,investment_action,keywords,stock_names,stock_codes,similar_count,analyzed_at)
+            VALUES ('{esc(data["title"][:500])}','{esc(data.get("url","")[:500])}','{esc(data.get("source","")[:100])}',
+                '{data.get("published_at",datetime.now().isoformat())}','{data["hash"]}',
+                '{a.get("sentiment","중립")}',{a.get("intensity",0)},'{esc(a.get("primary_stock",""))}',
+                '{json.dumps(a.get("affected_stocks",[]))}','{esc(a.get("reason","")[:500])}',
+                '{a.get("investment_action","관망")}','{json.dumps(data.get("keywords",[])[:20])}',
+                '{json.dumps([s["name"] for s in data.get("stocks",[])])}',
+                '{json.dumps([s["code"] for s in data.get("stocks",[])])}',
+                0,'{datetime.now().isoformat()}')
+            ON CONFLICT (hash) DO NOTHING""")
+        return "ok"
+    except Exception as e:
+        logger.warning("pipeline.err", title=item.get("title","")[:50], error=str(e))
+        return "error"
+
+# ── 수집 작업 ──────────────────────────────────────────────
+async def job_market():
+    if stats.running: return
+    stats.running = True
+    try:
+        async with httpx.AsyncClient() as c:
+            news = await crawl_market_news(c)
+            ok = 0
+            for item in news:
+                h = nhash(item["title"], item.get("url",""))
+                if await is_dup(h): stats.duplicates+=1; continue
+                stats.collected += 1
+                r = await pipeline(item, c)
+                if r == "ok": ok += 1; stats.processed += 1
+                await asyncio.sleep(0.5)
+            stats.last_run = datetime.now().isoformat()
+            logger.info("job.market", total=len(news), processed=ok)
+    except Exception as e:
+        stats.errors += 1; logger.error("job.market.err", error=str(e))
+    finally:
+        stats.running = False
+
+async def job_stocks():
+    if stats.running: return
+    stats.running = True
+    try:
+        async with httpx.AsyncClient() as c:
+            top = await get_top_stocks(c, 200)
+            ok = 0
+            for stock in top:
+                try:
+                    news = await crawl_stock_news(c, stock["code"], stock["name"])
+                    for item in news:
+                        h = nhash(item["title"], item.get("url",""))
+                        if await is_dup(h): stats.duplicates+=1; continue
+                        stats.collected += 1
+                        r = await pipeline(item, c)
+                        if r == "ok": ok += 1; stats.processed += 1
+                    await asyncio.sleep(0.3)
+                except: pass
+            stats.last_run = datetime.now().isoformat()
+            logger.info("job.stocks", stocks=len(top), processed=ok)
+    except Exception as e:
+        stats.errors += 1; logger.error("job.stocks.err", error=str(e))
+    finally:
+        stats.running = False
+
+# ── FastAPI ────────────────────────────────────────────────
+app = FastAPI(title="뉴스 수집기")
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+@app.on_event("startup")
+async def startup():
+    global pg_pool, redis_cl
+    pg_pool = await asyncpg.create_pool(host=PG_HOST,port=PG_PORT,database=PG_DB,user=PG_USER,password=PG_PASS,min_size=2,max_size=5)
+    redis_cl = aioredis.Redis(host=REDIS_HOST,port=6379,password=REDIS_PASSWORD,db=4,decode_responses=True)
+    scheduler.add_job(job_market,"cron",day_of_week="mon-fri",hour="8-18",minute="*/5",id="market",replace_existing=True)
+    scheduler.add_job(job_stocks,"cron",day_of_week="mon-fri",hour="9-16",minute="*/30",id="stocks",replace_existing=True)
+    scheduler.start()
+    logger.info("news-collector.started")
+
+@app.on_event("shutdown")
+async def shutdown():
+    scheduler.shutdown()
+    if pg_pool: await pg_pool.close()
+    if redis_cl: await redis_cl.aclose()
+
+@app.get("/health")
+async def health():
+    return JSONResponse(content={"status":"ok","collected":stats.collected,"processed":stats.processed,
+        "duplicates":stats.duplicates,"errors":stats.errors,"last_run":stats.last_run,"running":stats.running})
+
+@app.post("/collect/market")
+async def m_market():
+    asyncio.create_task(job_market()); return {"status":"started","type":"market"}
+
+@app.post("/collect/stocks")
+async def m_stocks():
+    asyncio.create_task(job_stocks()); return {"status":"started","type":"stocks"}
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8787
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8787", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,886 @@
+"""
+멀티소스 금융 뉴스 수집기 + AI 분석 파이프라인 (버핏 스타일 강화)
+- RSS 28개 소스 (네이버 크롤링 제거, 주식 전문 사이트 중심)
+- 매 5분마다 전체 소스 수집
+- 2단계 중복제거: URL해시(Redis) + 제목정규화해시(Redis) + 벡터유사도(Qdrant)
+- 수집 즉시 바른API → Ollama → Qdrant → Ollama(EXAONE) → PostgreSQL
+"""
+import asyncio, hashlib, json, os, re, random, time
+import xml.etree.ElementTree as ET
+from datetime import datetime, timedelta
+from typing import Optional
+import asyncpg, httpx, redis.asyncio as aioredis, structlog
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from bs4 import BeautifulSoup
+from fastapi import FastAPI, Query
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+REDIS_HOST = os.getenv("REDIS_HOST", "redis")
+REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", "")
+PG_HOST = os.getenv("POSTGRES_HOST", "postgres")
+PG_PORT = int(os.getenv("POSTGRES_PORT", "5432"))
+PG_DB = os.getenv("POSTGRES_DB", "trading_ai")
+PG_USER = os.getenv("POSTGRES_USER", "kyu")
+PG_PASS = os.getenv("POSTGRES_PASSWORD", "7895123")
+BAREUN_URL = os.getenv("BAREUN_API_URL", "http://bareunaapi:5757")
+OLLAMA_URL = os.getenv("OLLAMA_URL", "http://ollama:11434")
+QDRANT_URL = os.getenv("QDRANT_URL", "http://qdrant:6333")
+
+HEADERS = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"}
+
+pg_pool: Optional[asyncpg.Pool] = None
+redis_cl: Optional[aioredis.Redis] = None
+scheduler = AsyncIOScheduler(timezone="Asia/Seoul")
+
+class S:
+    collected = 0; processed = 0; duplicates = 0; errors = 0
+    noise = 0; off_topic = 0
+    last_run = ""; running = False
+stats = S()
+
+def nhash(title, url=""): return hashlib.sha256(f"{title.strip()}{url.strip()}".encode()).hexdigest()[:16]
+
+def normalize_title(title: str) -> str:
+    """[속보][단독](종합) 등 제거 후 특수문자·공백 제거 → 유사 제목 중복 감지용"""
+    t = re.sub(r'[\[\(【〔][^\]\)】〕]{0,10}[\]\)】〕]', '', title)
+    t = re.sub(r'[^\w가-힣a-zA-Z0-9]', '', t)
+    return t.lower().strip()
+
+def is_korean(text: str) -> bool:
+    if not text: return True
+    hangul = sum(1 for c in text if '가' <= c <= '힣')
+    cjk = sum(1 for c in text if '一' <= c <= '鿿' or '㐀' <= c <= '䶿')
+    if hangul == 0 and cjk > 2: return False
+    return True
+
+async def is_dup(title: str, url: str = "") -> bool:
+    if not redis_cl: return False
+    try:
+        # 1차: URL+제목 해시 (완전 동일 기사)
+        h_url = nhash(title, url)
+        if await redis_cl.set(f"news:u:{h_url}", "1", ex=172800, nx=True) is None:
+            return True
+        # 2차: 정규화된 제목 해시 (같은 뉴스 다른 소스)
+        h_norm = nhash(normalize_title(title))
+        if await redis_cl.set(f"news:t:{h_norm}", "1", ex=21600, nx=True) is None:
+            return True
+        return False
+    except: return False
+
+# ── RSS 멀티소스 수집 (28개 소스, 네이버 크롤링 제거) ──────
+
+RSS_SOURCES = [
+    # ── 주요 경제/금융 신문 ──
+    ("한국경제",       "https://www.hankyung.com/feed/all-news"),
+    ("한국경제",       "https://www.hankyung.com/feed/finance"),
+    ("매일경제",       "https://www.mk.co.kr/rss/30100041/"),
+    ("매일경제",       "https://www.mk.co.kr/rss/30200030/"),
+    ("머니투데이",     "https://news.mt.co.kr/mtview.php?type=rss&MTPub=E&section=E"),
+    ("머니투데이",     "https://news.mt.co.kr/mtview.php?type=rss&MTPub=S"),
+    ("이데일리",       "https://www.edaily.co.kr/rss/news/finance.xml"),
+    ("이데일리",       "https://www.edaily.co.kr/rss/news/markets.xml"),
+    ("연합뉴스",       "https://www.yna.co.kr/rss/economy.xml"),
+    ("연합뉴스",       "https://www.yna.co.kr/rss/market.xml"),
+    ("조선비즈",       "https://biz.chosun.com/rss/rss.htm?site=biz.chosun.com"),
+    ("헤럴드경제",     "https://biz.heraldcorp.com/rss/index.xml"),
+    ("파이낸셜뉴스",   "https://www.fnnews.com/rss/fn_economy.xml"),
+    ("서울경제",       "https://www.sedaily.com/RSS/Economy"),
+    ("아시아경제",     "https://www.asiae.co.kr/rss/economy.htm"),
+    ("뉴스1",          "https://www.news1.kr/rss/industry.xml"),
+    ("비즈니스포스트", "https://www.businesspost.co.kr/BP?command=rss&rssFeedType=0000"),
+    ("인포스탁",       "https://www.infostock.co.kr/rss/rss_news.xml"),
+    # ── 추가 소스 ──
+    ("뉴스핌",         "https://www.newspim.com/rss/view?outtype=1"),
+    ("뉴시스",         "https://www.newsis.com/rss/finance.xml"),
+    ("이투데이",       "https://www.etoday.co.kr/news/newsfeed.php?CateId=0102"),
+    ("글로벌이코노믹", "https://www.g-enews.com/rss/rss_economy.xml"),
+    ("전자신문",       "https://www.etnews.com/news/latest_news.xml"),
+    ("디지털타임스",   "https://www.dt.co.kr/rss/rss_economy.html"),
+    ("더벨",           "https://www.thebell.co.kr/free/content/xmlService.asp"),
+    ("스탁데일리",     "https://www.stockdaily.kr/rss/rss.xml"),
+    ("세계일보",       "https://www.segye.com/RSS/economyRss.xml"),
+    ("SBS Biz",        "https://news.sbs.co.kr/news/SectionRssFeed.do?sectionId=EC"),
+]
+
+def parse_rss_date(date_str: str) -> str:
+    if not date_str:
+        return datetime.now().isoformat()
+    for fmt in ("%a, %d %b %Y %H:%M:%S %z", "%a, %d %b %Y %H:%M:%S +0900",
+                "%Y-%m-%dT%H:%M:%S%z", "%Y-%m-%d %H:%M:%S"):
+        try:
+            return datetime.strptime(date_str.strip(), fmt).isoformat()
+        except: pass
+    return datetime.now().isoformat()
+
+async def crawl_rss_sources(client) -> list:
+    news = []
+    for source_name, rss_url in RSS_SOURCES:
+        try:
+            r = await client.get(rss_url, timeout=15,
+                headers={"User-Agent": "Mozilla/5.0", "Accept": "application/rss+xml, application/xml"})
+            if r.status_code != 200:
+                continue
+            try:
+                root = ET.fromstring(r.content)
+            except ET.ParseError:
+                # XML 파싱 실패 시 BeautifulSoup 폴백
+                soup = BeautifulSoup(r.content, "lxml-xml")
+                items = soup.find_all("item")
+                for item in items[:20]:
+                    title = item.find("title")
+                    link  = item.find("link")
+                    pub   = item.find("pubDate") or item.find("dc:date")
+                    if title and title.get_text(strip=True):
+                        t = title.get_text(strip=True)
+                        if len(t) > 5 and is_korean(t):
+                            news.append({
+                                "title": t,
+                                "url": link.get_text(strip=True) if link else "",
+                                "source": source_name,
+                                "content": "",
+                                "published_at": parse_rss_date(pub.get_text(strip=True) if pub else ""),
+                            })
+                continue
+
+            ns = {"dc": "http://purl.org/dc/elements/1.1/"}
+            items = root.findall(".//item")
+            for item in items[:20]:
+                title  = item.findtext("title", "").strip()
+                link   = item.findtext("link", "").strip()
+                pubdate = item.findtext("pubDate", "") or item.findtext("dc:date", "", ns)
+                desc   = item.findtext("description", "")
+                if title and len(title) > 5 and is_korean(title):
+                    news.append({
+                        "title": title,
+                        "url": link,
+                        "source": source_name,
+                        "content": re.sub(r"<[^>]+>", "", desc)[:300] if desc else "",
+                        "published_at": parse_rss_date(pubdate),
+                    })
+            await asyncio.sleep(0.2)
+        except Exception as e:
+            logger.debug("rss.fetch.err", source=source_name, error=str(e))
+    logger.info("rss.crawled", sources=len(RSS_SOURCES), items=len(news))
+    return news
+
+# ── 비주식 뉴스 필터 ───────────────────────────────────────
+# 명백한 비주식 카테고리 prefix → 수집 차단
+NOISE_PREFIXES = (
+    # 도서/문화
+    "[책마을]", "[책꽂이]", "[책&생각]", "[책]", "[리뷰]", "[영화]",
+    "[공연]", "[전시]", "[음악]", "[연예]",
+    # 의견/사설
+    "[사설]", "[칼럼]", "[기고]", "[오피니언]", "[기자수첩]", "[데스크칼럼]",
+    "[Why]", "[취재일기]", "[현장에서]", "[광화문]", "[로터리]",
+    # 인사/사회
+    "[부고]", "[인사]", "[동정]", "[포토]", "[화보]", "[지면]", "[그래픽]",
+    # 라이프
+    "[건강]", "[푸드]", "[여행]", "[패션]", "[뷰티]", "[자동차]", "[취미]",
+    "[운세]", "[날씨]",
+    # 스포츠
+    "[스포츠]", "[야구]", "[축구]", "[농구]", "[골프]",
+    # 광고성/이벤트
+    "[보도자료]", "[알림]", "[이벤트]", "[알립니다]",
+)
+
+# 종목/시장과 무관한 라이프스타일·클릭베이트 키워드 (제목에 등장하면 차단)
+NOISE_KEYWORDS = (
+    # 라이프/소비
+    "이직", "퇴사", "신의 직장", "마통", "돈복사", "전기료 폭탄",
+    "세탁기", "건조기", "에어컨", "다이어트", "맛집", "할인",
+    "결혼", "이혼", "연애", "임신", "출산", "육아",
+    # 사회/사건사고 (기업이 부수적으로 언급되는 경우 多)
+    "성폭행", "성추행", "음주운전", "마약", "도박", "사기",
+    "교통사고", "화재", "살인", "강도",
+    # 정치 (종목 영향 적은 일반 정치)
+    "선거", "투표", "여론조사", "지지율", "총선", "대선",
+    # 라이프스타일 클릭베이트
+    "충격", "발칵", "경악", "헉", "화들짝", "깜짝", "대체",
+)
+# 클릭베이트: 제목이 큰따옴표·작은따옴표로 시작하면 의심
+CLICKBAIT_QUOTE_RE = re.compile(r'^[\s]*[\"“‘’\']')
+# 종목 매칭 0건일 때 시장 관련성 판정용 키워드
+MARKET_KEYWORDS = (
+    # 지수/시장
+    "코스피", "코스닥", "KOSPI", "KOSDAQ", "나스닥", "다우", "S&P", "증시",
+    "지수", "시장", "장세", "장중", "거래소",
+    # 거시지표
+    "환율", "금리", "원달러", "원화", "달러", "엔화", "위안", "유가",
+    "금값", "은값", "원유", "WTI", "브렌트", "채권", "국채", "회사채",
+    "인플레이션", "물가", "CPI", "PPI", "성장률", "GDP", "경기", "경제",
+    # 수급
+    "외국인", "외인", "기관", "개인투자자", "프로그램매매", "공매도",
+    "신용잔고", "거래대금", "거래량",
+    # 종목/거래
+    "주식", "주가", "종목", "매수", "매도", "매매", "투자", "테마주", "관련주",
+    "급등", "급락", "상승", "하락", "신고가", "신저가", "호재", "악재",
+    # 펀드/상품
+    "펀드", "ETF", "ETN", "선물", "옵션", "파생", "리츠",
+    # 기업이벤트
+    "상장", "상폐", "공모", "IPO", "유상증자", "무상증자", "배당", "자사주",
+    "M&A", "인수", "합병", "분할", "지분", "스톡옵션",
+    # 실적/재무
+    "실적", "영업이익", "매출", "순이익", "EBITDA", "ROE", "PER", "PBR",
+    "수주", "계약", "수출", "공시",
+    # 산업/섹터
+    "산업", "섹터", "반도체", "바이오", "2차전지", "배터리", "전기차",
+    "AI", "로봇", "방산", "조선", "건설", "유통", "금융", "은행", "보험",
+    "증권", "통신", "에너지", "철강", "화학", "정유"
+)
+
+def _is_noise_title(title: str) -> bool:
+    if not title:
+        return True
+    if any(title.startswith(p) for p in NOISE_PREFIXES):
+        return True
+    # 클릭베이트: 따옴표 시작 + 시장 키워드 부재 시 차단
+    if CLICKBAIT_QUOTE_RE.match(title) and not any(kw in title for kw in MARKET_KEYWORDS):
+        return True
+    # 라이프스타일/사건사고 키워드 (제목에서 검사)
+    if any(nk in title for nk in NOISE_KEYWORDS):
+        return True
+    return False
+
+def _has_market_relevance(text: str) -> bool:
+    """종목 매칭이 없을 때만 호출. 시장 키워드 ≥2개 매칭 시 통과 (broad 1개로는 부족)"""
+    if not text:
+        return False
+    matches = sum(1 for kw in MARKET_KEYWORDS if kw in text)
+    return matches >= 2
+
+# ── 파이프라인 ─────────────────────────────────────────────
+_corp_cache: dict = {}
+async def _corp_name(code: str) -> str:
+    if code in _corp_cache:
+        return _corp_cache[code]
+    nm = code
+    try:
+        row = await pg_pool.fetchrow(
+            "SELECT corp_name FROM dart_corps WHERE stock_code=$1", code)
+        if row and row["corp_name"]:
+            nm = row["corp_name"]
+    except Exception:
+        pass
+    _corp_cache[code] = nm
+    return nm
+
+
+async def pipeline(item, client):
+    try:
+        # 0. 비주식 카테고리 prefix 차단 (바른API 호출 전)
+        if _is_noise_title(item.get("title", "")):
+            return "noise"
+
+        # 1. 바른API
+        br = await client.post(f"{BAREUN_URL}/analyze", json={
+            "title": item["title"], "content": item.get("content",""),
+            "url": item.get("url",""), "source": item.get("source",""),
+            "published_at": item.get("published_at","")}, timeout=30)
+        bd = br.json()
+        if bd.get("is_duplicate"): return "dup"
+
+        data = {**item, "hash": bd.get("hash",""), "stocks": bd.get("stocks",[]),
+                "keywords": bd.get("keywords",[]), "filtered_text": bd.get("filtered_text","")}
+
+        # raw 백필: 종목코드가 확정된 뉴스는 종목명 인식 실패해도 보존 (관련성 과필터 완화)
+        kc = item.get("stock_code")
+        if kc and not any(st.get("code") == kc for st in data["stocks"]):
+            data["stocks"] = [{"name": await _corp_name(kc), "code": kc}] + data["stocks"]
+
+        # 1.5. 종목 매칭 + 시장 키워드 둘 다 없으면 스킵 (LLM 비용 절감)
+        if not data["stocks"]:
+            full_text = (item.get("title","") or "") + " " + (item.get("content","") or "")
+            if not _has_market_relevance(full_text):
+                return "off_topic"
+
+        # 2. Ollama 임베딩
+        er = await client.post(f"{OLLAMA_URL}/api/embeddings",
+            json={"model":"bge-m3","prompt": data["filtered_text"] or data["title"]}, timeout=60)
+        emb = er.json().get("embedding")
+        if not emb: return "no_embed"
+
+        # 3. Qdrant 유사도
+        try:
+            sr = await client.post(f"{QDRANT_URL}/collections/news_vectors/points/search",
+                json={"vector":emb,"limit":6,"score_threshold":0.80,"with_payload":True}, timeout=15)
+            hits = sr.json().get("result",[])
+            if any(h["score"]>=0.92 for h in hits): return "sim_dup"  # ≥0.99 근접중복도 차단
+        except: hits = []
+
+        # 3.5. RAG 컨텍스트: 유사 과거뉴스 + 종목 재무·추세·점수 (버핏 판단 근거 주입)
+        ctx = []
+        rel = [h for h in hits if 0.80 <= h.get("score", 0) < 0.92][:4]
+        if rel:
+            ctx.append("· 유사 과거뉴스:")
+            for h in rel:
+                p = h.get("payload", {})
+                ctx.append(f"  - {(p.get('title') or '')[:55]} → "
+                           f"{p.get('sentiment','?')}/강도{p.get('intensity','?')}")
+        focus = data["stocks"][0] if data["stocks"] else None
+        if focus:
+            g = lambda v: "-" if v is None else v
+            try:
+                fin = await pg_pool.fetchrow("""
+                    SELECT ROUND(roe::numeric,1) roe, ROUND(operating_margin::numeric,1) om,
+                           ROUND(debt_ratio::numeric,1) dr, ROUND(fcf_ratio::numeric,2) fcf,
+                           ROUND(revenue_growth::numeric,1) rg
+                    FROM dart_financials WHERE stock_code=$1 AND reprt_code='11011'
+                    ORDER BY bsns_year DESC LIMIT 1""", focus["code"])
+                if fin:
+                    ctx.append(f"· {focus['name']} 재무: ROE {g(fin['roe'])}% "
+                               f"영업이익률 {g(fin['om'])}% 부채비율 {g(fin['dr'])}% "
+                               f"FCF {g(fin['fcf'])} 매출성장 {g(fin['rg'])}%")
+                his = await pg_pool.fetchrow("""
+                    SELECT COUNT(*) FILTER (WHERE sentiment='호재') pos,
+                           COUNT(*) FILTER (WHERE sentiment='악재') neg,
+                           ROUND(AVG(intensity)::numeric,1) ai,
+                           MODE() WITHIN GROUP (ORDER BY catalyst) cat
+                    FROM news_analysis
+                    WHERE primary_stock=$1 AND analyzed_at >= NOW()-INTERVAL '14 days'
+                    """, focus["code"])
+                if his and (his["pos"] or his["neg"]):
+                    ctx.append(f"· 최근14일 {focus['name']}: 호재 {his['pos']} / "
+                               f"악재 {his['neg']}, 평균강도 {g(his['ai'])}, "
+                               f"주catalyst {his['cat'] or '-'}")
+                recent = await pg_pool.fetch("""
+                    SELECT sentiment, intensity, catalyst, left(reason,50) reason
+                    FROM news_analysis
+                    WHERE primary_stock=$1 AND reason != '파싱실패'
+                      AND analyzed_at >= NOW()-INTERVAL '30 days'
+                    ORDER BY analyzed_at DESC LIMIT 2""", focus["code"])
+                if recent:
+                    ctx.append(f"· {focus['name']} 최근 분석:")
+                    for r in recent:
+                        ctx.append(f"  - {r['sentiment']}/강도{r['intensity']} "
+                                   f"[{r['catalyst'] or '-'}] {r['reason']}")
+                sc = await pg_pool.fetchrow("""
+                    SELECT ROUND(total_score::numeric,1) total_score, recommendation
+                    FROM stock_scores
+                    WHERE stock_code=$1 ORDER BY score_date DESC LIMIT 1""",
+                    focus["code"])
+                if sc:
+                    ctx.append(f"· 현재 퀀트: 종합점수 {g(sc['total_score'])} "
+                               f"({sc['recommendation'] or '-'})")
+            except Exception as e:
+                logger.debug("rag.ctx_err", code=focus["code"], error=str(e))
+        context_block = "\n".join(ctx)
+
+        # 4. Ollama EXAONE 분석 (버핏 관점 강화 프롬프트)
+        stocks_str = ", ".join([f'{s["name"]}({s["code"]})' for s in data["stocks"][:5]])
+        source = item.get("source", "")
+        content_preview = (data.get("filtered_text") or item.get("content") or "")[:400]
+        system_prompt = (
+            "당신은 워렌 버핏 스타일의 한국 주식 가치투자 애널리스트입니다.\n"
+            "뉴스 한 건을 읽고 기업 본질가치 관점에서 영향을 평가합니다.\n"
+            "판단 우선순위: 기업 본질가치(실적·FCF·경쟁우위) > 재무 리스크 > 단기 수급·테마\n\n"
+            "[sentiment] 호재 / 악재 / 중립 — 본질가치 또는 주가에 미치는 방향\n\n"
+            "[intensity] 1~5 정수\n"
+            " 5=상장폐지·대규모 횡령 등 기업 존폐\n"
+            " 4=연간실적 20% 이상 영향(대형 수주·어닝쇼크 등)\n"
+            " 3=분기실적에 유의미한 영향\n"
+            " 2=영업환경·업황 변화(직접 실적 수치 영향은 제한적)\n"
+            " 1=단순 정보성·정기공시·소폭 코멘트\n\n"
+            "[catalyst] 반드시 아래 6개 중 정확히 하나만 출력 (다른 단어·복합어·영문·한자 절대 금지):\n"
+            ' "실적"  = 영업이익·매출·순이익 증감, 흑자/적자 전환, 어닝 서프라이즈/쇼크, 실적 가이던스\n'
+            ' "수주"  = 신규 수주·공급계약·납품, M&A, 신사업·증설·신제품 등 성장 동인\n'
+            ' "배당"  = 배당 확대/축소, 자사주 매입·소각, 주주환원 정책\n'
+            ' "리스크" = 횡령·배임·소송·감사의견거절·부채급증·유상증자(희석)·경영진 돌발사임·규제/제재·상폐위험\n'
+            ' "모멘텀" = 본질가치 변화 없는 가격 동인(외국인/기관 수급, 목표주가 조정, 테마·정책 기대)\n'
+            ' "기타"  = 위 어디에도 명확히 속하지 않는 정보성 뉴스\n'
+            " 호재/악재면 catalyst를 '기타'로 두지 말고 실적·수주·배당·리스크·모멘텀 중 가장 가까운 것을 고르세요.\n\n"
+            "[참고] 블록은 보조 자료일 뿐 — 본 기사 내용으로 판단하고 과거 감성 흐름에 휩쓸리지 마세요.\n"
+            "동일 내용이 [참고]의 유사 과거뉴스에 반복되면 신규성이 낮으니 intensity를 보수적으로.\n\n"
+            "반드시 스키마에 맞는 유효한 JSON 객체 하나만 출력. 마크다운·설명문 금지."
+        )
+        user_prompt = (
+            f"[출처] {source}\n"
+            f"[제목] {data['title'][:200]}\n"
+            f"[내용] {content_preview}\n"
+            f"[관련종목] {stocks_str or '없음'}\n"
+            + (f"\n[참고]\n{context_block}\n" if context_block else "")
+            + "\ninvestment_action 규칙: 호재+intensity≥3→매수관심, 악재+intensity≥3→매도관심, 그 외→관망\n"
+            "primary_stock은 6자리 숫자코드(시장 전체 뉴스면 빈 문자열), "
+            "affected_stocks는 영향받는 다른 종목코드 배열, reason은 한 문장 핵심 근거.\n\n"
+            "JSON 스키마:\n"
+            '{"sentiment":"호재|악재|중립","intensity":1~5,"primary_stock":"005930",'
+            '"affected_stocks":["000660"],"reason":"핵심 근거 한 문장",'
+            '"investment_action":"매수관심|매도관심|관망",'
+            '"catalyst":"실적|수주|배당|리스크|모멘텀|기타"}\n'
+            "예시: "
+            '{"sentiment":"호재","intensity":4,"primary_stock":"005930",'
+            '"affected_stocks":["000660"],"reason":"HBM 대형 수주 확정으로 연간 반도체 실적 큰 폭 개선 기대",'
+            '"investment_action":"매수관심","catalyst":"수주"}'
+        )
+        # EXAONE 호출 + JSON 파싱 (실패 시 1회 재시도 → ~15% 유실 감소)
+        a = None
+        raw_prev = ""
+        for attempt in range(2):
+            msgs = [{"role":"system","content": system_prompt},
+                    {"role":"user","content": user_prompt}]
+            if attempt == 1:
+                msgs += [{"role":"assistant","content": raw_prev[:600]},
+                         {"role":"user","content":
+                          "위 응답이 유효한 JSON이 아닙니다. 설명·마크다운 없이 "
+                          "스키마에 맞는 JSON 객체 하나만 출력하세요."}]
+            try:
+                vr = await client.post(f"{OLLAMA_URL}/v1/chat/completions", json={
+                    "model":"exaone3.5:7.8b","messages": msgs,
+                    "max_tokens":400,"temperature":0.0 if attempt else 0.05},
+                    timeout=120)
+                raw_prev = vr.json()["choices"][0]["message"]["content"]
+                c = raw_prev.replace("```json","").replace("```","").strip()
+                if not c.startswith("{"):           # 앞뒤 설명 제거
+                    s, e = c.find("{"), c.rfind("}")
+                    if s != -1 and e > s: c = c[s:e+1]
+                a = json.loads(c)
+                break
+            except Exception:
+                continue
+        if a is None:
+            a = {"sentiment":"중립","intensity":0,"primary_stock":"","affected_stocks":[],"reason":"파싱실패","investment_action":"관망"}
+
+        # catalyst 6개 enum 강제 (score-engine CATALYST_WEIGHTS 정합 — 일탈값은 '기타')
+        if a.get("catalyst") not in ("실적","수주","배당","리스크","모멘텀","기타"):
+            a["catalyst"] = "기타"
+
+        # 5. Qdrant 저장
+        try:
+            await client.put(f"{QDRANT_URL}/collections/news_vectors/points", json={
+                "points":[{"id":random.randint(1,999999999),"vector":emb,
+                    "payload":{"title":data["title"],"hash":data["hash"],
+                               "sentiment":a.get("sentiment",""),"intensity":a.get("intensity",0),
+                               "primary_stock":a.get("primary_stock","")}}]}, timeout=15)
+        except: pass
+
+        # 6. PostgreSQL 저장 (파라미터 바인딩 — 인젝션·이스케이프 유실 방지)
+        pub = data.get("published_at") or ""
+        try:
+            pub_dt = datetime.fromisoformat(pub.replace("Z", "+00:00")) if pub else datetime.now()
+        except Exception:
+            pub_dt = datetime.now()
+        try:
+            iv = int(a.get("intensity", 0) or 0)
+        except Exception:
+            iv = 0
+        s = lambda v: str(v if v is not None else "")
+        await pg_pool.execute("""
+            INSERT INTO news_analysis (title,url,source,published_at,hash,sentiment,intensity,
+                primary_stock,affected_stocks,reason,investment_action,keywords,stock_names,stock_codes,
+                catalyst,similar_count,analyzed_at)
+            VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9::jsonb,$10,$11,$12::jsonb,$13::jsonb,$14::jsonb,$15,0,NOW())
+            ON CONFLICT (hash) DO NOTHING
+        """,
+            s(data.get("title"))[:500], s(data.get("url"))[:500], s(data.get("source"))[:100],
+            pub_dt, data["hash"], s(a.get("sentiment", "중립"))[:10], iv,
+            s(a.get("primary_stock"))[:10], json.dumps(a.get("affected_stocks", []), ensure_ascii=False),
+            s(a.get("reason"))[:500], s(a.get("investment_action", "관망"))[:10],
+            json.dumps(data.get("keywords", [])[:20], ensure_ascii=False),
+            json.dumps([s_["name"] for s_ in data.get("stocks", [])], ensure_ascii=False),
+            json.dumps([s_["code"] for s_ in data.get("stocks", [])], ensure_ascii=False),
+            s(a.get("catalyst", "기타"))[:20])
+        return "ok"
+    except Exception as e:
+        logger.warning("pipeline.err", title=item.get("title","")[:50], error=str(e))
+        return "error"
+
+# ── 수집 작업 ──────────────────────────────────────────────
+async def job_rss():
+    """RSS 28개 소스 수집 (5분마다, 2단계 중복제거)"""
+    if stats.running: return
+    stats.running = True
+    try:
+        async with httpx.AsyncClient() as c:
+            news = await crawl_rss_sources(c)
+            ok = 0
+            for item in news:
+                if await is_dup(item["title"], item.get("url", "")):
+                    stats.duplicates += 1
+                    continue
+                stats.collected += 1
+                r = await pipeline(item, c)
+                if r == "ok":
+                    ok += 1
+                    stats.processed += 1
+                elif r == "noise":
+                    stats.noise += 1
+                elif r == "off_topic":
+                    stats.off_topic += 1
+            stats.last_run = datetime.now().isoformat()
+            logger.info("job.rss", sources=len(RSS_SOURCES), total=len(news), processed=ok,
+                        duplicates=stats.duplicates)
+    except Exception as e:
+        stats.errors += 1
+        logger.error("job.rss.err", error=str(e))
+    finally:
+        stats.running = False
+
+NAVER_FINANCE_URLS = [
+    ("https://finance.naver.com/news/mainnews.naver", "네이버금융"),
+    ("https://finance.naver.com/news/news_list.nhn?mode=RANK&type=now", "네이버금융"),
+]
+
+async def crawl_naver_market(client) -> list:
+    news = []
+    for url, source in NAVER_FINANCE_URLS:
+        try:
+            r = await client.get(url, timeout=15, headers=HEADERS)
+            if r.status_code != 200:
+                continue
+            soup = BeautifulSoup(r.text, "html.parser")
+            anchors = (
+                soup.select("ul.newsList li a") or
+                soup.select(".headline_list li a") or
+                soup.select("a[href*='news_read']")
+            )
+            for a in anchors[:30]:
+                title = a.get_text(strip=True)
+                href = a.get("href", "")
+                if not href or not title or len(title) < 5:
+                    continue
+                if href.startswith("/"):
+                    href = f"https://finance.naver.com{href}"
+                if not is_korean(title):
+                    continue
+                news.append({"title": title, "url": href, "source": source,
+                             "content": "", "published_at": datetime.now().isoformat()})
+        except Exception as e:
+            logger.debug("naver.market.err", url=url, error=str(e))
+    logger.info("naver.market.crawled", items=len(news))
+    return news
+
+async def crawl_naver_stock_news(client, code: str, max_pages: int = 20, sleep_s: float = 0.3) -> list:
+    """네이버 모바일 종목 뉴스 API (JSON) — 백워드 페이징"""
+    items = []
+    for page in range(1, max_pages + 1):
+        url = f"https://m.stock.naver.com/api/news/stock/{code}?pageSize=20&page={page}"
+        try:
+            r = await client.get(url, headers=HEADERS, timeout=15)
+            if r.status_code != 200:
+                break
+            data = r.json()
+            if not isinstance(data, list) or not data:
+                break
+            page_items = data[0].get("items", []) if isinstance(data[0], dict) else []
+            if not page_items:
+                break
+            for it in page_items:
+                title = (it.get("title") or "").strip()
+                article_id = it.get("articleId") or ""
+                office_id = it.get("officeId") or ""
+                if not title or not article_id or not is_korean(title):
+                    continue
+                url_news = f"https://n.news.naver.com/article/{office_id}/{article_id}"
+                dt = it.get("datetime") or ""
+                pub = (f"{dt[:4]}-{dt[4:6]}-{dt[6:8]} {dt[8:10]}:{dt[10:12]}"
+                       if len(dt) >= 12 else "")
+                items.append({
+                    "title": title, "url": url_news,
+                    "source": it.get("officeName") or "네이버",
+                    "published_at": pub,
+                    "content": (it.get("body") or "")[:500],
+                })
+            await asyncio.sleep(sleep_s)
+        except Exception as e:
+            logger.debug("naver_mobile.err", code=code, page=page, error=str(e))
+            break
+    return items
+
+
+async def save_raw_news(items: list, stock_code: str) -> int:
+    """raw 뉴스 일괄 저장 (분석 스킵, 빠른 누적용)"""
+    if not items: return 0
+    saved = 0
+    async with pg_pool.acquire() as conn:
+        for it in items:
+            try:
+                url = it.get("url", "")
+                title = it.get("title", "")
+                if not url or not title: continue
+                url_hash = hashlib.sha256(url.encode()).hexdigest()
+                res = await conn.execute("""
+                    INSERT INTO news_raw (stock_code, title, url, url_hash, source, published_at_text)
+                    VALUES ($1,$2,$3,$4,$5,$6) ON CONFLICT (url_hash) DO NOTHING
+                """, stock_code, title, url, url_hash,
+                    it.get("source", ""), it.get("published_at", ""))
+                if "INSERT 0 1" in res:
+                    saved += 1
+            except Exception as e:
+                logger.debug("raw.save_err", error=str(e))
+    return saved
+
+
+async def job_historical_raw(count: int = 0, max_pages: int = 100):
+    """전체 활성종목 백워드 크롤링 → news_raw에 raw 저장 (LLM 분석 X, 빠름)
+    count=0 → 활성 종목 전체"""
+    if stats.running:
+        logger.warning("historical_raw.skip_running"); return
+    stats.running = True
+    try:
+        async with pg_pool.acquire() as conn:
+            if count > 0:
+                rows = await conn.fetch("""
+                    SELECT stock_code FROM dart_corps WHERE is_active=true
+                    ORDER BY stock_code LIMIT $1
+                """, count)
+            else:
+                rows = await conn.fetch(
+                    "SELECT stock_code FROM dart_corps WHERE is_active=true ORDER BY stock_code")
+        codes = [r["stock_code"] for r in rows]
+        logger.info("historical_raw.start", n_stocks=len(codes), max_pages=max_pages)
+        total_saved = 0
+        async with httpx.AsyncClient() as c:
+            for i, code in enumerate(codes):
+                try:
+                    items = await crawl_naver_stock_news(c, code, max_pages=max_pages)
+                    saved = await save_raw_news(items, code)
+                    total_saved += saved
+                    if (i + 1) % 100 == 0:
+                        logger.info("historical_raw.progress",
+                                    done=i+1, of=len(codes), total_saved=total_saved)
+                    await asyncio.sleep(0.4)
+                except Exception as e:
+                    stats.errors += 1
+                    logger.warning("historical_raw.stock_err", code=code, error=str(e))
+        stats.last_run = datetime.now().isoformat()
+        logger.info("historical_raw.done", n_stocks=len(codes), total_saved=total_saved)
+    finally:
+        stats.running = False
+
+
+async def job_process_raw(batch_size: int = 200):
+    """news_raw의 미처리 행을 batch_size만큼 EXAONE 분석 → news_analysis로 이동"""
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT id, stock_code, title, url, source, published_at_text
+            FROM news_raw WHERE processed=false
+            ORDER BY id DESC LIMIT $1
+        """, batch_size)
+    if not rows:
+        logger.info("process_raw.empty"); return 0
+    ok = 0
+    async with httpx.AsyncClient() as c:
+        for r in rows:
+            item = {
+                "title": r["title"], "url": r["url"],
+                "source": r["source"] or "네이버금융",
+                "published_at": r["published_at_text"] or "",
+                "content": "",
+                "stock_code": r["stock_code"],
+            }
+            try:
+                if await is_dup(item["title"], item["url"]):
+                    stats.duplicates += 1
+                else:
+                    stats.collected += 1
+                    res = await pipeline(item, c)
+                    if res == "ok": ok += 1; stats.processed += 1
+                    elif res == "noise": stats.noise += 1
+                    elif res == "off_topic": stats.off_topic += 1
+                async with pg_pool.acquire() as conn:
+                    await conn.execute(
+                        "UPDATE news_raw SET processed=true, processed_at=NOW() WHERE id=$1",
+                        r["id"])
+            except Exception as e:
+                stats.errors += 1
+                logger.warning("process_raw.err", id=r["id"], error=str(e))
+    logger.info("process_raw.batch_done", batch=len(rows), processed_ok=ok)
+    return ok
+
+
+async def job_historical(count: int = 500, max_pages: int = 20):
+    """과거 뉴스 백필: stock_scores 점수 매겨진 활성 종목 우선 + 종목당 max_pages 페이지"""
+    if stats.running:
+        logger.warning("historical.skip_running")
+        return
+    stats.running = True
+    try:
+        async with pg_pool.acquire() as conn:
+            # 우선순위: 최근 점수 매겨진 종목 → 그 외 활성 종목 alphabetical
+            rows = await conn.fetch("""
+                WITH scored AS (
+                    SELECT DISTINCT stock_code, MAX(score_date) AS last_d
+                    FROM stock_scores GROUP BY stock_code
+                ), priority AS (
+                    SELECT d.stock_code,
+                           CASE WHEN s.stock_code IS NOT NULL THEN 0 ELSE 1 END AS p,
+                           COALESCE(s.last_d, '1900-01-01'::date) AS d
+                    FROM dart_corps d LEFT JOIN scored s ON s.stock_code=d.stock_code
+                    WHERE d.is_active=true
+                )
+                SELECT stock_code FROM priority
+                ORDER BY p ASC, d DESC, stock_code ASC LIMIT $1
+            """, count)
+        codes = [r["stock_code"] for r in rows]
+        logger.info("historical.start", n_stocks=len(codes), max_pages=max_pages)
+        ok = 0; total_items = 0
+        async with httpx.AsyncClient() as c:
+            for i, code in enumerate(codes):
+                try:
+                    items = await crawl_naver_stock_news(c, code, max_pages=max_pages)
+                    total_items += len(items)
+                    for item in items:
+                        if await is_dup(item["title"], item.get("url", "")):
+                            stats.duplicates += 1
+                            continue
+                        stats.collected += 1
+                        item["stock_code"] = code
+                        r = await pipeline(item, c)
+                        if r == "ok":
+                            ok += 1; stats.processed += 1
+                        elif r == "noise": stats.noise += 1
+                        elif r == "off_topic": stats.off_topic += 1
+                    if (i + 1) % 50 == 0:
+                        logger.info("historical.progress", done=i+1, of=len(codes),
+                                    items=total_items, processed=ok)
+                    await asyncio.sleep(0.5)  # 종목 간격 (차단 회피)
+                except Exception as e:
+                    stats.errors += 1
+                    logger.warning("historical.stock_err", code=code, error=str(e))
+        stats.last_run = datetime.now().isoformat()
+        logger.info("historical.done", n_stocks=len(codes), items=total_items, processed=ok)
+    finally:
+        stats.running = False
+
+
+async def job_market():
+    if stats.running:
+        return
+    stats.running = True
+    try:
+        async with httpx.AsyncClient() as c:
+            news = await crawl_naver_market(c)
+            ok = 0
+            for item in news:
+                if await is_dup(item["title"], item.get("url", "")):
+                    stats.duplicates += 1
+                    continue
+                stats.collected += 1
+                r = await pipeline(item, c)
+                if r == "ok":
+                    ok += 1
+                    stats.processed += 1
+                elif r == "noise":
+                    stats.noise += 1
+                elif r == "off_topic":
+                    stats.off_topic += 1
+            stats.last_run = datetime.now().isoformat()
+            logger.info("job.market", total=len(news), processed=ok, duplicates=stats.duplicates)
+    except Exception as e:
+        stats.errors += 1
+        logger.error("job.market.err", error=str(e))
+    finally:
+        stats.running = False
+
+# ── FastAPI ────────────────────────────────────────────────
+app = FastAPI(title="뉴스 수집기 (멀티소스)")
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+async def init_news_tables():
+    """news_raw 테이블 (수집/분석 분리용 임시 저장소)"""
+    async with pg_pool.acquire() as conn:
+        await conn.execute("""
+            CREATE TABLE IF NOT EXISTS news_raw (
+                id SERIAL PRIMARY KEY,
+                stock_code VARCHAR(10),
+                title TEXT NOT NULL,
+                url TEXT NOT NULL,
+                url_hash VARCHAR(64) UNIQUE,
+                source VARCHAR(100),
+                published_at_text VARCHAR(50),
+                collected_at TIMESTAMP DEFAULT NOW(),
+                processed BOOLEAN DEFAULT FALSE,
+                processed_at TIMESTAMP
+            )
+        """)
+        await conn.execute(
+            "CREATE INDEX IF NOT EXISTS idx_news_raw_unprocessed ON news_raw(processed) WHERE processed=false")
+        await conn.execute(
+            "CREATE INDEX IF NOT EXISTS idx_news_raw_stock ON news_raw(stock_code)")
+
+
+@app.on_event("startup")
+async def startup():
+    global pg_pool, redis_cl
+    pg_pool = await asyncpg.create_pool(host=PG_HOST,port=PG_PORT,database=PG_DB,user=PG_USER,password=PG_PASS,min_size=2,max_size=5)
+    redis_cl = aioredis.Redis(host=REDIS_HOST,port=6379,password=REDIS_PASSWORD,db=4,decode_responses=True)
+    await init_news_tables()
+    # 평일 RSS: 8-18시 5분마다
+    scheduler.add_job(job_rss,"cron",day_of_week="mon-fri",hour="8-18",minute="*/5",id="rss_weekday",replace_existing=True)
+    # 주말 RSS: 8-22시 15분마다 (누적 학습용)
+    scheduler.add_job(job_rss,"cron",day_of_week="sat,sun",hour="8-22",minute="*/15",id="rss_weekend",replace_existing=True)
+    # 평일 마켓 (네이버 금융): 9-17시 10분마다
+    scheduler.add_job(job_market,"cron",day_of_week="mon-fri",hour="9-17",minute="*/10",id="market",replace_existing=True)
+    # raw 뉴스 분석: 24시간 30분마다 batch 200건 처리 (백로그 소화용)
+    scheduler.add_job(job_process_raw,"cron",hour="*",minute="*/30",
+                      id="process_raw",replace_existing=True,kwargs={"batch_size":200})
+    # 매주 일요일 새벽 2시 raw 백필 (전체 활성종목)
+    scheduler.add_job(job_historical_raw,"cron",day_of_week="sun",hour="2",minute="0",
+                      id="historical_raw_weekly",replace_existing=True,
+                      kwargs={"count":0,"max_pages":50})
+    scheduler.start()
+    logger.info("news-collector.started", sources=len(RSS_SOURCES))
+
+@app.on_event("shutdown")
+async def shutdown():
+    scheduler.shutdown()
+    if pg_pool: await pg_pool.close()
+    if redis_cl: await redis_cl.aclose()
+
+@app.get("/health")
+async def health():
+    return JSONResponse(content={"status":"ok","collected":stats.collected,"processed":stats.processed,
+        "duplicates":stats.duplicates,"noise":stats.noise,"off_topic":stats.off_topic,
+        "errors":stats.errors,"last_run":stats.last_run,"running":stats.running})
+
+@app.post("/collect/rss")
+async def m_rss():
+    asyncio.create_task(job_rss()); return {"status":"started","type":"rss","sources":len(RSS_SOURCES)}
+
+@app.post("/collect/market")
+async def m_market():
+    asyncio.create_task(job_market()); return {"status":"started","type":"market"}
+
+@app.post("/collect/historical")
+async def m_historical(count: int = 500, max_pages: int = 20):
+    asyncio.create_task(job_historical(count, max_pages))
+    return {"status":"started","type":"historical","count":count,"max_pages":max_pages}
+
+@app.post("/collect/historical-raw")
+async def m_historical_raw(count: int = 0, max_pages: int = 100):
+    """전체 활성종목 raw 백필 (LLM 분석 X). count=0 → 전체"""
+    asyncio.create_task(job_historical_raw(count, max_pages))
+    return {"status":"started","type":"historical_raw","count":count or "전체","max_pages":max_pages}
+
+@app.post("/process/raw")
+async def m_process_raw(batch_size: int = 200):
+    """news_raw 미처리분 batch 분석 → news_analysis로 이동"""
+    n = await job_process_raw(batch_size)
+    return {"status":"done","processed":n}
+
+@app.get("/raw/stats")
+async def raw_stats():
+    """raw 백필 진행 상태"""
+    async with pg_pool.acquire() as conn:
+        s = await conn.fetchrow("""
+            SELECT
+              COUNT(*) AS total,
+              COUNT(*) FILTER (WHERE processed=false) AS unprocessed,
+              COUNT(*) FILTER (WHERE processed=true) AS processed,
+              COUNT(DISTINCT stock_code) AS unique_stocks,
+              MIN(collected_at) AS first_collected,
+              MAX(collected_at) AS last_collected
+            FROM news_raw
+        """)
+    return dict(s) if s else {}
+
+@app.get("/sources")
+async def list_sources():
+    return [{"name": s, "url": u} for s, u in RSS_SOURCES]
@@ -0,0 +1,10 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+redis==5.0.4
+asyncpg==0.29.0
+beautifulsoup4==4.12.3
+lxml==5.2.2
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
@@ -0,0 +1,19 @@
+service:
+  http_port: 6333
+  grpc_port: 6334
+  max_request_size_mb: 32
+
+storage:
+  performance:
+    max_search_threads: 4
+    max_optimization_threads: 2
+
+  optimizers:
+    deleted_threshold: 0.2
+    vacuum_min_vector_number: 1000
+    default_segment_number: 4
+    indexing_threshold_kb: 20000
+    flush_interval_sec: 5
+    max_optimization_threads: 2
+
+log_level: INFO
@@ -0,0 +1,10 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+redis==5.0.4
+asyncpg==0.29.0
+beautifulsoup4==4.12.3
+lxml==5.2.2
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+EXPOSE 8686
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8686", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,13 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+asyncpg==0.29.0
+redis==5.0.4
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
+scikit-learn==1.5.0
+numpy==1.26.4
+xgboost==2.0.3
+arch==7.0.0
+scipy==1.13.1
@@ -0,0 +1,146 @@
+#!/bin/bash
+# PostgreSQL 로컬 설치 + 데이터 경로를 시놀로지 NAS(/mnt/nas)에 설정
+# 실행: sudo bash setup-postgres-local.sh
+
+set -e
+
+NAS_HOST="192.168.0.36"
+NAS_NFS_PATH="/volume1/trading"
+NFS_MOUNT_LOCAL="/mnt/nas"
+PG_DATA_DIR="${NFS_MOUNT_LOCAL}/postgresql/data"
+PG_VERSION="16"
+PG_USER="kyu"
+PG_PASSWORD="7895123"
+PG_DB="trading_ai"
+
+echo "=============================="
+echo "1. NFS 클라이언트 설치 확인"
+echo "=============================="
+apt-get install -y nfs-common 2>/dev/null || true
+
+echo ""
+echo "=============================="
+echo "2. NAS NFS 마운트"
+echo "=============================="
+mkdir -p "${NFS_MOUNT_LOCAL}"
+
+# 이미 마운트 되어있으면 스킵
+if mountpoint -q "${NFS_MOUNT_LOCAL}"; then
+    echo "이미 마운트됨: ${NFS_MOUNT_LOCAL}"
+else
+    mount -t nfs "${NAS_HOST}:${NAS_NFS_PATH}" "${NFS_MOUNT_LOCAL}" -o rw,hard,intr,timeo=30,retrans=3
+    echo "마운트 완료: ${NAS_HOST}:${NAS_NFS_PATH} -> ${NFS_MOUNT_LOCAL}"
+fi
+
+# /etc/fstab에 자동 마운트 등록 (없으면 추가)
+FSTAB_ENTRY="${NAS_HOST}:${NAS_NFS_PATH} ${NFS_MOUNT_LOCAL} nfs rw,hard,intr,timeo=30,retrans=3,_netdev 0 0"
+if ! grep -qF "${NFS_MOUNT_LOCAL}" /etc/fstab; then
+    echo "${FSTAB_ENTRY}" >> /etc/fstab
+    echo "fstab에 자동 마운트 추가됨"
+fi
+
+echo ""
+echo "=============================="
+echo "3. PostgreSQL 데이터 디렉토리 생성"
+echo "=============================="
+mkdir -p "${PG_DATA_DIR}"
+chown postgres:postgres "${PG_DATA_DIR}"
+chmod 700 "${PG_DATA_DIR}"
+echo "데이터 경로: ${PG_DATA_DIR}"
+
+echo ""
+echo "=============================="
+echo "4. PostgreSQL 서비스 설정"
+echo "=============================="
+
+# 기존 클러스터 중지 및 제거
+pg_ctlcluster ${PG_VERSION} main stop 2>/dev/null || true
+
+# 기존 기본 데이터 디렉토리를 NAS 경로로 교체
+CONF_FILE="/etc/postgresql/${PG_VERSION}/main/postgresql.conf"
+if [ -f "${CONF_FILE}" ]; then
+    # data_directory 변경
+    sed -i "s|^data_directory.*|data_directory = '${PG_DATA_DIR}'|" "${CONF_FILE}"
+    # 이미 변경된 경우가 아니면 추가
+    if ! grep -q "^data_directory" "${CONF_FILE}"; then
+        echo "data_directory = '${PG_DATA_DIR}'" >> "${CONF_FILE}"
+    fi
+    echo "postgresql.conf 업데이트 완료"
+fi
+
+echo ""
+echo "=============================="
+echo "5. 데이터 디렉토리 초기화 (최초 1회)"
+echo "=============================="
+if [ ! -f "${PG_DATA_DIR}/PG_VERSION" ]; then
+    sudo -u postgres /usr/lib/postgresql/${PG_VERSION}/bin/initdb \
+        -D "${PG_DATA_DIR}" \
+        --encoding=UTF8 \
+        --locale=ko_KR.UTF-8 \
+        --auth=md5 2>/dev/null || \
+    sudo -u postgres /usr/lib/postgresql/${PG_VERSION}/bin/initdb \
+        -D "${PG_DATA_DIR}" \
+        --encoding=UTF8 \
+        --locale=C.UTF-8 \
+        --auth=md5
+    echo "DB 초기화 완료"
+else
+    echo "이미 초기화된 데이터 디렉토리 (스킵)"
+fi
+
+echo ""
+echo "=============================="
+echo "6. 외부 접속 허용 설정"
+echo "=============================="
+# postgresql.conf - 모든 인터페이스 리슨
+sed -i "s|^#listen_addresses.*|listen_addresses = '*'|; s|^listen_addresses.*|listen_addresses = '*'|" "${PG_DATA_DIR}/postgresql.conf" 2>/dev/null || \
+    echo "listen_addresses = '*'" >> "${PG_DATA_DIR}/postgresql.conf"
+
+# pg_hba.conf - Docker 네트워크(172.30.0.0/16) 접속 허용
+PG_HBA="${PG_DATA_DIR}/pg_hba.conf"
+if ! grep -q "172.30.0.0/16" "${PG_HBA}" 2>/dev/null; then
+    cat >> "${PG_HBA}" << 'HBA'
+# Trading AI Docker 네트워크
+host    all             all             172.30.0.0/16           md5
+host    all             all             127.0.0.1/32            md5
+host    all             all             ::1/128                 md5
+HBA
+    echo "pg_hba.conf 업데이트 완료"
+fi
+
+echo ""
+echo "=============================="
+echo "7. PostgreSQL 서비스 시작"
+echo "=============================="
+systemctl start postgresql
+systemctl enable postgresql
+sleep 3
+systemctl is-active postgresql && echo "PostgreSQL 실행 중"
+
+echo ""
+echo "=============================="
+echo "8. 유저 및 DB 생성"
+echo "=============================="
+sudo -u postgres psql -c "DO \$\$ BEGIN
+  IF NOT EXISTS (SELECT FROM pg_roles WHERE rolname = '${PG_USER}') THEN
+    CREATE USER ${PG_USER} WITH SUPERUSER PASSWORD '${PG_PASSWORD}';
+  ELSE
+    ALTER USER ${PG_USER} WITH PASSWORD '${PG_PASSWORD}';
+  END IF;
+END \$\$;"
+
+sudo -u postgres psql -c "CREATE DATABASE ${PG_DB} OWNER ${PG_USER};" 2>/dev/null || \
+    echo "DB ${PG_DB} 이미 존재 (스킵)"
+
+echo ""
+echo "=============================="
+echo "완료! PostgreSQL 로컬 실행 중"
+echo "  Host: localhost (또는 127.0.0.1)"
+echo "  Port: 5432"
+echo "  DB:   ${PG_DB}"
+echo "  User: ${PG_USER}"
+echo "  Data: ${PG_DATA_DIR} (NAS)"
+echo "=============================="
+echo ""
+echo "다음 단계: 덤프 임포트"
+echo "  PGPASSWORD=${PG_PASSWORD} psql -h 127.0.0.1 -U ${PG_USER} -d ${PG_DB} -f /home/kyu/trading/pg_backup/trading_ai_backup.sql"
@@ -0,0 +1,111 @@
+#!/usr/bin/env bash
+# ============================================================
+# Trading AI - Setup Script
+# ============================================================
+set -euo pipefail
+
+RED='\033[0;31m'; GREEN='\033[0;32m'; YELLOW='\033[1;33m'; NC='\033[0m'
+log()  { echo -e "${GREEN}[✓]${NC} $1"; }
+warn() { echo -e "${YELLOW}[!]${NC} $1"; }
+err()  { echo -e "${RED}[✗]${NC} $1" >&2; exit 1; }
+
+[ -f .env ] && source .env || err ".env 파일이 없습니다."
+
+echo "============================================================"
+echo "  Trading AI System - Setup"
+echo "============================================================"
+
+# ── 1. Docker / NVIDIA 확인 ────────────────────────────────
+command -v docker >/dev/null 2>&1 || err "Docker가 없습니다."
+
+if ! command -v nvidia-smi >/dev/null 2>&1; then
+    err "NVIDIA 드라이버가 없습니다."
+fi
+
+if ! docker info 2>/dev/null | grep -q "Runtimes.*nvidia"; then
+    warn "nvidia-container-toolkit 설치 중..."
+    curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
+        sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+    curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
+        sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
+        sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
+    sudo apt-get update -qq
+    sudo apt-get install -y nvidia-container-toolkit
+    sudo nvidia-ctk runtime configure --runtime=docker
+    sudo systemctl restart docker
+    log "nvidia-container-toolkit 설치 완료"
+fi
+
+log "GPU 정보:"
+nvidia-smi --query-gpu=index,name,memory.total --format=csv,noheader | \
+    while IFS=, read -r idx name mem; do echo "  GPU $idx: $name ($mem)"; done
+
+# ── 2. 시스템 최적화 ────────────────────────────────────────
+if ! grep -q "vm.overcommit_memory" /etc/sysctl.conf 2>/dev/null; then
+    printf "vm.overcommit_memory = 1\nvm.swappiness = 10\nnet.core.somaxconn = 65535\n" | \
+        sudo tee -a /etc/sysctl.conf
+    sudo sysctl -p >/dev/null 2>&1
+    log "커널 파라미터 최적화"
+fi
+echo never | sudo tee /sys/kernel/mm/transparent_hugepage/enabled >/dev/null
+
+# ── 3. NFS 마운트 (스킵 - 나중에 수동 설정) ───────────────
+warn "NFS 마운트 스킵 (Synology 설정 후 수동으로 진행하세요)"
+sudo mkdir -p /mnt/nas/news /mnt/nas/models /mnt/nas/backups
+# ── 4. 기존 네트워크 정리 ──────────────────────────────────
+log "기존 컨테이너/네트워크 정리 중..."
+docker compose down --remove-orphans 2>/dev/null || true
+docker network prune -f >/dev/null 2>&1 || true
+
+# ── 5. 순차 시작 ───────────────────────────────────────────
+log "1단계: Redis, Qdrant, Bareun 시작..."
+docker compose up -d --build redis qdrant bareun
+
+log "헬스체크 대기 (20초)..."
+sleep 20
+
+log "2단계: bareunaapi 시작..."
+docker compose up -d --build bareunaapi
+
+log "3단계: Ollama 시작 (GPU 1 - RTX 3070)..."
+docker compose up -d ollama
+
+log "BGE-M3 모델 다운로드 중..."
+sleep 30
+docker exec trading-ollama ollama pull bge-m3 2>/dev/null || \
+    warn "BGE-M3 수동 다운로드 필요: docker exec trading-ollama ollama pull bge-m3"
+
+log "4단계: vLLM 시작 (GPU 0 - RTX 3060, 모델 로딩 2~5분)..."
+docker compose up -d vllm
+
+log "5단계: n8n 시작..."
+docker compose up -d n8n n8n-worker
+
+# ── 6. Qdrant 컬렉션 초기화 ────────────────────────────────
+log "Qdrant 컬렉션 초기화 중..."
+sleep 10
+HTTP=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:6333/collections/news_vectors)
+if [ "${HTTP}" = "404" ]; then
+    curl -s -X PUT http://localhost:6333/collections/news_vectors \
+        -H "Content-Type: application/json" \
+        -d '{
+            "vectors": {"size": 1024, "distance": "Cosine"},
+            "optimizers_config": {"default_segment_number": 4},
+            "on_disk_payload": false
+        }' >/dev/null
+    log "Qdrant 컬렉션 'news_vectors' 생성"
+else
+    warn "Qdrant 컬렉션 이미 존재"
+fi
+
+echo ""
+echo "============================================================"
+echo -e "${GREEN}  시작 완료!${NC}"
+echo "============================================================"
+echo "  n8n      : http://localhost:5678"
+echo "  Qdrant   : http://localhost:6333/dashboard"
+echo "  vLLM     : http://localhost:8000/docs"
+echo "  Ollama   : http://localhost:11434"
+echo "  바른API  : http://localhost:5757/docs"
+echo "============================================================"
+warn "vLLM 모델 로딩 확인: docker compose logs -f vllm"
@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+[ -f .env ] && source .env
+GREEN='\033[0;32m'; RED='\033[0;31m'; YELLOW='\033[1;33m'; BLUE='\033[0;34m'; NC='\033[0m'
+
+chk_http() {
+    local name=$1 url=$2
+    code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 4 "$url" 2>/dev/null || echo "000")
+    [[ "$code" =~ ^(200|204)$ ]] \
+        && echo -e "  ${GREEN}●${NC} $name" \
+        || echo -e "  ${RED}●${NC} $name (HTTP $code)"
+}
+
+chk_container() {
+    local name=$1
+    status=$(docker inspect --format='{{.State.Status}}' "$name" 2>/dev/null || echo "없음")
+    [ "$status" = "running" ] \
+        && echo -e "  ${GREEN}●${NC} $name" \
+        || echo -e "  ${RED}●${NC} $name ($status)"
+}
+
+echo ""
+echo -e "${BLUE}══ Trading AI Status ══════════════════════${NC}  $(date '+%Y-%m-%d %H:%M:%S')"
+
+echo -e "\n${BLUE}[컨테이너]${NC}"
+for c in trading-redis trading-qdrant trading-bareun trading-bareunaapi \
+          trading-ollama trading-vllm trading-n8n trading-n8n-worker; do
+    chk_container "$c"
+done
+
+echo -e "\n${BLUE}[서비스]${NC}"
+chk_http "Redis"    "http://localhost:6379"
+chk_http "Qdrant"   "http://localhost:6333/healthz"
+chk_http "Bareun"   "http://localhost:9902/health"
+chk_http "바른API"  "http://localhost:5757/health"
+chk_http "Ollama"   "http://localhost:11434/api/tags"
+chk_http "vLLM"     "http://localhost:8000/health"
+chk_http "n8n"      "http://localhost:5678/healthz"
+
+echo -e "\n${BLUE}[GPU]${NC}"
+nvidia-smi --query-gpu=index,name,utilization.gpu,memory.used,memory.total,temperature.gpu \
+    --format=csv,noheader 2>/dev/null | \
+    while IFS=, read -r i n u mu mt t; do
+        echo "  GPU$i $n | 사용률:$u | VRAM:$mu/$mt | 온도:$t"
+    done
+
+echo -e "\n${BLUE}[NFS]${NC}"
+mountpoint -q /mnt/nas 2>/dev/null \
+    && echo -e "  ${GREEN}●${NC} /mnt/nas (마운트됨)" \
+    || echo -e "  ${RED}●${NC} /mnt/nas (마운트 안됨)"
+
+echo ""
@@ -0,0 +1,84 @@
+#!/usr/bin/env bash
+# ============================================================
+# Phase 1~4 전체 설치 스크립트
+# ============================================================
+set -euo pipefail
+
+GREEN='\033[0;32m'; NC='\033[0m'
+log() { echo -e "${GREEN}[✓]${NC} $1"; }
+
+cd ~/trading
+
+# ── 1. 디렉토리 생성 ───────────────────────────────────
+log "디렉토리 생성..."
+mkdir -p news-collector kis-api score-engine dashboard-api
+
+# ── 2. .env 추가 ────────────────────────────────────────
+log ".env 업데이트..."
+grep -q "KIS_APP_KEY" .env 2>/dev/null || cat >> .env << 'EOF'
+
+# Phase 2: 한국투자증권 (없으면 빈값)
+KIS_APP_KEY=
+KIS_APP_SECRET=
+KIS_ACCOUNT_NO=
+KIS_IS_PAPER=true
+EOF
+
+# ── 3. 빌드 및 시작 ─────────────────────────────────────
+log "Phase 1: 뉴스 수집기 빌드..."
+docker compose up -d --build news-collector
+
+log "Phase 2: KIS API 빌드..."
+docker compose up -d --build kis-api
+
+log "Phase 3: 점수 엔진 빌드..."
+docker compose up -d --build score-engine
+
+log "Phase 4: 대시보드 API 빌드..."
+docker compose up -d --build dashboard-api
+
+# ── 4. 헬스체크 ──────────────────────────────────────────
+log "서비스 시작 대기 (30초)..."
+sleep 30
+
+echo ""
+echo "============================================================"
+echo -e "${GREEN}  Phase 1~4 설치 완료!${NC}"
+echo "============================================================"
+echo ""
+echo "  서비스 목록:"
+echo "  ─────────────────────────────────────────"
+echo "  뉴스 수집기    : http://192.168.0.60:8787"
+echo "  KIS 주가 API   : http://192.168.0.60:8585"
+echo "  점수 엔진      : http://192.168.0.60:8686"
+echo "  대시보드 API   : http://192.168.0.60:8989"
+echo "  DART 수집기    : http://192.168.0.60:8888"
+echo "  n8n            : http://192.168.0.60:5678"
+echo "  바른API        : http://192.168.0.60:5757"
+echo "  vLLM           : http://192.168.0.60:8000"
+echo ""
+echo "  사용법:"
+echo "  ─────────────────────────────────────────"
+echo "  # 뉴스 수동 수집"
+echo "  curl -s -X POST http://localhost:8787/collect/market"
+echo "  curl -s -X POST http://localhost:8787/collect/stocks"
+echo ""
+echo "  # 점수 수동 산출"
+echo "  curl -s -X POST http://localhost:8686/score/calculate"
+echo ""
+echo "  # 추천 종목 조회"
+echo "  curl -s http://localhost:8686/recommendations | python3 -m json.tool"
+echo ""
+echo "  # 종목 랭킹"
+echo "  curl -s http://localhost:8686/ranking | python3 -m json.tool"
+echo ""
+echo "  # 종목 상세 (삼성전자)"
+echo "  curl -s http://localhost:8686/stock/005930 | python3 -m json.tool"
+echo ""
+echo "  # 대시보드 API"
+echo "  curl -s http://localhost:8989/api/summary | python3 -m json.tool"
+echo "  curl -s http://localhost:8989/api/ranking | python3 -m json.tool"
+echo "  curl -s http://localhost:8989/api/recommendations | python3 -m json.tool"
+echo "============================================================"
+
+docker ps -a --format "table {{.Names}}\t{{.Status}}"
@@ -0,0 +1,162 @@
+"""
+KRX 전체 종목 동적 로딩
+- 서버 시작 시 1회 로딩
+- 매일 자정 자동 갱신
+- 실패 시 하드코딩 폴백
+"""
+
+import asyncio
+import json
+import io
+import time
+from datetime import datetime
+
+import httpx
+import structlog
+
+logger = structlog.get_logger()
+
+# 폴백 종목 (네트워크 실패 시 사용)
+FALLBACK_STOCKS = {
+    "코스피": "KOSPI", "코스닥": "KOSDAQ", "코스피200": "KOSPI200",
+    "삼성전자": "005930", "SK하이닉스": "000660", "LG에너지솔루션": "373220",
+    "현대차": "005380", "기아": "000270", "셀트리온": "068270",
+    "카카오": "035720", "네이버": "035420", "NAVER": "035420",
+    "삼성바이오로직스": "207940", "KB금융": "105560", "POSCO홀딩스": "005490",
+    "신한지주": "055550", "LG화학": "051910", "삼성SDI": "006400",
+    "현대모비스": "012330", "하나금융지주": "086790", "SK텔레콤": "017670",
+    "KT": "030200", "LG전자": "066570", "한화에어로스페이스": "012450",
+    "삼성물산": "028260", "HD현대중공업": "329180", "한국전력": "015760",
+    "HMM": "011200", "대한항공": "003490", "카카오뱅크": "323410",
+    "크래프톤": "259960", "하이브": "352820", "에코프로": "086520",
+    "알테오젠": "196170", "한미반도체": "042700",
+}
+
+# 약칭/별칭
+ALIASES = {
+    "삼전": "005930", "하닉": "000660", "현차": "005380",
+    "카뱅": "323410", "삼바": "207940", "삼성바이오": "207940",
+    "한에솔": "012450", "한화에어": "012450", "LG엔솔": "373220",
+    "SK하닉": "000660", "포홀": "005490", "에프엔에프": "383220",
+    "현대자동차": "005380", "네이버": "035420",
+}
+
+
+async def fetch_krx_stocks() -> dict[str, str]:
+    """KRX에서 전체 상장 종목 가져오기"""
+    stock_map = {
+        "코스피": "KOSPI", "코스닥": "KOSDAQ",
+        "코스피200": "KOSPI200", "KRX": "KRX",
+    }
+
+    headers = {
+        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36",
+        "Referer": "http://data.krx.co.kr/contents/MDC/MDI/mdiLoader/index.cmd",
+    }
+
+    async with httpx.AsyncClient(timeout=30, headers=headers) as client:
+        for mkt_id, mkt_name in [("STK", "KOSPI"), ("KSQ", "KOSDAQ")]:
+            try:
+                payload = {
+                    "bld": "dbms/MDC/STAT/standard/MDCSTAT01901",
+                    "locale": "ko_KR",
+                    "mktId": mkt_id,
+                    "share": "1",
+                    "csvxls_is498": "false",
+                }
+                resp = await client.post(
+                    "http://data.krx.co.kr/comm/bldAttend/getJsonData.cmd",
+                    data=payload,
+                )
+                if resp.status_code == 200:
+                    data = resp.json()
+                    items = data.get("OutBlock_1", [])
+                    for item in items:
+                        name = item.get("ISU_ABBRV", "").strip()
+                        code = item.get("ISU_SRT_CD", "").strip()
+                        if name and code and len(code) == 6:
+                            stock_map[name] = code
+                    logger.info("krx.loaded", market=mkt_name, count=len(items))
+                else:
+                    logger.warning("krx.http_error", market=mkt_name, status=resp.status_code)
+            except Exception as e:
+                logger.warning("krx.fetch_failed", market=mkt_name, error=str(e))
+
+    # KRX 실패 시 네이버 금융 폴백
+    if len(stock_map) < 100:
+        logger.info("krx.fallback_naver", reason="KRX returned too few stocks")
+        stock_map.update(await fetch_naver_stocks(client=None))
+
+    return stock_map
+
+
+async def fetch_naver_stocks() -> dict[str, str]:
+    """네이버 금융에서 종목 가져오기 (KRX 폴백)"""
+    stock_map = {}
+    headers = {"User-Agent": "Mozilla/5.0"}
+    import re
+
+    async with httpx.AsyncClient(timeout=15, headers=headers) as client:
+        for sosok in [0, 1]:  # 0=KOSPI, 1=KOSDAQ
+            for page in range(1, 45):
+                try:
+                    url = f"https://finance.naver.com/sise/sise_market_sum.naver?sosok={sosok}&page={page}"
+                    resp = await client.get(url)
+                    text = resp.content.decode("euc-kr", errors="ignore")
+                    rows = re.findall(
+                        r"main\.naver\?code=(\d{6})[^>]*>([^<]+)</a>", text
+                    )
+                    if not rows:
+                        break
+                    for code, name in rows:
+                        name = name.strip()
+                        if name and code:
+                            stock_map[name] = code
+                except Exception:
+                    break
+
+    logger.info("naver.loaded", count=len(stock_map))
+    return stock_map
+
+
+async def load_all_stocks() -> dict[str, str]:
+    """전체 종목 로딩 (KRX → 네이버 → 폴백)"""
+    try:
+        stock_map = await fetch_krx_stocks()
+        if len(stock_map) > 100:
+            stock_map.update(ALIASES)
+            logger.info("stocks.loaded", total=len(stock_map), source="KRX")
+            return stock_map
+    except Exception as e:
+        logger.warning("stocks.krx_failed", error=str(e))
+
+    try:
+        stock_map = await fetch_naver_stocks()
+        if len(stock_map) > 100:
+            stock_map.update({
+                "코스피": "KOSPI", "코스닥": "KOSDAQ",
+                "코스피200": "KOSPI200", "KRX": "KRX",
+            })
+            stock_map.update(ALIASES)
+            logger.info("stocks.loaded", total=len(stock_map), source="Naver")
+            return stock_map
+    except Exception as e:
+        logger.warning("stocks.naver_failed", error=str(e))
+
+    # 최종 폴백
+    fallback = {**FALLBACK_STOCKS, **ALIASES}
+    logger.warning("stocks.using_fallback", total=len(fallback))
+    return fallback
+
+
+async def auto_refresh_stocks(state_ref, interval_hours: int = 24):
+    """백그라운드에서 종목 목록 자동 갱신"""
+    while True:
+        await asyncio.sleep(interval_hours * 3600)
+        try:
+            new_map = await load_all_stocks()
+            if len(new_map) > 50:
+                state_ref.stock_map = new_map
+                logger.info("stocks.refreshed", total=len(new_map))
+        except Exception as e:
+            logger.warning("stocks.refresh_failed", error=str(e))
@@ -0,0 +1,6 @@
+FROM python:3.11-slim
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY main.py .
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8484", "--workers", "1"]
@@ -0,0 +1,965 @@
+"""
+기술적 분석 엔진 (Technical Analysis Engine)
+- 네이버 금융 차트 API (1차) / yfinance (2차 백업) OHLCV 수집
+- MA5/20/60/120, RSI(14), MACD(12,26,9), 볼린저밴드(20,2), 스토캐스틱(14,3)
+- 기술적 점수 (-100~100) 산출
+- 매수/매도 목표가 T1/T2/T3 + 손절가 자동 계산
+- vLLM AI 문장형 판단 생성
+- 보유 포지션 손익 + 맞춤 전략 분석
+"""
+import asyncio, json, os, re, math
+from datetime import datetime
+from typing import Optional, List, Dict
+import asyncpg, httpx, redis.asyncio as aioredis, structlog
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from fastapi import FastAPI, Query, Body
+from fastapi.responses import JSONResponse
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+
+structlog.configure(processors=[
+    structlog.processors.TimeStamper(fmt="iso"),
+    structlog.processors.add_log_level,
+    structlog.processors.JSONRenderer(),
+])
+logger = structlog.get_logger()
+
+REDIS_HOST     = os.getenv("REDIS_HOST", "redis")
+REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", "")
+PG_HOST        = os.getenv("POSTGRES_HOST", "postgres")
+PG_PORT        = int(os.getenv("POSTGRES_PORT", "5432"))
+PG_DB          = os.getenv("POSTGRES_DB", "trading_ai")
+PG_USER        = os.getenv("POSTGRES_USER", "kyu")
+PG_PASS        = os.getenv("POSTGRES_PASSWORD", "")
+OLLAMA_URL     = os.getenv("OLLAMA_URL", "http://ollama:11434")
+HEADERS        = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"}
+
+pg_pool: Optional[asyncpg.Pool] = None
+redis_cl: Optional[aioredis.Redis] = None
+scheduler = AsyncIOScheduler(timezone="Asia/Seoul")
+
+class Stats:
+    analyzed = 0; errors = 0; last_run = ""
+stats = Stats()
+
+# ── 기술적 지표 계산 ──────────────────────────────────────
+
+def _ema_series(values: List[float], period: int) -> List[float]:
+    if not values or len(values) < period:
+        return [values[-1]] * len(values) if values else []
+    k = 2.0 / (period + 1)
+    seed = sum(values[:period]) / period
+    out = [seed]
+    for v in values[period:]:
+        out.append(v * k + out[-1] * (1 - k))
+    return [out[0]] * (period - 1) + out
+
+def _ma(closes: List[float], n: int) -> float:
+    if not closes: return 0.0
+    data = closes[-n:] if len(closes) >= n else closes
+    return sum(data) / len(data)
+
+def _rsi(closes: List[float], period: int = 14) -> float:
+    if len(closes) < period + 1:
+        return 50.0
+    deltas = [closes[i] - closes[i-1] for i in range(1, len(closes))]
+    gains = [max(d, 0.0) for d in deltas]
+    losses = [max(-d, 0.0) for d in deltas]
+    ag = sum(gains[:period]) / period
+    al = sum(losses[:period]) / period
+    for i in range(period, len(gains)):
+        ag = (ag * (period - 1) + gains[i]) / period
+        al = (al * (period - 1) + losses[i]) / period
+    if al == 0:
+        return 100.0
+    return round(100 - 100 / (1 + ag / al), 2)
+
+def _macd(closes: List[float]) -> tuple:
+    if len(closes) < 26:
+        return 0.0, 0.0, 0.0
+    e12 = _ema_series(closes, 12)
+    e26 = _ema_series(closes, 26)
+    macd_line = [a - b for a, b in zip(e12, e26)]
+    signal_line = _ema_series(macd_line[-9:] if len(macd_line) >= 9 else macd_line, 9)
+    macd = macd_line[-1]
+    signal = signal_line[-1]
+    return round(macd, 4), round(signal, 4), round(macd - signal, 4)
+
+def _bollinger(closes: List[float], period: int = 20) -> tuple:
+    if len(closes) < period:
+        c = closes[-1] if closes else 0
+        return float(c), float(c), float(c), 0.5
+    recent = closes[-period:]
+    ma = sum(recent) / period
+    std = math.sqrt(sum((x - ma) ** 2 for x in recent) / period)
+    upper = ma + 2 * std
+    lower = ma - 2 * std
+    cur = closes[-1]
+    pct_b = (cur - lower) / (upper - lower) if upper != lower else 0.5
+    return round(upper), round(ma), round(lower), round(pct_b, 3)
+
+def _stochastic(highs: List[float], lows: List[float], closes: List[float], period: int = 14) -> tuple:
+    if len(closes) < period:
+        return 50.0, 50.0
+    h = max(highs[-period:])
+    l = min(lows[-period:])
+    k = ((closes[-1] - l) / (h - l) * 100) if h != l else 50.0
+    ks = []
+    for i in range(3):
+        idx = -(3 - i)
+        hh = max(highs[idx - period + 1:idx + 1] if idx != -1 else highs[-period:])
+        ll = min(lows[idx - period + 1:idx + 1] if idx != -1 else lows[-period:])
+        ks.append(((closes[idx] - ll) / (hh - ll) * 100) if hh != ll else 50.0)
+    d = sum(ks) / len(ks)
+    return round(k, 2), round(d, 2)
+
+def _vol_ratio(volumes: List[float], period: int = 20) -> float:
+    if len(volumes) < period + 1:
+        return 1.0
+    avg = sum(volumes[-period - 1:-1]) / period
+    return round(volumes[-1] / avg, 2) if avg > 0 else 1.0
+
+def _atr(highs: List[float], lows: List[float], closes: List[float], period: int = 14) -> float:
+    """Average True Range — 변동성 측정"""
+    if len(closes) < period + 1:
+        return 0.0
+    trs = []
+    for i in range(1, len(closes)):
+        tr = max(
+            highs[i] - lows[i],
+            abs(highs[i] - closes[i-1]),
+            abs(lows[i]  - closes[i-1]),
+        )
+        trs.append(tr)
+    recent = trs[-period:]
+    return round(sum(recent) / period, 2) if recent else 0.0
+
+
+def _obv(closes: List[float], volumes: List[float]) -> tuple:
+    """On-Balance Volume — 거래량 누적, 가격 상승일 +volume, 하락일 -volume"""
+    if len(closes) < 2: return 0, 0
+    obv = 0.0
+    obvs = [0.0]
+    for i in range(1, len(closes)):
+        if closes[i] > closes[i-1]: obv += volumes[i]
+        elif closes[i] < closes[i-1]: obv -= volumes[i]
+        obvs.append(obv)
+    # 최근 OBV 추세 (20일 평균 대비)
+    recent = obvs[-20:] if len(obvs) >= 20 else obvs
+    avg = sum(recent) / len(recent) if recent else 0
+    obv_trend = "상승" if obv > avg * 1.05 else ("하락" if obv < avg * 0.95 else "중립")
+    return int(obv), obv_trend
+
+
+def _vwap(highs: List[float], lows: List[float], closes: List[float], volumes: List[float],
+          period: int = 20) -> float:
+    """Volume Weighted Average Price — 거래량 가중 평균가, 기관 매매 기준선"""
+    if len(closes) < period: return float(closes[-1]) if closes else 0
+    cum_vp = sum(((highs[i] + lows[i] + closes[i]) / 3) * volumes[i]
+                 for i in range(-period, 0))
+    cum_v = sum(volumes[-period:])
+    return round(cum_vp / cum_v, 2) if cum_v > 0 else 0
+
+
+def _ichimoku(highs: List[float], lows: List[float], closes: List[float]) -> dict:
+    """일목균형표 (Ichimoku Kinko Hyo) — 5개 라인"""
+    if len(closes) < 52: return {}
+    # 전환선(Tenkan-sen): (9일 고가 + 9일 저가) / 2
+    tenkan = (max(highs[-9:]) + min(lows[-9:])) / 2
+    # 기준선(Kijun-sen): (26일 고가 + 26일 저가) / 2
+    kijun  = (max(highs[-26:]) + min(lows[-26:])) / 2
+    # 선행스팬1(Senkou Span A): (전환+기준)/2, 26일 후
+    span_a = (tenkan + kijun) / 2
+    # 선행스팬2(Senkou Span B): (52일 고가 + 52일 저가)/2, 26일 후
+    span_b = (max(highs[-52:]) + min(lows[-52:])) / 2
+    # 후행스팬(Chikou Span): 종가, 26일 전
+    chikou = closes[-26] if len(closes) > 26 else closes[-1]
+    cur = closes[-1]
+    cloud_top = max(span_a, span_b)
+    cloud_bot = min(span_a, span_b)
+    pos = "구름위" if cur > cloud_top else ("구름아래" if cur < cloud_bot else "구름속")
+    return {
+        "tenkan": int(tenkan), "kijun": int(kijun),
+        "span_a": int(span_a), "span_b": int(span_b), "chikou": int(chikou),
+        "cloud_pos": pos,
+    }
+
+
+def calc_indicators(ohlcv: List[dict]) -> dict:
+    if len(ohlcv) < 5:
+        return {}
+    closes  = [float(d["close"])  for d in ohlcv]
+    highs   = [float(d["high"])   for d in ohlcv]
+    lows    = [float(d["low"])    for d in ohlcv]
+    volumes = [float(d["volume"]) for d in ohlcv]
+
+    bb_upper, bb_mid, bb_lower, pct_b = _bollinger(closes, 20)
+    stoch_k, stoch_d                  = _stochastic(highs, lows, closes, 14)
+    macd, macd_signal, macd_hist      = _macd(closes)
+    atr14                             = _atr(highs, lows, closes, 14)
+    obv_val, obv_trend                = _obv(closes, volumes)
+    vwap_val                          = _vwap(highs, lows, closes, volumes, 20)
+    ichi                              = _ichimoku(highs, lows, closes)
+
+    return {
+        "price":       int(closes[-1]),
+        "ma5":         round(_ma(closes, 5)),
+        "ma20":        round(_ma(closes, 20)),
+        "ma60":        round(_ma(closes, 60)),
+        "ma120":       round(_ma(closes, 120)),
+        "rsi":         _rsi(closes, 14),
+        "macd":        macd,
+        "macd_signal": macd_signal,
+        "macd_hist":   macd_hist,
+        "bb_upper":    bb_upper,
+        "bb_mid":      bb_mid,
+        "bb_lower":    bb_lower,
+        "pct_b":       pct_b,
+        "stoch_k":     stoch_k,
+        "stoch_d":     stoch_d,
+        "vol_ratio":   _vol_ratio(volumes, 20),
+        "atr14":       atr14,
+        "high_52w":    int(max(highs[-min(len(highs), 252):])),
+        "low_52w":     int(min(lows[-min(len(lows), 252):])),
+        "obv":         obv_val,
+        "obv_trend":   obv_trend,
+        "vwap20":      vwap_val,
+        "ichimoku":    ichi,
+    }
+
+def calc_tech_score(ind: dict) -> tuple:
+    """기술적 점수 (-100~100)와 근거 신호 목록 반환"""
+    if not ind:
+        return 0.0, []
+
+    price = ind["price"]
+    score = 0.0
+    signals: List[str] = []
+
+    # ── 이동평균 (±40) ──────────────────────────
+    if ind["ma5"] > ind["ma20"]:
+        score += 10; signals.append("MA5>MA20 단기상승")
+    else:
+        score -= 10; signals.append("MA5<MA20 단기하락")
+
+    if ind["ma20"] > ind["ma60"]:
+        score += 8; signals.append("MA20>MA60 중기상승")
+    else:
+        score -= 8
+
+    if ind["ma60"] > ind["ma120"]:
+        score += 7
+    else:
+        score -= 7
+
+    if price > ind["ma20"]:
+        score += 8; signals.append("현재가 MA20 위")
+    elif price < ind["ma60"]:
+        score -= 8; signals.append("현재가 MA60 아래")
+
+    # 정배열/역배열
+    if ind["ma5"] > ind["ma20"] > ind["ma60"] > ind["ma120"]:
+        score += 7; signals.append("정배열")
+    elif ind["ma5"] < ind["ma20"] < ind["ma60"] < ind["ma120"]:
+        score -= 7; signals.append("역배열")
+
+    # ── RSI (±25) ───────────────────────────────
+    rsi = ind["rsi"]
+    if rsi <= 30:
+        score += 25; signals.append(f"RSI 과매도({rsi:.0f})")
+    elif rsi <= 40:
+        score += 15; signals.append(f"RSI 저점({rsi:.0f})")
+    elif rsi <= 60:
+        score += 5
+    elif rsi <= 70:
+        score -= 5
+    else:
+        score -= 20; signals.append(f"RSI 과매수({rsi:.0f})")
+
+    # ── MACD (±20) ──────────────────────────────
+    if ind["macd_hist"] > 0 and ind["macd"] > ind["macd_signal"]:
+        score += 20; signals.append("MACD 골든크로스")
+    elif ind["macd_hist"] > 0:
+        score += 8
+    elif ind["macd_hist"] < 0 and ind["macd"] < ind["macd_signal"]:
+        score -= 20; signals.append("MACD 데드크로스")
+    else:
+        score -= 5
+
+    # ── 볼린저밴드 (±15) ────────────────────────
+    pb = ind["pct_b"]
+    if pb < 0.1:
+        score += 15; signals.append("볼밴 하단(과매도)")
+    elif pb < 0.3:
+        score += 8
+    elif pb > 0.9:
+        score -= 15; signals.append("볼밴 상단(과매수)")
+    elif pb > 0.7:
+        score -= 5
+
+    # ── 스토캐스틱 (±10) ────────────────────────
+    sk, sd = ind["stoch_k"], ind["stoch_d"]
+    if sk < 20 and sk > sd:
+        score += 10; signals.append("스토캐스틱 바닥반등")
+    elif sk > 80 and sk < sd:
+        score -= 10; signals.append("스토캐스틱 고점하락")
+
+    # ── 거래량 보너스 (±5) ──────────────────────
+    if ind["vol_ratio"] > 1.5:
+        if score > 0:
+            score += 5; signals.append("거래량 급증(매수세)")
+        else:
+            score -= 5; signals.append("거래량 급증(매도세)")
+
+    return round(max(-100.0, min(100.0, score)), 1), signals
+
+def calc_price_targets(price: int, ind: dict, sig: str) -> dict:
+    """매수/매도 목표가(T1/T2/T3) + 손절가 계산 (10원 단위 반올림)"""
+    if not ind or price <= 0:
+        return {}
+
+    def r10(p): return int(round(p / 10) * 10)
+
+    h52    = ind.get("high_52w", price * 1.3)
+    l52    = ind.get("low_52w",  price * 0.7)
+    bb_up  = ind.get("bb_upper", price * 1.05)
+    bb_dn  = ind.get("bb_lower", price * 0.95)
+    ma20   = ind.get("ma20", price)
+    ma60   = ind.get("ma60", price)
+
+    if sig == "매수":
+        # 진입: 현재가 기준 -2% (기술지표 확인 후 매수)
+        entry = r10(price * 0.98)
+        # T1: +7% (단기), T2: +14% (중기), T3: min(+22%, 52주고가 -3%)
+        t1   = r10(price * 1.07)
+        t2   = r10(price * 1.14)
+        t3   = r10(min(price * 1.22, h52 * 0.97))
+        t3   = t3 if t3 > t2 else r10(price * 1.22)
+        # 손절: max(-8%, MA60 -5%) — 최대 -10% 이내 제한
+        raw_stop = max(price * 0.92, ma60 * 0.95)
+        stop = r10(max(raw_stop, price * 0.90))  # 최소 -10%
+        er1  = round((t1 - price) / price * 100, 1)
+        sl_r = round(abs(stop - price) / price * 100, 1)
+        # M3: ATR 기반 trailing stop (현재가 기준 2 ATR 아래)
+        atr = ind.get("atr14", 0)
+        atr_trailing = r10(price - 2 * atr) if atr > 0 else stop
+        return {
+            "entry_price": entry,
+            "t1": t1, "t1_pct": er1, "t1_sell_pct": 50,
+            "t2": t2, "t2_pct": round((t2 - price) / price * 100, 1), "t2_sell_pct": 30,
+            "t3": t3, "t3_pct": round((t3 - price) / price * 100, 1), "t3_sell_pct": 20,
+            "stop_loss": stop, "stop_pct": -sl_r,
+            "atr14": atr,
+            "trailing_stop": atr_trailing,
+            "risk_reward": round(er1 / sl_r, 2) if sl_r > 0 else 0,
+            "exit_strategy": "T1 50% + T2 30% + T3 20% 분할매도, 손절 또는 trailing(ATR×2) 도달시 전량",
+        }
+    else:  # 매도 / 관망(음수)
+        entry = r10(price * 1.02)
+        t1   = r10(price * 0.93)
+        t2   = r10(price * 0.86)
+        t3   = r10(max(price * 0.78, l52 * 1.03))
+        t3   = t3 if t3 < t2 else r10(price * 0.78)
+        raw_stop = min(price * 1.08, ma20 * 1.05)
+        stop = r10(min(raw_stop, price * 1.10))
+        er1  = round((price - t1) / price * 100, 1)
+        sl_r = round(abs(stop - price) / price * 100, 1)
+        return {
+            "entry_price": entry,
+            "t1": t1, "t1_pct": -er1,
+            "t2": t2, "t2_pct": -round((price - t2) / price * 100, 1),
+            "t3": t3, "t3_pct": -round((price - t3) / price * 100, 1),
+            "stop_loss": stop, "stop_pct": sl_r,
+            "risk_reward": round(er1 / sl_r, 2) if sl_r > 0 else 0,
+        }
+
+# ── OHLCV 수집 (네이버 차트 → yfinance 백업) ─────────────
+
+async def get_ohlcv_naver_chart(client: httpx.AsyncClient, code: str, count: int = 120) -> List[dict]:
+    """네이버 차트 API (fchart)"""
+    try:
+        r = await client.get(
+            f"https://fchart.stock.naver.com/sise.nhn?symbol={code}&timeframe=day&count={count}&requestType=0",
+            headers=HEADERS, timeout=10)
+        items = re.findall(r'data="([^"]+)"', r.text)
+        data = []
+        for item in items:
+            p = item.split("|")
+            if len(p) >= 6 and all(x.strip() for x in p[:5]):
+                try:
+                    data.append({
+                        "date": p[0], "open": int(p[1]), "high": int(p[2]),
+                        "low": int(p[3]), "close": int(p[4]),
+                        "volume": int(p[5]) if p[5].strip() else 0,
+                    })
+                except ValueError:
+                    pass
+        return data
+    except Exception:
+        return []
+
+async def get_ohlcv_naver_sise(client: httpx.AsyncClient, code: str, pages: int = 7) -> List[dict]:
+    """네이버 일별시세 페이지 (차트 API 실패 시 2차)"""
+    data = []
+    try:
+        for page in range(1, pages + 1):
+            r = await client.get(
+                f"https://finance.naver.com/item/sise_day.naver?code={code}&page={page}",
+                headers=HEADERS, timeout=12)
+            r.encoding = "euc-kr"
+            # 날짜+종가+전일비+시가+고가+저가+거래량 패턴
+            rows = re.findall(
+                r'(\d{4}\.\d{2}\.\d{2})[^<]*</span>.*?'
+                r'<span[^>]*>([\d,]+)</span>.*?'   # 종가
+                r'(?:.*?){3}'
+                r'<span[^>]*>([\d,]+)</span>.*?'   # 시가
+                r'<span[^>]*>([\d,]+)</span>.*?'   # 고가
+                r'<span[^>]*>([\d,]+)</span>.*?'   # 저가
+                r'<span[^>]*>([\d,]+)</span>',     # 거래량
+                r.text, re.DOTALL)
+            if not rows:
+                # 단순 종가만 추출 (더 넓은 패턴)
+                simple = re.findall(
+                    r'class="tah p10 gray03">(\d{4}\.\d{2}\.\d{2})<.*?'
+                    r'class="tah p11">([\d,]+)<',
+                    r.text, re.DOTALL)
+                for date_str, close_str in simple:
+                    close = int(close_str.replace(",", ""))
+                    if close > 0:
+                        data.append({"date": date_str.replace(".", ""),
+                                     "open": close, "high": close,
+                                     "low": close, "close": close, "volume": 0})
+            else:
+                for m in rows:
+                    date_str = m[0].replace(".", "")
+                    close = int(m[1].replace(",", ""))
+                    open_ = int(m[2].replace(",", "")) if m[2] else close
+                    high  = int(m[3].replace(",", "")) if m[3] else close
+                    low   = int(m[4].replace(",", "")) if m[4] else close
+                    vol   = int(m[5].replace(",", "")) if m[5] else 0
+                    if close > 0:
+                        data.append({"date": date_str, "open": open_,
+                                     "high": high, "low": low,
+                                     "close": close, "volume": vol})
+            if len(data) >= 120:
+                break
+            await asyncio.sleep(0.15)
+    except Exception as e:
+        logger.warning("ohlcv.sise.err", code=code, error=str(e))
+    return data[:120]
+
+async def get_ohlcv_yfinance(code: str) -> List[dict]:
+    """yfinance 최종 백업"""
+    try:
+        import yfinance as yf
+        loop = asyncio.get_event_loop()
+        def _fetch():
+            t = yf.Ticker(f"{code}.KS")
+            return t.history(period="1y")
+        h = await loop.run_in_executor(None, _fetch)
+        if h.empty:
+            return []
+        return [{"date": idx.strftime("%Y%m%d"),
+                 "open": int(row["Open"] or 0), "high": int(row["High"] or 0),
+                 "low": int(row["Low"] or 0), "close": int(row["Close"] or 0),
+                 "volume": int(row["Volume"] or 0)}
+                for idx, row in h.iterrows() if row["Close"] and row["High"]]
+    except Exception as e:
+        logger.warning("ohlcv.yfinance.err", code=code, error=str(e))
+        return []
+
+async def get_ohlcv(client: httpx.AsyncClient, code: str, count: int = 120) -> List[dict]:
+    data = await get_ohlcv_naver_chart(client, code, count)
+    if len(data) < 20:
+        data = await get_ohlcv_naver_sise(client, code)
+    if len(data) < 20:
+        logger.info("ohlcv.fallback.yfinance", code=code)
+        await asyncio.sleep(0.5)  # rate limit 방지
+        data = await get_ohlcv_yfinance(code)
+    return data
+
+# ── vLLM AI 판단문 생성 ───────────────────────────────────
+
+async def generate_ai_opinion(client: httpx.AsyncClient, code: str, name: str,
+                               ind: dict, tech_score: float, signals: List[str],
+                               targets: dict, news_score: float = 0) -> str:
+    """vLLM으로 문장형 투자 판단 생성"""
+    sig = "매수" if tech_score >= 30 else ("매도" if tech_score <= -30 else "관망")
+    price = ind.get("price", 0)
+    rsi   = ind.get("rsi", 50)
+    h52   = ind.get("high_52w", price)
+    l52   = ind.get("low_52w",  price)
+    pos52 = int((price - l52) / (h52 - l52) * 100) if h52 != l52 else 50
+
+    prompt = f"""다음 주식 데이터를 바탕으로 투자자에게 명확한 매매 판단을 3~5문장으로 설명하세요.
+한국어로, 구체적인 가격과 수치를 포함해서 작성하세요.
+
+종목: {name}({code})
+현재가: {price:,}원
+기술점수: {tech_score}점 / 신호: {sig}
+이동평균: MA5={ind.get('ma5',0):,} MA20={ind.get('ma20',0):,} MA60={ind.get('ma60',0):,}
+RSI: {rsi} / MACD히스토그램: {'양수(골든)' if ind.get('macd_hist',0)>0 else '음수(데드)'}
+볼린저%B: {ind.get('pct_b',0.5)*100:.0f}%
+52주위치: 하단에서 {pos52}%
+뉴스감성점수: {news_score:.0f}
+기술신호: {', '.join(signals[:4])}
+{f"1차목표가: {targets.get('t1',0):,}원 / 손절가: {targets.get('stop_loss',0):,}원" if targets else ""}
+
+투자 판단 (3~5문장):"""
+
+    try:
+        r = await client.post(f"{OLLAMA_URL}/v1/chat/completions", json={
+            "model": "exaone3.5:7.8b",
+            "messages": [
+                {"role": "system", "content": "당신은 한국 주식 전문 애널리스트입니다. 기술적 분석 데이터를 바탕으로 명확하고 실용적인 투자 의견을 제시합니다."},
+                {"role": "user",   "content": prompt}
+            ],
+            "max_tokens": 300, "temperature": 0.2
+        }, timeout=60)
+        return r.json()["choices"][0]["message"]["content"].strip()
+    except Exception as e:
+        logger.warning("ai_opinion.err", code=code, error=str(e))
+        return ""
+
+# ── 포지션 손익 분석 ──────────────────────────────────────
+
+class PositionRequest(BaseModel):
+    code: str
+    name: str = ""
+    buy_price: int
+    qty: int
+
+def analyze_position(price: int, buy_price: int, qty: int,
+                     ind: dict, tech_score: float) -> dict:
+    """보유 포지션 기반 맞춤 전략 계산"""
+    pnl      = (price - buy_price) * qty
+    pnl_pct  = (price - buy_price) / buy_price * 100
+    total_buy = buy_price * qty
+    h52 = ind.get("high_52w", price * 1.3)
+    l52 = ind.get("low_52w",  price * 0.7)
+    ma20 = ind.get("ma20", price)
+    ma60 = ind.get("ma60", price)
+    bbu  = ind.get("bb_upper", price * 1.05)
+
+    def r10(p): return int(round(p / 10) * 10)
+
+    # 손절선: 매입가 -8% 또는 MA60 -3% 중 높은 것
+    stop = max(r10(buy_price * 0.92), r10(ma60 * 0.97))
+
+    # 목표가
+    t1 = r10(max(buy_price * 1.08, bbu * 0.97))   # 본전+8% 또는 볼밴 상단
+    t2 = r10(max((price + h52) / 2, buy_price * 1.15))
+    t3 = r10(max(h52 * 0.97, buy_price * 1.25))
+
+    # 추가매수 구간 (물타기) - 현재가 -5%, -10%
+    avg_down1_price = r10(price * 0.95)
+    avg_down1_qty   = max(1, qty // 3)
+    avg_down1_avg   = (total_buy + avg_down1_price * avg_down1_qty) / (qty + avg_down1_qty)
+
+    avg_down2_price = r10(price * 0.90)
+    avg_down2_qty   = max(1, qty // 2)
+    avg_down2_avg   = (total_buy + avg_down2_price * avg_down2_qty) / (qty + avg_down2_qty)
+
+    return {
+        "pnl":          pnl,
+        "pnl_pct":      round(pnl_pct, 2),
+        "total_buy":    total_buy,
+        "current_value": price * qty,
+        "stop_loss":    stop,
+        "stop_pnl":     (stop - buy_price) * qty,
+        "t1": t1, "t1_pnl": (t1 - buy_price) * qty,
+        "t2": t2, "t2_pnl": (t2 - buy_price) * qty,
+        "t3": t3, "t3_pnl": (t3 - buy_price) * qty,
+        "avg_down": [
+            {"price": avg_down1_price, "add_qty": avg_down1_qty,
+             "new_avg": round(avg_down1_avg), "add_cost": avg_down1_price * avg_down1_qty},
+            {"price": avg_down2_price, "add_qty": avg_down2_qty,
+             "new_avg": round(avg_down2_avg), "add_cost": avg_down2_price * avg_down2_qty},
+        ],
+        "action": (
+            "손절 고려" if price <= stop else
+            "추가매수 검토" if pnl_pct < -5 and tech_score >= 0 else
+            "홀드" if -5 <= pnl_pct < 5 else
+            "1차 익절 고려" if pnl_pct >= 10 else "홀드"
+        ),
+    }
+
+# ── 단일 종목 분석 ────────────────────────────────────────
+
+EXCLUDE_KEYWORDS = (
+    "기업인수목적", "선박투자회사", "부동산투자회사", "특별자산", "인프라투자",
+    "사모투자", "맥쿼리", "리츠", "REITs",
+)
+
+async def analyze_stock(client: httpx.AsyncClient, code: str, name: str = "",
+                        with_ai: bool = False, news_score: float = 0) -> Optional[dict]:
+    # 이름이 없으면 DB에서 조회
+    if not name and pg_pool:
+        try:
+            async with pg_pool.acquire() as conn:
+                name = await conn.fetchval(
+                    "SELECT corp_name FROM dart_corps WHERE stock_code=$1", code) or ""
+        except: pass
+
+    # SPAC·REITs·선박펀드 등 제외
+    if any(kw in name for kw in EXCLUDE_KEYWORDS):
+        return None
+
+    ohlcv = await get_ohlcv(client, code, 120)
+    if len(ohlcv) < 20:
+        return None
+
+    ind = calc_indicators(ohlcv)
+    if not ind:
+        return None
+
+    # 주가 500원 미만 penny stock 제외
+    if ind.get("price", 0) < 500:
+        return None
+
+    tech_score, signals = calc_tech_score(ind)
+    sig = "매수" if tech_score >= 30 else ("매도" if tech_score <= -30 else "관망")
+    # 관망도 목표가 계산 (기술점수 양수면 매수 기준, 음수면 매도 기준)
+    tgt_sig = "매수" if tech_score >= 0 else "매도"
+    targets = calc_price_targets(ind["price"], ind, tgt_sig)
+
+    ai_opinion = ""
+    if with_ai and sig != "관망":
+        ai_opinion = await generate_ai_opinion(
+            client, code, name, ind, tech_score, signals, targets, news_score)
+
+    result = {
+        "code": code, "name": name,
+        "tech_score": tech_score, "signal": sig,
+        "signals": signals, "indicators": ind, "targets": targets,
+        "ai_opinion": ai_opinion,
+        "analyzed_at": datetime.now().isoformat(),
+    }
+
+    if redis_cl:
+        try:
+            await redis_cl.set(f"ta:{code}", json.dumps(result, ensure_ascii=False), ex=1800)
+        except: pass
+
+    if pg_pool:
+        try:
+            async with pg_pool.acquire() as conn:
+                await conn.execute("""
+                    INSERT INTO stock_technical (
+                        stock_code, stock_name, price,
+                        ma5, ma20, ma60, ma120,
+                        rsi, macd, macd_signal, macd_hist,
+                        bb_upper, bb_mid, bb_lower, pct_b,
+                        stoch_k, stoch_d, vol_ratio,
+                        tech_score, signal, signals, targets, analyzed_at
+                    ) VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14,$15,$16,$17,$18,$19,$20,$21,$22,$23)
+                    ON CONFLICT (stock_code) DO UPDATE SET
+                        stock_name=$2, price=$3,
+                        ma5=$4, ma20=$5, ma60=$6, ma120=$7,
+                        rsi=$8, macd=$9, macd_signal=$10, macd_hist=$11,
+                        bb_upper=$12, bb_mid=$13, bb_lower=$14, pct_b=$15,
+                        stoch_k=$16, stoch_d=$17, vol_ratio=$18,
+                        tech_score=$19, signal=$20, signals=$21, targets=$22, analyzed_at=$23
+                """,
+                    code, name, ind["price"],
+                    ind["ma5"], ind["ma20"], ind["ma60"], ind["ma120"],
+                    ind["rsi"], ind["macd"], ind["macd_signal"], ind["macd_hist"],
+                    ind["bb_upper"], ind["bb_mid"], ind["bb_lower"], ind["pct_b"],
+                    ind["stoch_k"], ind["stoch_d"], ind["vol_ratio"],
+                    tech_score, sig,
+                    json.dumps(signals, ensure_ascii=False),
+                    json.dumps(targets, ensure_ascii=False),
+                    datetime.now())
+        except Exception as e:
+            logger.warning("ta.db.err", code=code, error=str(e))
+
+    return result
+
+# ── DB 초기화 ─────────────────────────────────────────────
+
+async def init_db():
+    async with pg_pool.acquire() as conn:
+        await conn.execute("""
+            CREATE TABLE IF NOT EXISTS stock_technical (
+                id          SERIAL PRIMARY KEY,
+                stock_code  VARCHAR(10) UNIQUE NOT NULL,
+                stock_name  VARCHAR(100) DEFAULT '',
+                price       INTEGER DEFAULT 0,
+                ma5         FLOAT DEFAULT 0,
+                ma20        FLOAT DEFAULT 0,
+                ma60        FLOAT DEFAULT 0,
+                ma120       FLOAT DEFAULT 0,
+                rsi         FLOAT DEFAULT 50,
+                macd        FLOAT DEFAULT 0,
+                macd_signal FLOAT DEFAULT 0,
+                macd_hist   FLOAT DEFAULT 0,
+                bb_upper    FLOAT DEFAULT 0,
+                bb_mid      FLOAT DEFAULT 0,
+                bb_lower    FLOAT DEFAULT 0,
+                pct_b       FLOAT DEFAULT 0.5,
+                stoch_k     FLOAT DEFAULT 50,
+                stoch_d     FLOAT DEFAULT 50,
+                vol_ratio   FLOAT DEFAULT 1,
+                tech_score  FLOAT DEFAULT 0,
+                signal      VARCHAR(10) DEFAULT '관망',
+                signals     JSONB DEFAULT '[]'::jsonb,
+                targets     JSONB DEFAULT '{}'::jsonb,
+                analyzed_at TIMESTAMP DEFAULT NOW()
+            )
+        """)
+        await conn.execute("CREATE INDEX IF NOT EXISTS idx_ta_score  ON stock_technical(tech_score DESC)")
+        await conn.execute("CREATE INDEX IF NOT EXISTS idx_ta_signal ON stock_technical(signal)")
+    logger.info("ta.db.initialized")
+
+# ── 전체 분석 작업 ────────────────────────────────────────
+
+async def job_analyze():
+    logger.info("ta.job.start")
+    async with httpx.AsyncClient() as client:
+        codes: List[tuple] = []
+
+        if pg_pool:
+            try:
+                rows = await pg_pool.fetch("SELECT stock_code, corp_name FROM dart_corps LIMIT 500")
+                codes = [(r["stock_code"], r["corp_name"] or "") for r in rows if r["stock_code"]]
+            except: pass
+
+        if not codes:
+            for sosok in [0, 1]:
+                for page in range(1, 30):
+                    try:
+                        r = await client.get(
+                            f"https://finance.naver.com/sise/sise_market_sum.naver?sosok={sosok}&page={page}",
+                            headers=HEADERS, timeout=15)
+                        r.encoding = "euc-kr"
+                        found = re.findall(r'main\.naver\?code=(\d{6})[^>]*>([^<]+)</a>', r.text)
+                        if not found: break
+                        codes.extend([(c.strip(), n.strip()) for c, n in found])
+                        await asyncio.sleep(0.2)
+                    except: break
+                    if len(codes) >= 500: break
+
+        ok = 0
+        for code, name in codes[:500]:
+            if not code or len(code) != 6: continue
+            try:
+                result = await analyze_stock(client, code, name)
+                if result: ok += 1
+            except Exception as e:
+                stats.errors += 1
+                logger.warning("ta.analyze.err", code=code, error=str(e))
+            await asyncio.sleep(0.4)
+
+        stats.analyzed += ok
+        stats.last_run = datetime.now().isoformat()
+        logger.info("ta.job.done", analyzed=ok)
+
+# ── FastAPI ────────────────────────────────────────────────
+
+app = FastAPI(title="기술적 분석 엔진")
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+@app.on_event("startup")
+async def startup():
+    global pg_pool, redis_cl
+    pg_pool = await asyncpg.create_pool(
+        host=PG_HOST, port=PG_PORT, database=PG_DB,
+        user=PG_USER, password=PG_PASS, min_size=2, max_size=5)
+    redis_cl = aioredis.Redis(
+        host=REDIS_HOST, port=6379, password=REDIS_PASSWORD, db=5, decode_responses=True)
+    await init_db()
+    scheduler.add_job(job_analyze, "cron", day_of_week="mon-fri",
+                      hour="9-16", minute="*/30", id="ta_30m", replace_existing=True)
+    scheduler.add_job(job_analyze, "cron", day_of_week="mon-fri",
+                      hour=16, minute=15, id="ta_close", replace_existing=True)
+    scheduler.start()
+    logger.info("ta-engine.started")
+
+@app.on_event("shutdown")
+async def shutdown():
+    scheduler.shutdown()
+    if pg_pool: await pg_pool.close()
+    if redis_cl: await redis_cl.aclose()
+
+@app.get("/health")
+async def health():
+    return {"status": "ok", "analyzed": stats.analyzed,
+            "errors": stats.errors, "last_run": stats.last_run}
+
+@app.get("/technical/{code}")
+async def technical(code: str):
+    if redis_cl:
+        cached = await redis_cl.get(f"ta:{code}")
+        if cached:
+            return JSONResponse(content=json.loads(cached))
+    if pg_pool:
+        async with pg_pool.acquire() as conn:
+            row = await conn.fetchrow("SELECT * FROM stock_technical WHERE stock_code=$1", code)
+            if row:
+                d = dict(row)
+                d["analyzed_at"] = str(d["analyzed_at"])
+                for k in ("signals", "targets"):
+                    if isinstance(d[k], str):
+                        d[k] = json.loads(d[k])
+                return JSONResponse(content=d)
+    # 실시간 분석
+    async with httpx.AsyncClient() as client:
+        result = await analyze_stock(client, code)
+        if result:
+            return JSONResponse(content=result)
+    return JSONResponse(content={"error": "not found"}, status_code=404)
+
+@app.get("/ranking")
+async def ranking(limit: int = Query(default=30), signal: str = Query(default="")):
+    async with pg_pool.acquire() as conn:
+        if signal:
+            rows = await conn.fetch(
+                "SELECT * FROM stock_technical WHERE signal=$1 ORDER BY tech_score DESC LIMIT $2",
+                signal, limit)
+        else:
+            rows = await conn.fetch(
+                "SELECT * FROM stock_technical ORDER BY tech_score DESC LIMIT $1", limit)
+    result = []
+    for row in rows:
+        d = dict(row)
+        d["analyzed_at"] = str(d["analyzed_at"])
+        for k in ("signals", "targets"):
+            if isinstance(d[k], str):
+                d[k] = json.loads(d[k])
+        result.append(d)
+    return result
+
+@app.get("/buy-candidates")
+async def buy_candidates(limit: int = Query(default=20)):
+    """기술적 매수 후보 (점수 30 이상) + 펀더멘탈 점수 합산"""
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT t.*,
+                   s.news_score, s.dart_score, s.recommendation AS fundamental_rec,
+                   s.total_score AS fundamental_total
+            FROM stock_technical t
+            LEFT JOIN stock_scores s
+                   ON t.stock_code = s.stock_code
+                  AND s.score_date = (SELECT MAX(score_date) FROM stock_scores)
+            WHERE t.signal = '매수' AND t.tech_score >= 30
+            ORDER BY (t.tech_score + COALESCE(s.total_score, 0)) DESC
+            LIMIT $1
+        """, limit)
+    result = []
+    for row in rows:
+        d = dict(row)
+        d["analyzed_at"] = str(d["analyzed_at"])
+        for k in ("signals", "targets"):
+            if isinstance(d.get(k), str):
+                d[k] = json.loads(d[k])
+        result.append(d)
+    return result
+
+@app.post("/analyze/all")
+async def analyze_all():
+    asyncio.create_task(job_analyze())
+    return {"status": "started"}
+
+@app.post("/analyze/{code}")
+async def analyze_single(code: str, ai: bool = False):
+    async with httpx.AsyncClient() as client:
+        result = await analyze_stock(client, code, with_ai=ai)
+        if result:
+            return JSONResponse(content=result)
+    return JSONResponse(content={"error": "analysis failed"}, status_code=500)
+
+# ── 보유 포지션 맞춤 분석 ────────────────────────────────
+
+@app.post("/position")
+async def position_analysis(req: PositionRequest, ai: bool = False):
+    """보유 종목 매입가/수량 기반 맞춤 손익 + 전략 분석"""
+    code = req.code
+
+    # 캐시 확인
+    result = None
+    if redis_cl:
+        try:
+            cached = await redis_cl.get(f"ta:{code}")
+            if cached:
+                result = json.loads(cached)
+        except: pass
+
+    if not result:
+        async with httpx.AsyncClient() as client:
+            result = await analyze_stock(client, code, req.name, with_ai=False)
+
+    if not result:
+        return JSONResponse(content={"error": "종목 분석 실패"}, status_code=500)
+
+    ind        = result.get("indicators", {})
+    tech_score = result.get("tech_score", 0)
+    price      = ind.get("price", 0)
+
+    pos = analyze_position(price, req.buy_price, req.qty, ind, tech_score)
+
+    # AI 판단문 (요청 시)
+    ai_opinion = result.get("ai_opinion", "")
+    if ai and not ai_opinion:
+        async with httpx.AsyncClient() as client:
+            ai_opinion = await generate_ai_opinion(
+                client, code, req.name or result.get("name", code),
+                ind, tech_score, result.get("signals", []),
+                result.get("targets", {}))
+
+    return {
+        "code": code,
+        "name": req.name or result.get("name", code),
+        "buy_price": req.buy_price,
+        "qty": req.qty,
+        "current_price": price,
+        "tech_score": tech_score,
+        "signal": result.get("signal"),
+        "signals": result.get("signals", []),
+        "indicators": ind,
+        "position": pos,
+        "targets": result.get("targets", {}),
+        "ai_opinion": ai_opinion,
+        "analyzed_at": result.get("analyzed_at"),
+    }
+
+# ── 종목 전체 리포트 (AI 판단문 포함) ────────────────────
+
+@app.get("/report/{code}")
+async def full_report(code: str):
+    """기술적 분석 + AI 판단문 + 뉴스감성 통합 리포트"""
+    news_score = 0.0
+    if pg_pool:
+        try:
+            async with pg_pool.acquire() as conn:
+                row = await conn.fetchrow(
+                    "SELECT news_score FROM stock_scores WHERE stock_code=$1 "
+                    "ORDER BY score_date DESC LIMIT 1", code)
+                if row:
+                    news_score = float(row["news_score"] or 0)
+        except: pass
+
+    async with httpx.AsyncClient() as client:
+        result = await analyze_stock(client, code, with_ai=True, news_score=news_score)
+
+    if not result:
+        return JSONResponse(content={"error": "분석 실패"}, status_code=500)
+
+    # DB에서 추가 정보
+    extra = {}
+    if pg_pool:
+        try:
+            async with pg_pool.acquire() as conn:
+                score_row = await conn.fetchrow(
+                    "SELECT * FROM stock_scores WHERE stock_code=$1 "
+                    "ORDER BY score_date DESC LIMIT 1", code)
+                news_rows = await conn.fetch(
+                    "SELECT title, sentiment, intensity, reason "
+                    "FROM news_analysis WHERE primary_stock=$1 "
+                    "ORDER BY analyzed_at DESC LIMIT 5", code)
+                if score_row:
+                    extra["score"] = dict(score_row)
+                    extra["score"]["score_date"] = str(extra["score"]["score_date"])
+                extra["recent_news"] = [dict(r) for r in news_rows]
+        except: pass
+
+    return {**result, **extra, "news_score": news_score}
@@ -0,0 +1,10 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+httpx==0.27.0
+redis==5.0.4
+asyncpg==0.29.0
+apscheduler==3.10.4
+orjson==3.10.3
+structlog==24.2.0
+yfinance==0.2.54
+pydantic==2.7.1
@@ -0,0 +1,7 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+CMD ["python", "-u", "main.py"]
@@ -0,0 +1,499 @@
+"""
+Trading AI Telegram Bot (양방향 대화)
+
+명령어:
+  /start, /help  — 안내
+  /추천 또는 /buy  — 오늘 강력매수 top 5
+  /매도 또는 /sell — 오늘 강력매도 top 3
+  /종목 005930 또는 /stock 005930 — 특정 종목 상세
+  /시장 또는 /market — 시장 상황 요약
+
+자유 텍스트 → EXAONE 3.5가 시스템 DB 데이터 + 사용자 질문 종합해 답변
+"""
+import os
+import re
+import asyncio
+import json
+import logging
+from datetime import datetime
+from typing import Optional, List
+
+import asyncpg
+import httpx
+import structlog
+from telegram import Update, BotCommand
+from telegram.constants import ParseMode
+from telegram.ext import (Application, CommandHandler, MessageHandler,
+                           filters, ContextTypes)
+
+logging.basicConfig(level=logging.INFO,
+                     format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+logger = structlog.get_logger()
+
+PG = {
+    "host": os.getenv("POSTGRES_HOST", "postgres"),
+    "port": int(os.getenv("POSTGRES_PORT", 5432)),
+    "database": os.getenv("POSTGRES_DB", "trading_ai"),
+    "user": os.getenv("POSTGRES_USER", "kyu"),
+    "password": os.getenv("POSTGRES_PASSWORD", ""),
+}
+TG_TOKEN = os.getenv("TELEGRAM_BOT_TOKEN", "")
+ALLOWED_CHAT_ID = os.getenv("TELEGRAM_CHAT_ID", "")  # 허용된 사용자만
+OLLAMA_URL = os.getenv("OLLAMA_URL", "http://ollama:11434")
+EXAONE_MODEL = os.getenv("EXAONE_MODEL", "exaone3.5:7.8b")
+SCORE_API = os.getenv("SCORE_API_URL", "http://score-engine:8686")
+
+pg_pool: Optional[asyncpg.Pool] = None
+
+
+# ─────────────────────────────────────────────────────────────
+# 권한 체크
+# ─────────────────────────────────────────────────────────────
+def is_allowed(update: Update) -> bool:
+    if not ALLOWED_CHAT_ID:
+        return True
+    return str(update.effective_chat.id) == ALLOWED_CHAT_ID
+
+
+# ─────────────────────────────────────────────────────────────
+# DB 조회 헬퍼
+# ─────────────────────────────────────────────────────────────
+async def db_top_buys(limit: int = 5) -> list:
+    async with pg_pool.acquire() as conn:
+        return await conn.fetch("""
+            SELECT s.stock_code, s.stock_name, s.total_score, s.recommendation,
+                   s.us_overnight_adj, s.us_pair_top,
+                   s.insider_signal, s.consensus_signal, s.macro_signal,
+                   s.inst_flow_signal, s.valuation_pct,
+                   s.top_reasons
+            FROM stock_scores s
+            WHERE s.score_date=(SELECT MAX(score_date) FROM stock_scores)
+              AND s.recommendation IN ('강력매수','매수관심')
+            ORDER BY s.total_score DESC LIMIT $1
+        """, limit)
+
+
+async def db_top_sells(limit: int = 3) -> list:
+    async with pg_pool.acquire() as conn:
+        return await conn.fetch("""
+            SELECT stock_code, stock_name, total_score, recommendation, top_reasons
+            FROM stock_scores
+            WHERE score_date=(SELECT MAX(score_date) FROM stock_scores)
+              AND recommendation IN ('강력매도','매도관심')
+            ORDER BY total_score ASC LIMIT $1
+        """, limit)
+
+
+async def db_stock_detail(code: str) -> Optional[dict]:
+    async with pg_pool.acquire() as conn:
+        row = await conn.fetchrow("""
+            SELECT s.*,
+                   d.corp_name, d.sector_name,
+                   f.roe, f.operating_margin, f.debt_ratio, f.revenue_growth,
+                   f.fcf_ratio
+            FROM stock_scores s
+            JOIN dart_corps d ON d.stock_code=s.stock_code
+            LEFT JOIN dart_financials f ON f.stock_code=s.stock_code
+              AND f.reprt_code='11011'
+              AND f.bsns_year=(SELECT MAX(bsns_year) FROM dart_financials f2
+                               WHERE f2.stock_code=s.stock_code AND f2.reprt_code='11011')
+            WHERE s.stock_code=$1
+              AND s.score_date=(SELECT MAX(score_date) FROM stock_scores
+                                WHERE stock_code=$1)
+        """, code)
+        return dict(row) if row else None
+
+
+async def db_market_regime() -> Optional[dict]:
+    async with pg_pool.acquire() as conn:
+        row = await conn.fetchrow(
+            "SELECT regime, regime_adj, dt FROM market_regime ORDER BY dt DESC LIMIT 1")
+        macro = await conn.fetch("""
+            SELECT DISTINCT ON (indicator) indicator, trade_date, value
+            FROM macro_daily ORDER BY indicator, trade_date DESC
+        """)
+        return {
+            "regime": dict(row) if row else None,
+            "macro": {r["indicator"]: float(r["value"]) for r in macro}
+        }
+
+
+async def find_stock_code_by_name(name: str) -> Optional[str]:
+    """이름으로 종목코드 검색"""
+    async with pg_pool.acquire() as conn:
+        row = await conn.fetchrow("""
+            SELECT stock_code FROM dart_corps
+            WHERE is_active=true AND corp_name ILIKE $1
+            ORDER BY CASE WHEN corp_name=$2 THEN 0 ELSE 1 END
+            LIMIT 1
+        """, f"%{name}%", name)
+        return row["stock_code"] if row else None
+
+
+# ─────────────────────────────────────────────────────────────
+# EXAONE 호출
+# ─────────────────────────────────────────────────────────────
+SYSTEM_PROMPT = """당신은 워렌 버핏 스타일의 한국 주식 가치투자 분석가입니다.
+사용자 질문에 대해 제공된 시스템 데이터를 우선 참고하여 답변하세요.
+
+원칙:
+- 데이터에 근거한 객관적 답변
+- 짧고 명확하게 (3~5문장)
+- ROE/부채비율/영업이익률 중심의 가치투자 관점
+- 투자는 본인 책임이라는 점 마지막에 짧게 언급
+- 데이터에 없는 내용은 추측하지 말고 "모르겠습니다"라고 말할 것
+"""
+
+
+async def ask_exaone(user_msg: str, context: str = "") -> str:
+    prompt = f"{SYSTEM_PROMPT}\n\n=== 시스템 데이터 ===\n{context}\n\n=== 사용자 질문 ===\n{user_msg}"
+    try:
+        async with httpx.AsyncClient(timeout=120) as cl:
+            r = await cl.post(f"{OLLAMA_URL}/api/generate", json={
+                "model": EXAONE_MODEL, "prompt": prompt, "stream": False,
+                "options": {"temperature": 0.3, "num_predict": 400}
+            })
+        if r.status_code != 200:
+            return f"⚠️ LLM 호출 실패 (status {r.status_code})"
+        j = r.json()
+        return j.get("response", "").strip() or "(빈 응답)"
+    except Exception as e:
+        return f"⚠️ LLM 오류: {e}"
+
+
+# ─────────────────────────────────────────────────────────────
+# 포매팅 헬퍼
+# ─────────────────────────────────────────────────────────────
+def fmt_pct(v) -> str:
+    try: return f"{float(v):+.1f}"
+    except: return "-"
+
+
+def fmt_buys(rows: list) -> str:
+    if not rows:
+        return "📭 오늘 매수 추천 없음"
+    lines = ["🟢 <b>오늘 매수 추천</b>"]
+    for i, r in enumerate(rows, 1):
+        emoji = "🔥" if r["recommendation"] == "강력매수" else "✨"
+        line = f"{emoji} <b>{i}. {r['stock_name']}({r['stock_code']})</b> {r['total_score']:.1f}점"
+        lines.append(line)
+        sigs = []
+        if r["us_overnight_adj"] and r["us_overnight_adj"] != 0:
+            sigs.append(f"US{fmt_pct(r['us_overnight_adj'])}")
+        if r["insider_signal"] and r["insider_signal"] != 0:
+            sigs.append(f"내부자{fmt_pct(r['insider_signal'])}")
+        if r["consensus_signal"] and r["consensus_signal"] != 0:
+            sigs.append(f"컨센{fmt_pct(r['consensus_signal'])}")
+        if r["valuation_pct"] and r["valuation_pct"] != 0:
+            sigs.append(f"밸류{fmt_pct(r['valuation_pct'])}")
+        if r["inst_flow_signal"] and r["inst_flow_signal"] != 0:
+            sigs.append(f"기관{fmt_pct(r['inst_flow_signal'])}")
+        if sigs:
+            lines.append(f"   ▸ {' / '.join(sigs)}")
+        if r["top_reasons"]:
+            lines.append(f"   ▸ {r['top_reasons'][:80]}")
+    return "\n".join(lines)
+
+
+def fmt_sells(rows: list) -> str:
+    if not rows:
+        return "📭 오늘 매도 추천 없음"
+    lines = ["🔴 <b>오늘 매도 신호</b>"]
+    for i, r in enumerate(rows, 1):
+        lines.append(f"<b>{i}. {r['stock_name']}({r['stock_code']})</b> {r['total_score']:.1f}점 · {r['recommendation']}")
+        if r["top_reasons"]:
+            lines.append(f"   ▸ {r['top_reasons'][:80]}")
+    return "\n".join(lines)
+
+
+def fmt_stock(d: dict) -> str:
+    if not d:
+        return "❓ 종목을 찾을 수 없습니다"
+    lines = [
+        f"📊 <b>{d['corp_name']}({d['stock_code']})</b>",
+        f"종합점수 <b>{d['total_score']:.1f}</b> · {d['recommendation']}",
+        f"섹터: {d.get('sector_name') or '미분류'}",
+        "",
+        "<b>📈 시그널</b>",
+    ]
+    sigs = [
+        ("뉴스", d.get("news_score")),
+        ("공시", d.get("dart_score")),
+        ("기술", d.get("technical_score")),
+        ("외국인", d.get("foreign_score")),
+        ("공매도", d.get("short_score")),
+        ("미국동조", d.get("us_overnight_adj")),
+        ("내부자", d.get("insider_signal")),
+        ("컨센", d.get("consensus_signal")),
+        ("매크로", d.get("macro_signal")),
+        ("기관 5d", d.get("inst_flow_signal")),
+        ("밸류", d.get("valuation_pct")),
+    ]
+    for k, v in sigs:
+        if v and float(v) != 0:
+            lines.append(f"  {k}: {fmt_pct(v)}")
+    lines.append("")
+    lines.append("<b>💰 재무</b>")
+    if d.get("roe") is not None:
+        lines.append(f"  ROE {d['roe']:.1f}% · 부채 {d.get('debt_ratio',0):.0f}% · 영업이익률 {d.get('operating_margin',0):.1f}%")
+    if d.get("intrinsic_value"):
+        lines.append(f"  내재가치 {int(d['intrinsic_value']):,}원 · 안전마진 {d.get('margin_of_safety',0):.0f}%")
+    if d.get("us_pair_top"):
+        lines.append("")
+        lines.append(f"🇺🇸 {d['us_pair_top']}")
+    if d.get("top_reasons"):
+        lines.append("")
+        lines.append(f"<i>{d['top_reasons'][:200]}</i>")
+    return "\n".join(lines)
+
+
+# ─────────────────────────────────────────────────────────────
+# 핸들러
+# ─────────────────────────────────────────────────────────────
+async def cmd_start(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    msg = (
+        "👋 <b>Trading AI 봇</b>\n\n"
+        "명령어:\n"
+        "/buy — 오늘 매수 추천\n"
+        "/sell — 오늘 매도 신호\n"
+        "/stock 005930 — 특정 종목 상세\n"
+        "/deep 005930 — AI 심층분석(RAG+EXAONE)\n"
+        "/market — 시장 상황\n"
+        "/help — 도움말\n\n"
+        "자유 텍스트로 질문해도 됩니다. 예:\n"
+        "<i>\"에스엠 사도 돼?\"</i>\n"
+        "<i>\"오늘 추천 알려줘\"</i>\n"
+        "<i>\"삼성전자 어때?\"</i>"
+    )
+    await update.message.reply_text(msg, parse_mode=ParseMode.HTML)
+
+
+async def cmd_help(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    await cmd_start(update, ctx)
+
+
+async def cmd_buys(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    rows = await db_top_buys(10)
+    await update.message.reply_text(fmt_buys(rows), parse_mode=ParseMode.HTML)
+
+
+async def cmd_sells(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    rows = await db_top_sells(5)
+    await update.message.reply_text(fmt_sells(rows), parse_mode=ParseMode.HTML)
+
+
+async def cmd_stock(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    args = ctx.args
+    if not args:
+        await update.message.reply_text("종목코드 또는 이름을 입력하세요. 예: /종목 005930 또는 /종목 삼성전자")
+        return
+    q = " ".join(args).strip()
+    code = q if re.match(r"^\d{6}$", q) else await find_stock_code_by_name(q)
+    if not code:
+        await update.message.reply_text(f"❓ '{q}' 종목을 찾을 수 없습니다")
+        return
+    d = await db_stock_detail(code)
+    await update.message.reply_text(fmt_stock(d), parse_mode=ParseMode.HTML)
+
+
+def fmt_deep(r: dict) -> str:
+    if r.get("error"):
+        return f"⚠️ {r.get('code')}: {r['error']}"
+    icon = {"강력매수": "🟢🟢", "매수": "🟢", "중립": "⚪",
+            "매도": "🔴", "강력매도": "🔴🔴"}.get(r.get("recommendation"), "⚪")
+    lines = [
+        f"{icon} <b>{r.get('name')}({r.get('code')})</b> — AI심층분석",
+        f"판단: <b>{r.get('recommendation')}</b> (확신도 {r.get('conviction')}/5) "
+        f"· 퀀트 {r.get('quant_score',0):.0f}점",
+        f"밸류: {r.get('valuation_view','-')} · 기간: {r.get('time_horizon','-')}",
+        "",
+        f"<b>📝 투자논거</b>\n{r.get('thesis','')}",
+    ]
+    if r.get("key_points"):
+        lines.append("\n<b>✅ 핵심근거</b>")
+        lines += [f"  • {p}" for p in r["key_points"][:5]]
+    if r.get("risks"):
+        lines.append("\n<b>⚠️ 리스크</b>")
+        lines += [f"  • {p}" for p in r["risks"][:4]]
+    if r.get("catalyst_watch"):
+        lines.append("\n<b>👀 관전포인트</b>")
+        lines += [f"  • {p}" for p in r["catalyst_watch"][:3]]
+    tp, sl = r.get("target_price", 0) or 0, r.get("stop_loss", 0) or 0
+    if tp or sl:
+        lines.append(f"\n🎯 목표가 {tp:,}원 · 손절가 {sl:,}원")
+    lines.append("\n<i>※ 투자 판단·책임은 본인에게 있습니다</i>")
+    return "\n".join(lines)
+
+
+async def cmd_deep(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    args = ctx.args
+    if not args:
+        await update.message.reply_text("종목코드/이름 입력. 예: /deep 005930 또는 /deep 삼성전자")
+        return
+    q = " ".join(args).strip()
+    code = q if re.match(r"^\d{6}$", q) else await find_stock_code_by_name(q)
+    if not code:
+        await update.message.reply_text(f"❓ '{q}' 종목을 찾을 수 없습니다")
+        return
+    await update.message.reply_text("🧠 RAG+EXAONE 심층분석 중… (최대 1~2분)")
+    try:
+        async with httpx.AsyncClient(timeout=200) as cl:
+            resp = await cl.get(f"{SCORE_API}/deep-analysis/{code}",
+                                params={"refresh": "true", "notify": "false"})
+            r = resp.json()
+    except Exception as e:
+        await update.message.reply_text(f"⚠️ 분석 실패: {e}")
+        return
+    await update.message.reply_text(fmt_deep(r), parse_mode=ParseMode.HTML)
+
+
+async def cmd_market(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    m = await db_market_regime()
+    regime = m.get("regime") or {}
+    macro = m.get("macro") or {}
+    lines = ["🌍 <b>시장 상황</b>"]
+    if regime:
+        lines.append(f"  레짐: <b>{regime.get('regime','?')}</b> (보정 {regime.get('regime_adj',0):+.0f})")
+        lines.append(f"  날짜: {regime.get('dt')}")
+    lines.append("")
+    lines.append("<b>매크로</b>")
+    if "usdkrw" in macro: lines.append(f"  USD/KRW {macro['usdkrw']:.1f}원")
+    if "kor_10y" in macro: lines.append(f"  국고채 10년 {macro['kor_10y']:.2f}%")
+    if "kor_3y"  in macro: lines.append(f"  국고채 3년  {macro['kor_3y']:.2f}%")
+    if "kospi"   in macro: lines.append(f"  KOSPI {macro['kospi']:,.1f}")
+    await update.message.reply_text("\n".join(lines), parse_mode=ParseMode.HTML)
+
+
+# 자유 텍스트
+async def on_text(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    if not is_allowed(update): return
+    text = (update.message.text or "").strip()
+    if not text:
+        return
+
+    # 한글 단축 키워드 → 명령 라우팅 (빠른 응답)
+    if text in ("추천", "매수", "오늘추천", "buy"):
+        rows = await db_top_buys(10)
+        await update.message.reply_text(fmt_buys(rows), parse_mode=ParseMode.HTML)
+        return
+    if text in ("매도", "팔것", "sell"):
+        rows = await db_top_sells(5)
+        await update.message.reply_text(fmt_sells(rows), parse_mode=ParseMode.HTML)
+        return
+    if text in ("시장", "장", "market"):
+        await cmd_market(update, ctx)
+        return
+
+    await update.message.chat.send_action("typing")
+
+    # 컨텍스트 구성: 종목명 언급되면 그 종목 데이터, 아니면 시장 전반
+    context_parts = []
+    # 종목코드 / 종목명 감지
+    code = None
+    code_match = re.search(r"\b(\d{6})\b", text)
+    if code_match:
+        code = code_match.group(1)
+    if not code:
+        # 종목명 추정 (2~10자 한글)
+        for cand in re.findall(r"[가-힣A-Z]{2,10}", text):
+            if cand in ("매수", "매도", "추천", "오늘", "지금", "사도", "팔아", "어때"):
+                continue
+            c = await find_stock_code_by_name(cand)
+            if c:
+                code = c
+                break
+
+    if code:
+        d = await db_stock_detail(code)
+        if d:
+            context_parts.append(f"[종목 {d['corp_name']}({code}) 데이터]")
+            context_parts.append(f"종합점수 {d.get('total_score',0):.1f}점, 등급 {d.get('recommendation','-')}, 섹터 {d.get('sector_name') or '미분류'}")
+            context_parts.append(
+                f"세부 점수 — 뉴스 {d.get('news_score',0):.0f}, 기술 {d.get('technical_score',0):.0f}, "
+                f"공시 {d.get('dart_score',0):.0f}, 외국인 {d.get('foreign_score',0):.0f}, "
+                f"미국동조 {d.get('us_overnight_adj',0):.1f}, 내부자 {d.get('insider_signal',0):.1f}, "
+                f"컨센서스 {d.get('consensus_signal',0):.1f}, 기관5d {d.get('inst_flow_signal',0):.1f}, "
+                f"밸류 percentile {d.get('valuation_pct',0):.1f}")
+            if d.get("roe") is not None:
+                context_parts.append(
+                    f"재무 — ROE {d['roe']:.1f}%, 영업이익률 {d.get('operating_margin',0):.1f}%, "
+                    f"부채비율 {d.get('debt_ratio',0):.0f}%, 매출성장 {d.get('revenue_growth',0):.1f}%")
+            if d.get("intrinsic_value"):
+                context_parts.append(
+                    f"DCF 내재가치 {int(d['intrinsic_value']):,}원, 안전마진 {d.get('margin_of_safety',0):.0f}%")
+            if d.get("us_pair_top"):
+                context_parts.append(f"미국 페어: {d['us_pair_top']}")
+            if d.get("top_reasons"):
+                context_parts.append(f"근거: {d['top_reasons']}")
+    else:
+        # 시장 전반
+        buys = await db_top_buys(5)
+        sells = await db_top_sells(3)
+        m = await db_market_regime()
+        regime = (m.get("regime") or {})
+        macro = m.get("macro") or {}
+        context_parts.append(f"[시장 상황] 레짐: {regime.get('regime','?')}")
+        if macro:
+            context_parts.append(
+                f"환율 {macro.get('usdkrw',0):.0f}원, 10년물 {macro.get('kor_10y',0):.2f}%, "
+                f"KOSPI {macro.get('kospi',0):,.0f}")
+        if buys:
+            context_parts.append("[오늘 매수 추천 top]")
+            for r in buys[:5]:
+                context_parts.append(f"- {r['stock_name']}({r['stock_code']}) {r['total_score']:.0f}점 {r['recommendation']}")
+        if sells:
+            context_parts.append("[오늘 매도 신호]")
+            for r in sells[:3]:
+                context_parts.append(f"- {r['stock_name']}({r['stock_code']}) {r['total_score']:.0f}점 {r['recommendation']}")
+
+    context_str = "\n".join(context_parts) if context_parts else "(관련 데이터 없음)"
+    answer = await ask_exaone(text, context_str)
+    # 답변 길이 제한
+    if len(answer) > 3500:
+        answer = answer[:3500] + "..."
+    await update.message.reply_text(answer)
+
+
+# ─────────────────────────────────────────────────────────────
+# 부트
+# ─────────────────────────────────────────────────────────────
+async def post_init(app: Application):
+    global pg_pool
+    pg_pool = await asyncpg.create_pool(**PG, min_size=2, max_size=5)
+    # 커맨드 메뉴 등록
+    await app.bot.set_my_commands([
+        BotCommand("buy", "오늘 매수 추천"),
+        BotCommand("sell", "오늘 매도 신호"),
+        BotCommand("stock", "종목 상세 (예: /stock 005930)"),
+        BotCommand("market", "시장 상황"),
+        BotCommand("help", "도움말"),
+    ])
+    logger.info("telegram-bot.ready")
+
+
+def main():
+    if not TG_TOKEN:
+        logger.error("missing TELEGRAM_BOT_TOKEN")
+        return
+    app = Application.builder().token(TG_TOKEN).post_init(post_init).build()
+    # 한국어/영어 명령어 둘 다
+    app.add_handler(CommandHandler("start", cmd_start))
+    app.add_handler(CommandHandler("help", cmd_help))
+    app.add_handler(CommandHandler(["buy", "buys"], cmd_buys))
+    app.add_handler(CommandHandler(["sell", "sells"], cmd_sells))
+    app.add_handler(CommandHandler(["stock"], cmd_stock))
+    app.add_handler(CommandHandler(["deep"], cmd_deep))
+    app.add_handler(CommandHandler(["market"], cmd_market))
+    app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, on_text))
+    logger.info("telegram-bot.start", model=EXAONE_MODEL)
+    app.run_polling(close_loop=False)
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,5 @@
+python-telegram-bot[asyncio]==21.5
+asyncpg==0.29.0
+httpx==0.27.0
+structlog==24.2.0
+orjson==3.10.3
@@ -0,0 +1,268 @@
+import { useState, useEffect, useCallback } from "react";
+
+const API = "http://192.168.0.60:8989/api";
+
+const SENTIMENT_MAP = { "호재": { color: "#00E676", icon: "▲", bg: "rgba(0,230,118,0.08)" }, "악재": { color: "#FF5252", icon: "▼", bg: "rgba(255,82,82,0.08)" }, "중립": { color: "#78909C", icon: "●", bg: "rgba(120,144,156,0.06)" } };
+const REC_MAP = { "강력매수": { color: "#00E676", weight: 900 }, "매수관심": { color: "#69F0AE", weight: 700 }, "관망": { color: "#78909C", weight: 500 }, "매도관심": { color: "#FF8A80", weight: 700 }, "강력매도": { color: "#FF5252", weight: 900 } };
+
+function useFetch(endpoint, interval = 60000) {
+  const [data, setData] = useState(null);
+  const [loading, setLoading] = useState(true);
+  const load = useCallback(() => {
+    fetch(`${API}${endpoint}`).then(r => r.json()).then(d => { setData(d); setLoading(false); }).catch(() => setLoading(false));
+  }, [endpoint]);
+  useEffect(() => { load(); const t = setInterval(load, interval); return () => clearInterval(t); }, [load, interval]);
+  return { data, loading, reload: load };
+}
+
+function formatTime(ts) {
+  if (!ts) return "-";
+  const d = new Date(ts);
+  const now = new Date();
+  const diff = (now - d) / 1000 / 60;
+  if (diff < 60) return `${Math.floor(diff)}분 전`;
+  if (diff < 1440) return `${Math.floor(diff / 60)}시간 전`;
+  return d.toLocaleDateString("ko-KR", { month: "short", day: "numeric" });
+}
+
+function ScoreBar({ value, max = 100 }) {
+  const pct = Math.min(100, Math.max(0, ((value + max) / (2 * max)) * 100));
+  const color = value >= 30 ? "#00E676" : value >= -30 ? "#78909C" : "#FF5252";
+  return (
+    <div style={{ width: "100%", height: 6, background: "rgba(255,255,255,0.05)", borderRadius: 3, overflow: "hidden" }}>
+      <div style={{ width: `${pct}%`, height: "100%", background: color, borderRadius: 3, transition: "width 0.8s ease" }} />
+    </div>
+  );
+}
+
+function SummaryCard({ icon, label, value, sub, color }) {
+  return (
+    <div style={{ background: "rgba(255,255,255,0.03)", border: "1px solid rgba(255,255,255,0.06)", borderRadius: 16, padding: "24px 20px", display: "flex", flexDirection: "column", gap: 8 }}>
+      <div style={{ fontSize: 13, color: "#90A4AE", letterSpacing: 1 }}>{icon} {label}</div>
+      <div style={{ fontSize: 36, fontWeight: 800, color: color || "#E0E0E0", fontFamily: "'JetBrains Mono', monospace", lineHeight: 1 }}>{value}</div>
+      {sub && <div style={{ fontSize: 12, color: "#607D8B" }}>{sub}</div>}
+    </div>
+  );
+}
+
+function Header({ onRefresh }) {
+  const [time, setTime] = useState(new Date());
+  useEffect(() => { const t = setInterval(() => setTime(new Date()), 1000); return () => clearInterval(t); }, []);
+  return (
+    <div style={{ display: "flex", justifyContent: "space-between", alignItems: "center", padding: "20px 0", borderBottom: "1px solid rgba(255,255,255,0.06)", marginBottom: 24 }}>
+      <div>
+        <h1 style={{ margin: 0, fontSize: 22, fontWeight: 800, color: "#E0E0E0", letterSpacing: -0.5 }}>
+          <span style={{ color: "#00E676" }}>◆</span> Trading AI Dashboard
+        </h1>
+        <div style={{ fontSize: 12, color: "#546E7A", marginTop: 4 }}>뉴스 + DART 공시 기반 종목 분석 시스템</div>
+      </div>
+      <div style={{ display: "flex", alignItems: "center", gap: 16 }}>
+        <div style={{ fontSize: 13, color: "#607D8B", fontFamily: "'JetBrains Mono', monospace" }}>
+          {time.toLocaleString("ko-KR", { hour: "2-digit", minute: "2-digit", second: "2-digit" })}
+        </div>
+        <button onClick={onRefresh} style={{ background: "rgba(0,230,118,0.1)", border: "1px solid rgba(0,230,118,0.2)", borderRadius: 8, padding: "8px 16px", color: "#00E676", fontSize: 12, cursor: "pointer", fontWeight: 600 }}>
+          ↻ 새로고침
+        </button>
+      </div>
+    </div>
+  );
+}
+
+function RecommendationCard({ item }) {
+  const rec = REC_MAP[item.recommendation] || REC_MAP["관망"];
+  return (
+    <div style={{ background: "rgba(255,255,255,0.02)", border: "1px solid rgba(255,255,255,0.05)", borderRadius: 12, padding: 16, display: "flex", flexDirection: "column", gap: 10 }}>
+      <div style={{ display: "flex", justifyContent: "space-between", alignItems: "flex-start" }}>
+        <div>
+          <div style={{ fontSize: 15, fontWeight: 700, color: "#E0E0E0" }}>{item.stock_name}</div>
+          <div style={{ fontSize: 11, color: "#546E7A", fontFamily: "'JetBrains Mono', monospace" }}>{item.stock_code}</div>
+        </div>
+        <div style={{ background: `${rec.color}15`, border: `1px solid ${rec.color}30`, borderRadius: 6, padding: "4px 10px", fontSize: 12, fontWeight: rec.weight, color: rec.color }}>
+          {item.recommendation}
+        </div>
+      </div>
+      <div style={{ display: "flex", gap: 12, fontSize: 11, color: "#78909C" }}>
+        <span>종합 <b style={{ color: "#E0E0E0", fontSize: 14 }}>{item.total_score?.toFixed(1)}</b></span>
+        <span>뉴스 <b style={{ color: "#69F0AE" }}>{item.news_score?.toFixed(0)}</b></span>
+        <span>공시 <b style={{ color: "#40C4FF" }}>{item.dart_score?.toFixed(0)}</b></span>
+        <span>가격 <b style={{ color: "#FFD740" }}>{item.price_score?.toFixed(0)}</b></span>
+      </div>
+      <ScoreBar value={item.total_score} />
+      {item.top_reasons && (
+        <div style={{ fontSize: 11, color: "#607D8B", lineHeight: 1.5, borderTop: "1px solid rgba(255,255,255,0.04)", paddingTop: 8 }}>
+          {item.top_reasons.split("|")[0]?.trim().substring(0, 120)}...
+        </div>
+      )}
+    </div>
+  );
+}
+
+function NewsItem({ item }) {
+  const s = SENTIMENT_MAP[item.sentiment] || SENTIMENT_MAP["중립"];
+  return (
+    <div style={{ display: "flex", gap: 12, padding: "12px 0", borderBottom: "1px solid rgba(255,255,255,0.03)" }}>
+      <div style={{ minWidth: 40, textAlign: "center" }}>
+        <div style={{ fontSize: 18, color: s.color }}>{s.icon}</div>
+        <div style={{ fontSize: 10, color: s.color, fontWeight: 700 }}>{item.intensity || 0}</div>
+      </div>
+      <div style={{ flex: 1, minWidth: 0 }}>
+        <div style={{ fontSize: 13, color: "#CFD8DC", lineHeight: 1.4, overflow: "hidden", textOverflow: "ellipsis", whiteSpace: "nowrap" }}>
+          {item.title}
+        </div>
+        <div style={{ display: "flex", gap: 8, marginTop: 4, fontSize: 11 }}>
+          {item.primary_stock && <span style={{ color: "#40C4FF" }}>{item.primary_stock}</span>}
+          <span style={{ color: "#546E7A" }}>{item.source}</span>
+          <span style={{ color: "#455A64" }}>{formatTime(item.analyzed_at)}</span>
+        </div>
+        {item.reason && <div style={{ fontSize: 11, color: "#607D8B", marginTop: 4, lineHeight: 1.4 }}>{item.reason.substring(0, 100)}</div>}
+      </div>
+      <div style={{ background: s.bg, borderRadius: 6, padding: "4px 8px", fontSize: 11, color: s.color, fontWeight: 600, height: "fit-content", whiteSpace: "nowrap" }}>
+        {item.investment_action || "관망"}
+      </div>
+    </div>
+  );
+}
+
+function AlertItem({ item }) {
+  const s = SENTIMENT_MAP[item.sentiment] || SENTIMENT_MAP["중립"];
+  return (
+    <div style={{ background: `${s.color}08`, border: `1px solid ${s.color}20`, borderRadius: 10, padding: 12, marginBottom: 8 }}>
+      <div style={{ display: "flex", justifyContent: "space-between", alignItems: "center" }}>
+        <div style={{ fontSize: 13, fontWeight: 600, color: s.color }}>
+          {s.icon} {item.sentiment} (강도 {item.intensity})
+        </div>
+        <div style={{ fontSize: 10, color: "#546E7A" }}>{formatTime(item.analyzed_at)}</div>
+      </div>
+      <div style={{ fontSize: 12, color: "#CFD8DC", marginTop: 6, lineHeight: 1.4 }}>{item.title}</div>
+      <div style={{ fontSize: 11, color: "#78909C", marginTop: 4 }}>{item.reason?.substring(0, 100)}</div>
+    </div>
+  );
+}
+
+function Tab({ active, label, onClick }) {
+  return (
+    <button onClick={onClick} style={{
+      background: active ? "rgba(0,230,118,0.1)" : "transparent",
+      border: active ? "1px solid rgba(0,230,118,0.2)" : "1px solid transparent",
+      borderRadius: 8, padding: "8px 16px", color: active ? "#00E676" : "#607D8B",
+      fontSize: 13, fontWeight: active ? 700 : 500, cursor: "pointer", transition: "all 0.2s"
+    }}>{label}</button>
+  );
+}
+
+export default function Dashboard() {
+  const [tab, setTab] = useState("overview");
+  const { data: summary, reload: r1 } = useFetch("/summary", 30000);
+  const { data: ranking } = useFetch("/ranking", 60000);
+  const { data: recs, reload: r2 } = useFetch("/recommendations", 60000);
+  const { data: recent } = useFetch("/recent?limit=30", 30000);
+  const { data: alerts } = useFetch("/alerts", 30000);
+  const { data: timeline } = useFetch("/timeline?hours=48", 60000);
+
+  const refresh = () => { r1(); r2(); };
+
+  return (
+    <div style={{ minHeight: "100vh", background: "#0D1117", color: "#E0E0E0", fontFamily: "'Noto Sans KR', -apple-system, sans-serif", padding: "0 24px 40px" }}>
+      <style>{`
+        @import url('https://fonts.googleapis.com/css2?family=Noto+Sans+KR:wght@300;400;500;700;800;900&family=JetBrains+Mono:wght@400;600;700&display=swap');
+        * { box-sizing: border-box; margin: 0; }
+        ::-webkit-scrollbar { width: 6px; }
+        ::-webkit-scrollbar-track { background: transparent; }
+        ::-webkit-scrollbar-thumb { background: rgba(255,255,255,0.1); border-radius: 3px; }
+        body { background: #0D1117 !important; }
+      `}</style>
+      
+      <div style={{ maxWidth: 1400, margin: "0 auto" }}>
+        <Header onRefresh={refresh} />
+
+        {/* Tabs */}
+        <div style={{ display: "flex", gap: 4, marginBottom: 24 }}>
+          <Tab active={tab === "overview"} label="📊 종합" onClick={() => setTab("overview")} />
+          <Tab active={tab === "recommend"} label="🏆 추천종목" onClick={() => setTab("recommend")} />
+          <Tab active={tab === "news"} label="📰 뉴스피드" onClick={() => setTab("news")} />
+          <Tab active={tab === "alerts"} label="🚨 알림" onClick={() => setTab("alerts")} />
+        </div>
+
+        {/* Overview */}
+        {tab === "overview" && (
+          <div style={{ display: "flex", flexDirection: "column", gap: 24 }}>
+            {/* Summary Cards */}
+            <div style={{ display: "grid", gridTemplateColumns: "repeat(auto-fit, minmax(180px, 1fr))", gap: 12 }}>
+              <SummaryCard icon="📰" label="분석 뉴스" value={summary?.total || 0} sub="최근 7일" />
+              <SummaryCard icon="▲" label="호재" value={summary?.positive || 0} color="#00E676" sub={`${summary?.sentiment_ratio || 50}%`} />
+              <SummaryCard icon="▼" label="악재" value={summary?.negative || 0} color="#FF5252" />
+              <SummaryCard icon="📋" label="DART 공시" value={summary?.dart || 0} />
+              <SummaryCard icon="🏢" label="분석 종목" value={summary?.stocks_analyzed || 0} />
+            </div>
+
+            {/* Two column: Ranking + Recent */}
+            <div style={{ display: "grid", gridTemplateColumns: "1fr 1fr", gap: 16 }}>
+              {/* Ranking */}
+              <div style={{ background: "rgba(255,255,255,0.02)", border: "1px solid rgba(255,255,255,0.05)", borderRadius: 16, padding: 20 }}>
+                <h3 style={{ fontSize: 14, fontWeight: 700, color: "#90A4AE", marginBottom: 16 }}>🏆 종목 랭킹 (최신)</h3>
+                {ranking && ranking.length > 0 ? ranking.slice(0, 10).map((item, i) => (
+                  <div key={i} style={{ display: "flex", alignItems: "center", gap: 12, padding: "10px 0", borderBottom: "1px solid rgba(255,255,255,0.03)" }}>
+                    <div style={{ width: 24, height: 24, borderRadius: "50%", background: i < 3 ? "rgba(0,230,118,0.15)" : "rgba(255,255,255,0.04)", display: "flex", alignItems: "center", justifyContent: "center", fontSize: 11, fontWeight: 700, color: i < 3 ? "#00E676" : "#546E7A" }}>{i + 1}</div>
+                    <div style={{ flex: 1 }}>
+                      <div style={{ fontSize: 13, fontWeight: 600, color: "#CFD8DC" }}>{item.stock_name}</div>
+                      <div style={{ fontSize: 10, color: "#546E7A" }}>{item.stock_code}</div>
+                    </div>
+                    <div style={{ textAlign: "right" }}>
+                      <div style={{ fontSize: 14, fontWeight: 700, color: (REC_MAP[item.recommendation] || {}).color || "#78909C" }}>{item.total_score?.toFixed(1)}</div>
+                      <div style={{ fontSize: 10, color: (REC_MAP[item.recommendation] || {}).color || "#78909C" }}>{item.recommendation}</div>
+                    </div>
+                  </div>
+                )) : <div style={{ color: "#546E7A", fontSize: 13, padding: 20, textAlign: "center" }}>데이터 수집 중...</div>}
+              </div>
+
+              {/* Recent News */}
+              <div style={{ background: "rgba(255,255,255,0.02)", border: "1px solid rgba(255,255,255,0.05)", borderRadius: 16, padding: 20, maxHeight: 500, overflow: "auto" }}>
+                <h3 style={{ fontSize: 14, fontWeight: 700, color: "#90A4AE", marginBottom: 16 }}>📰 최근 분석</h3>
+                {recent && recent.length > 0 ? recent.slice(0, 15).map((item, i) => (
+                  <NewsItem key={i} item={item} />
+                )) : <div style={{ color: "#546E7A", fontSize: 13, padding: 20, textAlign: "center" }}>데이터 수집 중...</div>}
+              </div>
+            </div>
+          </div>
+        )}
+
+        {/* Recommendations */}
+        {tab === "recommend" && (
+          <div>
+            <div style={{ fontSize: 12, color: "#546E7A", marginBottom: 16 }}>뉴스 감성 + DART 공시 + 가격 모멘텀 기반 종합 점수</div>
+            <div style={{ display: "grid", gridTemplateColumns: "repeat(auto-fill, minmax(320px, 1fr))", gap: 12 }}>
+              {recs && recs.length > 0 ? recs.map((item, i) => (
+                <RecommendationCard key={i} item={item} />
+              )) : <div style={{ color: "#546E7A", fontSize: 13, padding: 40, textAlign: "center", gridColumn: "1/-1" }}>추천 데이터가 아직 없습니다. 16:30 자동 산출 후 업데이트됩니다.</div>}
+            </div>
+          </div>
+        )}
+
+        {/* News Feed */}
+        {tab === "news" && (
+          <div style={{ background: "rgba(255,255,255,0.02)", border: "1px solid rgba(255,255,255,0.05)", borderRadius: 16, padding: 20 }}>
+            <h3 style={{ fontSize: 14, fontWeight: 700, color: "#90A4AE", marginBottom: 16 }}>전체 뉴스 피드</h3>
+            {recent && recent.length > 0 ? recent.map((item, i) => (
+              <NewsItem key={i} item={item} />
+            )) : <div style={{ color: "#546E7A", fontSize: 13, padding: 40, textAlign: "center" }}>뉴스 데이터 수집 중...</div>}
+          </div>
+        )}
+
+        {/* Alerts */}
+        {tab === "alerts" && (
+          <div>
+            <div style={{ fontSize: 12, color: "#546E7A", marginBottom: 16 }}>강도 3 이상 뉴스/공시 (최근 24시간)</div>
+            {alerts && alerts.length > 0 ? alerts.map((item, i) => (
+              <AlertItem key={i} item={item} />
+            )) : <div style={{ background: "rgba(255,255,255,0.02)", borderRadius: 16, padding: 40, textAlign: "center", color: "#546E7A", fontSize: 13 }}>현재 긴급 알림이 없습니다.</div>}
+          </div>
+        )}
+
+        {/* Footer */}
+        <div style={{ marginTop: 40, padding: "16px 0", borderTop: "1px solid rgba(255,255,255,0.04)", textAlign: "center", fontSize: 11, color: "#37474F" }}>
+          Trading AI System • 뉴스/공시 기반 자동 분석 • 투자 판단의 참고 자료로만 활용하세요
+        </div>
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,8 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir --default-timeout=180 --retries=5 -r requirements.txt
+COPY . .
+EXPOSE 8383
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8383", "--workers", "1", "--log-level", "info"]
@@ -0,0 +1,838 @@
+"""
+US Market Sync Service (port 8383, 172.30.0.24)
+
+미국증시 → 한국증시 동조 시그널 생성:
+  A) 섹터 ETF 동조 (SOXX, XBI, LIT 등 14개) → 한국 같은 섹터에 ±5점
+  B) 개별 페어 (NVDA↔SK하이닉스 등) 60일 회귀 베타 → ±10점
+  D) 자동 페어 발굴 (코스피200 × S&P500 60일 상관계수)
+
+매일 KST 07:30 미국 정규장 마감 후 수집, 08:00 시그널 계산.
+"""
+import os
+import asyncio
+import json
+from datetime import date, datetime, timedelta
+from typing import Optional, List, Dict
+
+import asyncpg
+import orjson
+import structlog
+from fastapi import FastAPI, Query, BackgroundTasks
+from apscheduler.schedulers.asyncio import AsyncIOScheduler
+from apscheduler.triggers.cron import CronTrigger
+from pytz import timezone
+
+import httpx
+import pandas as pd
+import numpy as np
+from scipy import stats
+
+# ─────────────────────────────────────────────────────────────
+# 설정
+# ─────────────────────────────────────────────────────────────
+PG = {
+    "host": os.getenv("POSTGRES_HOST", "postgres"),
+    "port": int(os.getenv("POSTGRES_PORT", 5432)),
+    "database": os.getenv("POSTGRES_DB", "trading_ai"),
+    "user": os.getenv("POSTGRES_USER", "kyu"),
+    "password": os.getenv("POSTGRES_PASSWORD", ""),
+}
+KST = timezone("Asia/Seoul")
+FINNHUB_KEY = os.getenv("FINNHUB_API_KEY", "")
+FINNHUB_BASE = "https://finnhub.io/api/v1"
+AV_KEY = os.getenv("ALPHAVANTAGE_API_KEY", "")
+AV_BASE = "https://www.alphavantage.co/query"
+AV_DAILY_LIMIT = 25  # free tier: 25 calls/day
+
+logger = structlog.get_logger()
+app = FastAPI(title="US Market Sync")
+pg_pool: Optional[asyncpg.Pool] = None
+scheduler = AsyncIOScheduler(timezone=KST)
+
+# ─────────────────────────────────────────────────────────────
+# ETF → 한국 섹터 키워드 매핑 (dart_corps.sector LIKE 매칭용)
+# 섹터 컬럼은 KSIC 한글 (예: "반도체 및 전자집적회로 제조업")
+# ─────────────────────────────────────────────────────────────
+SECTOR_ETFS: Dict[str, Dict] = {
+    "SOXX": {"keywords": ["반도체", "전자집적", "전자부품"],
+             "desc": "iShares Semiconductor"},
+    "SMH":  {"keywords": ["반도체", "전자집적"],
+             "desc": "VanEck Semiconductor"},
+    "XLK":  {"keywords": ["소프트웨어", "정보서비스", "컴퓨터"],
+             "desc": "Tech Select Sector"},
+    "QQQ":  {"keywords": ["소프트웨어", "인터넷", "포털"],
+             "desc": "Nasdaq 100"},
+    "XBI":  {"keywords": ["바이오", "생물의약", "의약품"],
+             "desc": "S&P Biotech"},
+    "IBB":  {"keywords": ["바이오", "생물의약"],
+             "desc": "Nasdaq Biotech"},
+    "LIT":  {"keywords": ["전지", "축전지", "이차전지"],
+             "desc": "Global Lithium"},
+    "XLE":  {"keywords": ["석유", "정제", "가스", "원유"],
+             "desc": "Energy Select"},
+    "XLF":  {"keywords": ["은행", "보험", "증권", "금융"],
+             "desc": "Financial Select"},
+    "XLV":  {"keywords": ["의료", "병원", "의료용기기"],
+             "desc": "Health Care Select"},
+    "XLI":  {"keywords": ["기계", "산업용", "건설"],
+             "desc": "Industrials Select"},
+    "XLP":  {"keywords": ["식품", "음료", "가공식품"],
+             "desc": "Consumer Staples"},
+    "XLY":  {"keywords": ["자동차", "여가", "의류", "백화점"],
+             "desc": "Consumer Discretionary"},
+    "ITA":  {"keywords": ["항공", "방위", "조선"],
+             "desc": "Aerospace & Defense"},
+}
+
+# ─────────────────────────────────────────────────────────────
+# 검증된 핵심 페어 (seed) — 학술/시장 통념상 강한 동조
+# (us_ticker, kr_code, kr_name_hint)  kr_name_hint는 검증용 코멘트
+# ─────────────────────────────────────────────────────────────
+SEED_PAIRS: List = [
+    # 반도체
+    ("NVDA", "000660"),  # SK하이닉스
+    ("NVDA", "005930"),  # 삼성전자
+    ("AMD",  "000660"),
+    ("AMD",  "005930"),
+    ("MU",   "000660"),
+    ("INTC", "005930"),
+    ("TSM",  "000660"),
+    ("TSM",  "005930"),
+    # 2차전지
+    ("TSLA", "373220"),  # LG에너지솔루션
+    ("TSLA", "247540"),  # 에코프로비엠
+    ("TSLA", "006400"),  # 삼성SDI
+    ("ALB",  "006400"),  # 알버말 (리튬)
+    ("ALB",  "373220"),
+    # 자동차
+    ("F",    "005380"),  # 현대차
+    ("GM",   "000270"),  # 기아
+    ("TSLA", "012330"),  # 현대모비스
+    ("TM",   "005380"),  # 도요타→현대차
+    # 인터넷/IT
+    ("GOOGL","035420"),  # NAVER
+    ("META", "035720"),  # 카카오
+    ("AAPL", "011070"),  # LG이노텍
+    ("AAPL", "009150"),  # 삼성전기
+    # 바이오/제약
+    ("PFE",  "068270"),  # 셀트리온
+    ("MRK",  "207940"),  # 삼성바이오로직스
+    # 조선/방산
+    ("LMT",  "079550"),  # LIG넥스원
+    ("LMT",  "329180"),  # HD현대중공업
+    # 철강/소재
+    ("NUE",  "005490"),  # POSCO홀딩스
+    # 화학
+    ("DOW",  "051910"),  # LG화학
+    ("LYB",  "011170"),  # 롯데케미칼
+    # 게임/엔터
+    ("NTES", "036570"),  # 엔씨소프트
+    ("NTES", "251270"),  # 넷마블
+    ("DIS",  "035250"),  # 강원랜드 (엔터)
+    # 유통
+    ("AMZN", "139480"),  # 이마트
+]
+
+# 자동 발굴용 미국 종목 후보 (S&P500 대표 + 한국 영향 큰 종목)
+DISCOVERY_US_TICKERS = [
+    # 반도체
+    "NVDA", "AMD", "MU", "INTC", "TSM", "AVGO", "QCOM", "TXN", "AMAT", "LRCX", "KLAC",
+    # 빅테크
+    "AAPL", "MSFT", "GOOGL", "AMZN", "META", "TSLA", "NFLX", "ORCL", "ADBE", "CRM",
+    # 금융
+    "JPM", "BAC", "GS", "MS", "WFC", "C", "BLK",
+    # 에너지/소재
+    "XOM", "CVX", "COP", "NUE", "FCX",
+    # 헬스/바이오
+    "JNJ", "PFE", "MRK", "ABBV", "LLY", "BMY", "GILD",
+    # 소비재/유통
+    "WMT", "COST", "HD", "MCD", "NKE", "SBUX", "DIS",
+    # 산업/방산
+    "BA", "LMT", "RTX", "GD", "CAT", "DE",
+    # 자동차
+    "F", "GM", "TM",
+    # 2차전지/리튬
+    "ALB", "RIVN",
+    # 화학
+    "DOW", "LYB",
+]
+
+# ─────────────────────────────────────────────────────────────
+# DB 초기화
+# ─────────────────────────────────────────────────────────────
+DDL = """
+CREATE TABLE IF NOT EXISTS us_market_daily (
+    ticker      VARCHAR(20),
+    trade_date  DATE,
+    open_price  DOUBLE PRECISION,
+    close_price DOUBLE PRECISION,
+    prev_close  DOUBLE PRECISION,
+    change_pct  DOUBLE PRECISION,
+    volume      BIGINT,
+    created_at  TIMESTAMP DEFAULT NOW(),
+    PRIMARY KEY (ticker, trade_date)
+);
+CREATE INDEX IF NOT EXISTS idx_us_daily_ticker ON us_market_daily(ticker, trade_date DESC);
+
+CREATE TABLE IF NOT EXISTS us_sector_etf_map (
+    etf_ticker      VARCHAR(20) PRIMARY KEY,
+    sector_keywords TEXT[],
+    description     TEXT,
+    updated_at      TIMESTAMP DEFAULT NOW()
+);
+
+CREATE TABLE IF NOT EXISTS us_kr_pairs (
+    us_ticker       VARCHAR(20),
+    kr_code         VARCHAR(10),
+    beta_60d        DOUBLE PRECISION,
+    correlation_60d DOUBLE PRECISION,
+    sample_size     INTEGER,
+    source          VARCHAR(20) DEFAULT 'seed',
+    updated_at      TIMESTAMP DEFAULT NOW(),
+    PRIMARY KEY (us_ticker, kr_code)
+);
+CREATE INDEX IF NOT EXISTS idx_pairs_kr ON us_kr_pairs(kr_code);
+
+CREATE TABLE IF NOT EXISTS us_overnight_signal (
+    kr_code            VARCHAR(10),
+    signal_date        DATE,
+    sector_adj         DOUBLE PRECISION DEFAULT 0,
+    pair_adj           DOUBLE PRECISION DEFAULT 0,
+    total_adj          DOUBLE PRECISION DEFAULT 0,
+    contributing_pairs JSONB,
+    created_at         TIMESTAMP DEFAULT NOW(),
+    PRIMARY KEY (kr_code, signal_date)
+);
+CREATE INDEX IF NOT EXISTS idx_overnight_date ON us_overnight_signal(signal_date DESC);
+"""
+
+# ─────────────────────────────────────────────────────────────
+# 시작/종료
+# ─────────────────────────────────────────────────────────────
+@app.on_event("startup")
+async def on_start():
+    global pg_pool
+    pg_pool = await asyncpg.create_pool(**PG, min_size=2, max_size=10)
+    async with pg_pool.acquire() as conn:
+        await conn.execute(DDL)
+    await seed_etfs_and_pairs()
+
+    scheduler.add_job(
+        collect_us_daily, CronTrigger(hour=7, minute=30),
+        id="us_collect", replace_existing=True)
+    scheduler.add_job(
+        calc_overnight_signals_all, CronTrigger(hour=8, minute=0),
+        id="overnight_calc", replace_existing=True)
+    scheduler.add_job(
+        recalc_pair_betas, CronTrigger(day_of_week="sun", hour=2, minute=0),
+        id="pair_beta", replace_existing=True)
+    scheduler.add_job(
+        discover_new_pairs, CronTrigger(day_of_week="sun", hour=3, minute=0),
+        id="pair_discover", replace_existing=True)
+    # 매일 06:00 Alpha Vantage 백필 (25 ticker/일, 60일+ 채워질 때까지)
+    scheduler.add_job(
+        backfill_yfinance, CronTrigger(hour=6, minute=0),
+        id="av_backfill", replace_existing=True)
+    scheduler.start()
+    logger.info("us-market.started")
+
+
+@app.on_event("shutdown")
+async def on_stop():
+    if scheduler.running:
+        scheduler.shutdown()
+    if pg_pool:
+        await pg_pool.close()
+
+
+# ─────────────────────────────────────────────────────────────
+# Seed: ETF + 핵심 페어 등록
+# ─────────────────────────────────────────────────────────────
+async def seed_etfs_and_pairs():
+    async with pg_pool.acquire() as conn:
+        for etf, meta in SECTOR_ETFS.items():
+            await conn.execute("""
+                INSERT INTO us_sector_etf_map (etf_ticker, sector_keywords, description)
+                VALUES ($1, $2, $3)
+                ON CONFLICT (etf_ticker) DO UPDATE
+                  SET sector_keywords=$2, description=$3, updated_at=NOW()
+            """, etf, meta["keywords"], meta["desc"])
+        for us, kr in SEED_PAIRS:
+            await conn.execute("""
+                INSERT INTO us_kr_pairs (us_ticker, kr_code, source)
+                VALUES ($1, $2, 'seed')
+                ON CONFLICT (us_ticker, kr_code) DO NOTHING
+            """, us, kr)
+    logger.info("seed.done", etfs=len(SECTOR_ETFS), pairs=len(SEED_PAIRS))
+
+
+# ─────────────────────────────────────────────────────────────
+# Finnhub 헬퍼 — /quote (무료 OK, /stock/candle은 2024년부터 유료)
+# 응답: c(current), pc(prev_close), dp(percent), o(open), h, l, t
+# Free tier: 60 calls/분
+# ─────────────────────────────────────────────────────────────
+async def fetch_finnhub_quote(client: httpx.AsyncClient, ticker: str
+                                ) -> Optional[dict]:
+    """Finnhub /quote 호출. 실패 시 None."""
+    if not FINNHUB_KEY:
+        return None
+    try:
+        r = await client.get(f"{FINNHUB_BASE}/quote", params={
+            "symbol": ticker, "token": FINNHUB_KEY,
+        }, timeout=15)
+    except Exception as e:
+        logger.warning("finnhub.req_err", ticker=ticker, err=str(e))
+        return None
+    if r.status_code == 429:
+        await asyncio.sleep(5)
+        return None
+    if r.status_code != 200:
+        return None
+    try:
+        j = r.json()
+    except Exception:
+        return None
+    # 휴장/잘못된 ticker 시 c=0
+    if not j or j.get("c", 0) <= 0:
+        return None
+    return j
+
+
+# ─────────────────────────────────────────────────────────────
+# 수집: /quote으로 미국 일간 종가 누적
+# 매일 호출하면 us_market_daily에 시계열 점진 누적 → 베타 학습 가능
+# ─────────────────────────────────────────────────────────────
+async def collect_us_daily(days: int = 1):
+    """매일 KST 07:30 호출. days 인자는 호환용(무시) — /quote은 단일 시점."""
+    if not FINNHUB_KEY:
+        logger.error("us.no_api_key")
+        return {"saved": 0, "err": "FINNHUB_API_KEY missing — set in .env"}
+
+    tickers = sorted(
+        set(SECTOR_ETFS.keys())
+        | {t for t, _ in SEED_PAIRS}
+        | set(DISCOVERY_US_TICKERS)
+    )
+    saved = 0
+    failed: List[str] = []
+    async with httpx.AsyncClient() as client:
+        for i, ticker in enumerate(tickers):
+            q = await fetch_finnhub_quote(client, ticker)
+            if not q:
+                failed.append(ticker)
+            else:
+                # t 타임스탬프(미국 ET 종가 시점) → trade_date
+                ts = q.get("t", 0)
+                trade_dt = (datetime.fromtimestamp(ts).date() if ts
+                            else date.today() - timedelta(days=1))
+                async with pg_pool.acquire() as conn:
+                    await conn.execute("""
+                        INSERT INTO us_market_daily
+                          (ticker, trade_date, open_price, close_price,
+                           prev_close, change_pct, volume)
+                        VALUES ($1, $2, $3, $4, $5, $6, $7)
+                        ON CONFLICT (ticker, trade_date) DO UPDATE
+                          SET open_price=$3, close_price=$4, prev_close=$5,
+                              change_pct=$6
+                    """, ticker, trade_dt,
+                        float(q.get("o", 0) or 0), float(q["c"]),
+                        float(q.get("pc", 0) or 0), float(q.get("dp", 0) or 0),
+                        0)
+                    saved += 1
+            # Rate limit: 60/분 → 1.1초 간격
+            if i < len(tickers) - 1:
+                await asyncio.sleep(1.1)
+    logger.info("us.collected", rows=saved, ok=len(tickers) - len(failed),
+                failed=len(failed))
+    return {"saved": saved, "tickers": len(tickers),
+            "ok": len(tickers) - len(failed), "failed": failed[:10]}
+
+
+# ─────────────────────────────────────────────────────────────
+# Alpha Vantage 백필 — TIME_SERIES_DAILY로 ticker당 100일 히스토리.
+# Free tier 25 calls/day → 자동으로 매일 25개씩 분할 처리 (75 ticker × 3일).
+# ─────────────────────────────────────────────────────────────
+async def fetch_av_daily(client: httpx.AsyncClient, ticker: str
+                          ) -> Optional[List[dict]]:
+    """Alpha Vantage TIME_SERIES_DAILY 호출. 실패/한도초과 시 None."""
+    try:
+        r = await client.get(AV_BASE, params={
+            "function": "TIME_SERIES_DAILY",
+            "symbol": ticker,
+            "outputsize": "compact",  # 100일치 (full=20+년)
+            "apikey": AV_KEY,
+        }, timeout=20)
+    except Exception as e:
+        logger.warning("av.req_err", ticker=ticker, err=str(e))
+        return None
+    if r.status_code != 200:
+        return None
+    try:
+        j = r.json()
+    except Exception:
+        return None
+    if "Note" in j or "Information" in j or "Error Message" in j:
+        logger.warning("av.limit_or_err",
+                       ticker=ticker, msg=str(j)[:200])
+        return None
+    ts = j.get("Time Series (Daily)")
+    if not ts:
+        return None
+    rows = []
+    for dt_str, ohlcv in ts.items():
+        try:
+            rows.append({
+                "trade_date": date.fromisoformat(dt_str),
+                "open":   float(ohlcv["1. open"]),
+                "high":   float(ohlcv["2. high"]),
+                "low":    float(ohlcv["3. low"]),
+                "close":  float(ohlcv["4. close"]),
+                "volume": int(float(ohlcv["5. volume"])),
+            })
+        except (KeyError, ValueError):
+            continue
+    rows.sort(key=lambda x: x["trade_date"])
+    return rows
+
+
+async def backfill_yfinance(days: int = 180, max_tickers: int = 0):
+    """Alpha Vantage로 히스토리 백필.
+    - max_tickers=0 (기본): AV_DAILY_LIMIT(=25)개만 처리 → 일일 한도 자동 준수
+    - max_tickers>0: 명시값 사용
+    - 이미 60일+ 데이터 있는 ticker는 건너뜀 → 3일치 분산 자동 진행
+    """
+    if not AV_KEY:
+        return {"saved": 0, "err": "ALPHAVANTAGE_API_KEY missing — set in .env"}
+    all_tickers = sorted(
+        set(SECTOR_ETFS.keys())
+        | {t for t, _ in SEED_PAIRS}
+        | set(DISCOVERY_US_TICKERS)
+    )
+    # 이미 60일+ 누적된 ticker는 스킵
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch(
+            "SELECT ticker, COUNT(*) AS n FROM us_market_daily "
+            "GROUP BY ticker HAVING COUNT(*) >= 60")
+        done = {r["ticker"] for r in rows}
+    pending = [t for t in all_tickers if t not in done]
+    limit = max_tickers if max_tickers > 0 else AV_DAILY_LIMIT
+    targets = pending[:limit]
+
+    saved = 0
+    failed: List[str] = []
+    async with httpx.AsyncClient() as client:
+        for i, ticker in enumerate(targets):
+            rows = await fetch_av_daily(client, ticker)
+            if not rows:
+                failed.append(ticker)
+                # 한도 초과면 즉시 중단
+                if i > 0 and len(failed) > 3 and len(failed) > i // 2:
+                    logger.warning("av.likely_quota", processed=i)
+                    break
+            else:
+                prev_close = None
+                async with pg_pool.acquire() as conn:
+                    async with conn.transaction():
+                        for row in rows:
+                            pc = prev_close if prev_close is not None else 0.0
+                            dp = ((row["close"] - pc) / pc * 100.0) if pc > 0 else 0.0
+                            await conn.execute("""
+                                INSERT INTO us_market_daily
+                                  (ticker, trade_date, open_price, close_price,
+                                   prev_close, change_pct, volume)
+                                VALUES ($1, $2, $3, $4, $5, $6, $7)
+                                ON CONFLICT (ticker, trade_date) DO UPDATE
+                                  SET open_price=$3, close_price=$4, prev_close=$5,
+                                      change_pct=$6, volume=$7
+                            """, ticker, row["trade_date"], row["open"],
+                                row["close"], pc, dp, row["volume"])
+                            saved += 1
+                            prev_close = row["close"]
+            # AV free tier 5 calls/min → 12초 간격
+            if i < len(targets) - 1:
+                await asyncio.sleep(12.5)
+
+    logger.info("us.av_backfill", saved=saved,
+                processed=len(targets) - len(failed),
+                failed=len(failed),
+                pending_remaining=len(pending) - len(targets) + len(failed))
+    return {"saved": saved, "processed": len(targets),
+            "ok": len(targets) - len(failed),
+            "failed": failed[:10],
+            "pending_after": max(0, len(pending) - len(targets) + len(failed)),
+            "source": "alphavantage"}
+
+
+# ─────────────────────────────────────────────────────────────
+# 시그널 계산: 한국 종목별 overnight 보정 점수
+# ─────────────────────────────────────────────────────────────
+async def calc_overnight_signals_all(target_date: Optional[date] = None):
+    """오늘 자(=어제 미국장 마감) 보정 점수 계산.
+
+    A. 섹터 ETF 동조: ETF change_pct → 같은 sector 한국 종목에 ±5점
+    B. 페어 동조: 페어별 미국주 change_pct × beta → ±10점 (집계)
+    """
+    target = target_date or date.today()
+    async with pg_pool.acquire() as conn:
+        # 1) 최신 미국 거래일의 ETF/주식 change_pct
+        us_rows = await conn.fetch("""
+            SELECT DISTINCT ON (ticker)
+              ticker, trade_date, change_pct
+            FROM us_market_daily
+            WHERE trade_date <= $1 AND change_pct IS NOT NULL
+            ORDER BY ticker, trade_date DESC
+        """, target)
+        us_chg = {r["ticker"]: float(r["change_pct"]) for r in us_rows}
+        if not us_chg:
+            logger.warning("overnight.no_us_data")
+            return {"saved": 0, "err": "no us data"}
+
+        # 2) 활성 한국 종목 + 섹터
+        kr_rows = await conn.fetch("""
+            SELECT stock_code, sector FROM dart_corps WHERE is_active=true
+        """)
+
+        # 3) 페어 매핑 (us_ticker → list of (kr_code, beta))
+        pair_rows = await conn.fetch("""
+            SELECT us_ticker, kr_code, beta_60d, correlation_60d
+            FROM us_kr_pairs
+        """)
+        pairs_by_kr: Dict[str, List] = {}
+        for r in pair_rows:
+            kr = r["kr_code"]
+            pairs_by_kr.setdefault(kr, []).append({
+                "us": r["us_ticker"],
+                "beta": float(r["beta_60d"]) if r["beta_60d"] else 1.0,
+                "corr": float(r["correlation_60d"]) if r["correlation_60d"] else 0.0,
+            })
+
+        # 4) ETF 매핑
+        etf_rows = await conn.fetch("SELECT etf_ticker, sector_keywords FROM us_sector_etf_map")
+
+        saved = 0
+        for kr in kr_rows:
+            code = kr["stock_code"]
+            sector = (kr["sector"] or "")
+
+            # A. 섹터 ETF 동조
+            sector_adj = 0.0
+            matched_etfs = []
+            for er in etf_rows:
+                kws = er["sector_keywords"] or []
+                if not sector or not kws:
+                    continue
+                if any(kw and kw in sector for kw in kws):
+                    pct = us_chg.get(er["etf_ticker"])
+                    if pct is None:
+                        continue
+                    matched_etfs.append({"etf": er["etf_ticker"], "pct": pct})
+            if matched_etfs:
+                # 매칭된 ETF 평균 변동률 → ±5 클램프
+                avg_pct = sum(m["pct"] for m in matched_etfs) / len(matched_etfs)
+                sector_adj = max(-5.0, min(5.0, avg_pct * 1.5))
+
+            # B. 페어 베타 기반
+            pair_adj = 0.0
+            contributing = []
+            kr_pairs = pairs_by_kr.get(code, [])
+            for p in kr_pairs:
+                pct = us_chg.get(p["us"])
+                if pct is None:
+                    continue
+                # 예상 갭 = 미국주 변동률 × 베타
+                exp_gap = pct * p["beta"]
+                # 상관계수 가중 (|corr|이 낮으면 신뢰도 ↓)
+                weight = max(0.3, abs(p["corr"])) if p["corr"] else 0.5
+                contrib = exp_gap * weight
+                pair_adj += contrib
+                contributing.append({
+                    "us": p["us"], "pct": round(pct, 2),
+                    "beta": round(p["beta"], 2),
+                    "corr": round(p["corr"], 2),
+                    "contribution": round(contrib, 2),
+                })
+            if contributing:
+                # 다중 페어 평균 + 클램프
+                pair_adj = max(-10.0, min(10.0, pair_adj / len(contributing) * 2.0))
+
+            total_adj = sector_adj + pair_adj
+            if abs(total_adj) < 0.1 and not matched_etfs and not contributing:
+                continue  # 영향 없는 종목은 저장 스킵
+
+            await conn.execute("""
+                INSERT INTO us_overnight_signal
+                  (kr_code, signal_date, sector_adj, pair_adj, total_adj, contributing_pairs)
+                VALUES ($1, $2, $3, $4, $5, $6)
+                ON CONFLICT (kr_code, signal_date) DO UPDATE
+                  SET sector_adj=$3, pair_adj=$4, total_adj=$5,
+                      contributing_pairs=$6, created_at=NOW()
+            """, code, target, sector_adj, pair_adj, total_adj,
+                json.dumps({"etfs": matched_etfs, "pairs": contributing}))
+            saved += 1
+    logger.info("overnight.calculated", saved=saved, date=str(target))
+    return {"saved": saved, "date": str(target)}
+
+
+# ─────────────────────────────────────────────────────────────
+# 페어 베타 재계산 (주 1회 일요일)
+# ─────────────────────────────────────────────────────────────
+async def recalc_pair_betas(window_days: int = 60):
+    """등록된 페어에 대해 최근 N일 일간수익률로 회귀 → beta, correlation 갱신."""
+    since = date.today() - timedelta(days=window_days * 2)  # 거래일 여유
+
+    async with pg_pool.acquire() as conn:
+        pairs = await conn.fetch("SELECT us_ticker, kr_code FROM us_kr_pairs")
+        updated, skipped = 0, 0
+        for p in pairs:
+            us_t, kr_c = p["us_ticker"], p["kr_code"]
+            # 미국 시계열
+            us_rows = await conn.fetch("""
+                SELECT trade_date, close_price FROM us_market_daily
+                WHERE ticker=$1 AND trade_date >= $2 ORDER BY trade_date
+            """, us_t, since)
+            if len(us_rows) < 30:
+                skipped += 1
+                continue
+            # 한국 시계열 (stock_prices.collected_at)
+            kr_rows = await conn.fetch("""
+                SELECT collected_at::date AS dt,
+                       AVG(price)::float AS close
+                FROM stock_prices
+                WHERE stock_code=$1 AND collected_at::date >= $2
+                GROUP BY collected_at::date ORDER BY dt
+            """, kr_c, since)
+            if len(kr_rows) < 30:
+                skipped += 1
+                continue
+
+            us_df = pd.DataFrame([(r["trade_date"], r["close_price"]) for r in us_rows],
+                                  columns=["dt", "us"])
+            kr_df = pd.DataFrame([(r["dt"], r["close"]) for r in kr_rows],
+                                  columns=["dt", "kr"])
+            us_df["dt"] = pd.to_datetime(us_df["dt"])
+            kr_df["dt"] = pd.to_datetime(kr_df["dt"])
+
+            # 한국 종가는 미국 다음날에 영향 받음 → 미국 시계열을 +1일 시프트해서 매칭
+            us_df["dt"] = us_df["dt"] + pd.Timedelta(days=1)
+            merged = pd.merge(us_df, kr_df, on="dt", how="inner")
+            if len(merged) < window_days // 2:
+                skipped += 1
+                continue
+
+            us_ret = merged["us"].pct_change().dropna()
+            kr_ret = merged["kr"].pct_change().dropna()
+            n = min(len(us_ret), len(kr_ret))
+            if n < 20:
+                skipped += 1
+                continue
+            us_ret, kr_ret = us_ret.iloc[-n:].values, kr_ret.iloc[-n:].values
+
+            # 회귀: kr_ret = beta * us_ret + intercept
+            slope, _, r_val, _, _ = stats.linregress(us_ret, kr_ret)
+            beta = float(slope)
+            corr = float(r_val)
+            await conn.execute("""
+                UPDATE us_kr_pairs
+                  SET beta_60d=$1, correlation_60d=$2, sample_size=$3, updated_at=NOW()
+                WHERE us_ticker=$4 AND kr_code=$5
+            """, beta, corr, n, us_t, kr_c)
+            updated += 1
+    logger.info("pair_beta.recalc", updated=updated, skipped=skipped)
+    return {"updated": updated, "skipped": skipped}
+
+
+# ─────────────────────────────────────────────────────────────
+# 자동 페어 발굴 (월 1회)
+# 코스피200 시총상위 50개 × DISCOVERY_US 후보 → |corr|≥0.5인 것만 등록
+# ─────────────────────────────────────────────────────────────
+async def discover_new_pairs(min_abs_corr: float = 0.5, top_kr: int = 50,
+                              window_days: int = 60):
+    since = date.today() - timedelta(days=window_days * 2)
+    async with pg_pool.acquire() as conn:
+        # 시총상위 한국 종목
+        top_rows = await conn.fetch("""
+            SELECT DISTINCT ON (stock_code) stock_code, market_cap
+            FROM stock_prices
+            WHERE collected_at::date >= CURRENT_DATE - INTERVAL '7 days'
+              AND market_cap > 0
+            ORDER BY stock_code, collected_at DESC
+        """)
+        kr_top = sorted(top_rows, key=lambda r: -(r["market_cap"] or 0))[:top_kr]
+        added = 0
+        for kr in kr_top:
+            kr_c = kr["stock_code"]
+            kr_rows = await conn.fetch("""
+                SELECT collected_at::date AS dt, AVG(price)::float AS close
+                FROM stock_prices
+                WHERE stock_code=$1 AND collected_at::date >= $2
+                GROUP BY collected_at::date ORDER BY dt
+            """, kr_c, since)
+            if len(kr_rows) < 30:
+                continue
+            kr_df = pd.DataFrame([(r["dt"], r["close"]) for r in kr_rows],
+                                  columns=["dt", "kr"])
+            kr_df["dt"] = pd.to_datetime(kr_df["dt"])
+
+            for us_t in DISCOVERY_US_TICKERS:
+                us_rows = await conn.fetch("""
+                    SELECT trade_date, close_price FROM us_market_daily
+                    WHERE ticker=$1 AND trade_date >= $2 ORDER BY trade_date
+                """, us_t, since)
+                if len(us_rows) < 30:
+                    continue
+                us_df = pd.DataFrame([(r["trade_date"], r["close_price"]) for r in us_rows],
+                                      columns=["dt", "us"])
+                us_df["dt"] = pd.to_datetime(us_df["dt"]) + pd.Timedelta(days=1)
+                merged = pd.merge(us_df, kr_df, on="dt", how="inner")
+                if len(merged) < window_days // 2:
+                    continue
+                us_ret = merged["us"].pct_change().dropna()
+                kr_ret = merged["kr"].pct_change().dropna()
+                n = min(len(us_ret), len(kr_ret))
+                if n < 20:
+                    continue
+                slope, _, r_val, _, _ = stats.linregress(
+                    us_ret.iloc[-n:].values, kr_ret.iloc[-n:].values)
+                if abs(r_val) < min_abs_corr:
+                    continue
+                await conn.execute("""
+                    INSERT INTO us_kr_pairs
+                      (us_ticker, kr_code, beta_60d, correlation_60d, sample_size, source)
+                    VALUES ($1, $2, $3, $4, $5, 'discovered')
+                    ON CONFLICT (us_ticker, kr_code) DO UPDATE
+                      SET beta_60d=$3, correlation_60d=$4, sample_size=$5, updated_at=NOW()
+                """, us_t, kr_c, float(slope), float(r_val), n)
+                added += 1
+    logger.info("pair_discover.done", added=added)
+    return {"added": added}
+
+
+# ─────────────────────────────────────────────────────────────
+# REST API
+# ─────────────────────────────────────────────────────────────
+@app.get("/health")
+async def health():
+    return {"ok": True, "service": "us-market", "ts": datetime.now(KST).isoformat()}
+
+
+@app.post("/collect")
+async def manual_collect(days: int = Query(default=7, ge=1, le=365),
+                          bg: BackgroundTasks = None):
+    """수동 수집. days=7 일상, 백필은 180 권장."""
+    if bg:
+        bg.add_task(collect_us_daily, days)
+        return {"status": "queued", "days": days}
+    return await collect_us_daily(days)
+
+
+@app.post("/collect/backfill")
+async def backfill(days: int = Query(default=180, ge=30, le=365),
+                    bg: BackgroundTasks = None):
+    """대규모 백필 — 페어 베타 학습용 180일 권장. 76 ticker × 1.1초 ≈ 85초."""
+    if bg:
+        bg.add_task(collect_us_daily, days)
+        return {"status": "queued", "days": days}
+    return await collect_us_daily(days)
+
+
+@app.post("/collect/yfinance-backfill")
+async def yfinance_backfill_ep(days: int = Query(default=180, ge=30, le=730),
+                                bg: BackgroundTasks = None):
+    """yfinance로 일괄 히스토리 다운로드 (페어 베타 학습 초기 시드용)."""
+    if bg:
+        bg.add_task(backfill_yfinance, days)
+        return {"status": "queued", "days": days}
+    return await backfill_yfinance(days)
+
+
+@app.post("/signal/calculate")
+async def manual_signal(target: Optional[str] = None):
+    d = date.fromisoformat(target) if target else date.today()
+    return await calc_overnight_signals_all(d)
+
+
+@app.get("/signal/{kr_code}")
+async def get_signal(kr_code: str, days: int = Query(default=7, ge=1, le=90)):
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT signal_date, sector_adj, pair_adj, total_adj, contributing_pairs
+            FROM us_overnight_signal
+            WHERE kr_code=$1 AND signal_date >= CURRENT_DATE - $2::int
+            ORDER BY signal_date DESC
+        """, kr_code, days)
+        return [dict(r) for r in rows]
+
+
+@app.get("/signal/latest")
+async def latest_signals(limit: int = Query(default=50, le=500)):
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("""
+            SELECT DISTINCT ON (kr_code)
+              kr_code, signal_date, sector_adj, pair_adj, total_adj
+            FROM us_overnight_signal
+            ORDER BY kr_code, signal_date DESC
+            LIMIT $1
+        """, limit)
+        return [dict(r) for r in rows]
+
+
+@app.get("/pairs")
+async def list_pairs(kr_code: Optional[str] = None, us_ticker: Optional[str] = None):
+    async with pg_pool.acquire() as conn:
+        if kr_code:
+            rows = await conn.fetch(
+                "SELECT * FROM us_kr_pairs WHERE kr_code=$1 ORDER BY ABS(correlation_60d) DESC NULLS LAST",
+                kr_code)
+        elif us_ticker:
+            rows = await conn.fetch(
+                "SELECT * FROM us_kr_pairs WHERE us_ticker=$1 ORDER BY ABS(correlation_60d) DESC NULLS LAST",
+                us_ticker)
+        else:
+            rows = await conn.fetch(
+                "SELECT * FROM us_kr_pairs ORDER BY ABS(correlation_60d) DESC NULLS LAST LIMIT 200")
+        return [dict(r) for r in rows]
+
+
+@app.post("/pairs/recalc-beta")
+async def manual_recalc(window: int = Query(default=60, ge=20, le=250)):
+    return await recalc_pair_betas(window)
+
+
+@app.post("/pairs/discover")
+async def manual_discover(min_corr: float = Query(default=0.5, ge=0.1, le=0.9),
+                           top_kr: int = Query(default=50, ge=10, le=200)):
+    return await discover_new_pairs(min_corr, top_kr)
+
+
+@app.get("/etfs")
+async def list_etfs():
+    async with pg_pool.acquire() as conn:
+        rows = await conn.fetch("SELECT * FROM us_sector_etf_map ORDER BY etf_ticker")
+        return [dict(r) for r in rows]
+
+
+@app.get("/etfs/{etf}/latest")
+async def etf_latest(etf: str):
+    async with pg_pool.acquire() as conn:
+        row = await conn.fetchrow("""
+            SELECT * FROM us_market_daily WHERE ticker=$1
+            ORDER BY trade_date DESC LIMIT 1
+        """, etf.upper())
+        return dict(row) if row else {"err": "no data"}
+
+
+@app.get("/stats")
+async def stats_endpoint():
+    async with pg_pool.acquire() as conn:
+        r1 = await conn.fetchrow(
+            "SELECT COUNT(*) AS rows, COUNT(DISTINCT ticker) AS tickers,"
+            " MIN(trade_date) AS earliest, MAX(trade_date) AS latest FROM us_market_daily")
+        r2 = await conn.fetchrow(
+            "SELECT COUNT(*) AS total, COUNT(*) FILTER (WHERE source='seed') AS seed,"
+            " COUNT(*) FILTER (WHERE source='discovered') AS discovered,"
+            " COUNT(*) FILTER (WHERE beta_60d IS NOT NULL) AS with_beta"
+            " FROM us_kr_pairs")
+        r3 = await conn.fetchrow(
+            "SELECT COUNT(*) AS rows, COUNT(DISTINCT kr_code) AS codes,"
+            " MAX(signal_date) AS latest FROM us_overnight_signal")
+        return {"us_daily": dict(r1), "pairs": dict(r2), "signals": dict(r3)}
@@ -0,0 +1,11 @@
+fastapi==0.111.0
+uvicorn[standard]==0.30.1
+asyncpg==0.29.0
+apscheduler==3.10.4
+structlog==24.2.0
+orjson==3.10.3
+httpx==0.27.0
+pandas==2.2.2
+numpy==1.26.4
+scipy==1.13.1
+pytz==2024.1
				`@@ -0,0 +1 @@`
				{"id":"669c2fcf-b774-40a1-b874-21bbb5027474","name":"📈 AI 예측 성과 추적 (매일 16:10)","nodes":[{"id":"n1","name":"매일 16:10","type":"n8n-nodes-base.scheduleTrigger","typeVersion":1.1,"position":[100,300],"parameters":{"rule":{"interval":[{"field":"cronExpression","expression":"10 16 * * 1-5"}]}}},{"id":"n2","name":"추천종목 랭킹조회","type":"n8n-nodes-base.httpRequest","typeVersion":4.2,"position":[350,300],"parameters":{"method":"GET","url":"http://score-engine:8686/recommendations","options":{"timeout":30000}}},{"id":"n3","name":"오늘 성과 브리핑","type":"n8n-nodes-base.httpRequest","typeVersion":4.2,"position":[600,300],"parameters":{"method":"POST","url":"http://score-engine:8686/briefing/send","options":{"timeout":30000}}}],"connections":{"매일 16:10":{"main":[[{"node":"추천종목 랭킹조회","type":"main","index":0}]]},"추천종목 랭킹조회":{"main":[[{"node":"오늘 성과 브리핑","type":"main","index":0}]]}},"active":true}