docs(web-ai): rewrite CLAUDE.md with Phase 0-4 complete context

Replaces Phase-2-era placeholder. Adds:
- V1 (signal_v1 :8000 LSTM bot) vs V2 (signal_v2 :8001 Confidence Pipeline) split
- start.bat invocation for each + KIS rate limit warning (do NOT run both)
- Phase 0-7 status table, Phase 4 completed 2026-05-17
- signal_v2/ module-level inventory + new test count (56)
- Phase 4 buy/sell rule summary (absolute spread amendment included)
- 11 known traps + Phase 7 backlog
- Cross-repo workflow note (code in web-ai, spec/plan in web-ui)

CLAUDE.md

@@ -1,24 +1,141 @@
 # web-ai — Workspace 가이드
 
-Windows AI 머신 (AMD 9800X3D + RTX 5070 Ti) 의 두 시그널 파이프라인 컨테이너.
+Windows AI 머신 (AMD 9800X3D + RTX 5070 Ti 16GB) 의 두 신호 파이프라인.
+**Confidence Signal Pipeline V2 의 Windows-side 구현체** (NAS stock 백엔드와 HTTP 연동).
+
+상위 워크스페이스 컨텍스트는 `../CLAUDE.md` 참조.
+
+---
 
 ## 디렉토리 구조
 
-| 경로 | 역할 | 상태 |
-|------|------|------|
-| `signal_v1/` | V1 자체 자동매매 시스템 (main_server.py + Trading Bot + Telegram Bot + LSTM + Ollama + KIS 자동주문) | 운영 중. Confidence Signal Pipeline V2 Phase 6 에서 deprecation 예정 |
-| `signal_v2/` | V2 신호 파이프라인 (stock pull worker + Chronos-2 + signal API client) | Phase 2 에서 신설 |
-| `.env` | V1 + V2 환경변수 공유 | KIS_*, TELEGRAM_*, STOCK_API_URL, WEBAI_API_KEY 등 |
-| `start.bat` | V1 진입 (signal_v1 디렉토리 안 main_server.py 실행) | V2 별도 start 스크립트는 signal_v2/start.bat |
+| 경로 | 역할 | 포트 | 상태 |
+|------|------|------|------|
+| `signal_v1/` | 레거시 자동매매 시스템 (LSTM 7-features + Gemini Flash + Telegram Bot + KIS 자동주문) | `:8000` | 운영 중. **V2 Phase 6 에서 deprecation 예정** |
+| `signal_v2/` | Confidence Signal Pipeline V2 (Chronos-bolt + 분봉 모멘텀 + KIS WebSocket + 신호 생성) | `:8001` | **Phase 4 완료 (2026-05-17)**, Phase 5 대기 |
+| `.env` | V1 + V2 환경변수 공유 | — | `KIS_REAL_*`, `TELEGRAM_*`, `STOCK_API_URL`, `WEBAI_API_KEY`, `LOG_LEVEL` |
+| `start.bat` | V1 진입점 | — | `signal_v1/main_server.py` 실행 |
+| `signal_v2/start.bat` | V2 진입점 | — | `signal_v2/main.py` uvicorn 실행 |
+| `requirements.txt` | 공용 의존성 | — | torch, chronos-forecasting, fastapi, httpx, websockets 등 |
 
-## 운영 가이드
+`.venv` 는 **구조적으로 깨짐**: `pyvenv.cfg` 가 한글 사용자 경로(`C:\Users\박재오\...`) 를 포함하여 콘솔 코드페이지가 roundtrip 못함. 테스트는 시스템 Python 으로 실행: `C:\Users\jaeoh\AppData\Local\Programs\Python\Python312\python.exe -m pytest signal_v2/tests -q`.
 
-- V1 시작: `start.bat` 또는 `cd signal_v1 && python main_server.py`
-- V2 시작 (Phase 2 이후): `cd signal_v2 && python -m uvicorn main:app --port 8001`
-- 둘 다 동시 실행 가능 (포트 분리: V1=8000, V2=8001)
+---
+
+## 서버 시작 방식
+
+### V1 단독 (운영 기본)
+```bat
+cd C:\Users\jaeoh\Desktop\workspace\web-ai
+.\start.bat
+```
+기대 로그: `[Bot] Cycle Start ...`, `[AI] 005930: NN epochs ...`, `[Ensemble] tech=... news=... lstm=...`, `Score: 0.xx [HOLD]`
+
+### V2 단독 (smoke/검증)
+```bat
+cd C:\Users\jaeoh\Desktop\workspace\web-ai\signal_v2
+.\start.bat
+```
+기대 로그: `Uvicorn running on http://0.0.0.0:8001`, `poll_loop started`, `[KIS] minute bars ... OK`, `[Chronos] predicted N tickers`, `signal emit XXXXXX buy conf=0.xxx`.
+
+휴장일/장 외 시간엔 `poll_loop` 만 idle. `Application startup complete` 만 보이면 정상.
+
+### V1 + V2 동시 실행 — **권장 안 함**
+**KIS app_key 초당 2회 한도 (EGW00201)** 충돌. V1 cycle + V2 분봉 cron 이 같은 KIS app_key 로 동시 호출하면 rate limit. 채택 해결책: V2 임시 종료 (Phase 3a 결정), Phase 6 V1 deprecation 시 자연 해소. 별도 app_key 발급은 옵션 B.
+
+---
 
 ## Phase 진행 상태 (Confidence Signal Pipeline V2)
 
-`web-ui/docs/superpowers/specs/2026-05-15-confidence-signal-pipeline-v2-architecture.md` 참조.
+| Phase | 내용 | 상태 |
+|-------|------|------|
+| 0 | Architecture & contract spec | ✅ Chronos-2 + Qwen3 14B 채택 |
+| 1 | stock 백엔드 WebAI API 보강 (NAS) | ✅ 102/102 tests, 운영 배포 |
+| 1.5 | V1 → `signal_v1/` rename | ✅ V1 정상 기동 |
+| 2 | signal_v2 pull worker + signal API client + scheduler | ✅ 19/19 tests, `:8001` 기동 |
+| 3a | KIS REST 분봉 + WebSocket 호가 + NXT 스케줄 | ✅ 33/33 tests |
+| 3b | Chronos-bolt-base 추론 + 5분봉 모멘텀 분류기 | ✅ 45/45 tests, 실 KIS+Chronos chain 검증 |
+| 4 | Signal Generator (매수/매도 룰) + pull_worker 통합 + 로깅 | ✅ **2026-05-17 완료, 56/56 tests, push 완료** |
+| 5 | agent-office `/signal` + Ollama Qwen3 14B + 이중 텔레그램 | ⏳ 2주 예상 |
+| 6 | signal_v1 deprecation | ⏳ 1주 |
+| 7 | 운영 모니터링 + 4주 IC 검증 | ⏳ 1주 + 4주 |
 
-자세한 V1 가이드는 `signal_v1/CLAUDE.md` 참조.
+상세 spec/plan: `../web-ui/docs/superpowers/specs/` 및 `../web-ui/docs/superpowers/plans/` (web-ui repo 안에 보관됨 — V2 자체 코드와 분리 보관).
+
+---
+
+## signal_v2 디렉토리 내부
+
+| 파일 | 역할 |
+|------|------|
+| `main.py` | FastAPI app + lifespan (StockClient + KISClient + KISWebSocket + ChronosPredictor + SignalDedup 초기화). poll_loop task 생성 |
+| `config.py` | Settings dataclass — 환경변수 로드. Phase 4 추가 6 필드: `stop_loss_pct`, `take_profit_pct`, `chronos_spread_threshold`, `asking_bid_ratio_threshold`, `confidence_threshold`, `min_momentum_for_buy` |
+| `state.py` | PollState (process-wide singleton) — portfolio, screener_preview, news_sentiment, chronos_predictions, minute_bars, asking_price, **signals** (Phase 4) |
+| `stock_client.py` | NAS stock 백엔드 pull (X-WebAI-Key + 메모리 cache 60s/300s/60s + retry) |
+| `kis_client.py` | KIS REST 분봉/호가 — V1 토큰 read-only 공유 (mtime cache) + 초당 2회 throttle + 지수 backoff |
+| `kis_websocket.py` | KIS WebSocket H0STASP0 호가 + approval_key + 재연결 (1→2→4→max 30s) |
+| `chronos_predictor.py` | `amazon/chronos-bolt-base` zero-shot quantile (FP32 강제 — FP16 overflow 회피) |
+| `minute_momentum.py` | 5분봉 → strong_up/weak_up/neutral/weak_down/strong_down 5단계 분류 |
+| `signal_generator.py` | **Phase 4 — 매수/매도 룰 엔진**. `generate_signals(state, dedup, settings)` 진입. sell-first → buy 순서. 신호 emit/skip INFO/DEBUG 로그 |
+| `pull_worker.py` | asyncio cron — 장전 5분 / 장중 1분 / 장후 5분 / NXT / dead zone skip. cycle 끝에 `generate_signals` 호출 |
+| `scheduler.py` | polling window 판정 (KST 캘린더 + 휴장일) |
+| `rate_limit.py` | 초당 N회 token bucket |
+| `dedup.py` | SignalDedup SQLite WAL — `(ticker, action)` PK 24h |
+| `tests/` | 56 tests (pytest + respx HTTP mock + monkeypatch) |
+| `data/` | dedup.db (SQLite WAL) + `holidays.json` (NAS stock 에서 manual copy) |
+| `start.bat` | V2 진입 |
+
+---
+
+## 신호 룰 요약 (Phase 4)
+
+### 매수 (screener Top-N + portfolio, sell 신호 받은 종목은 skip)
+모두 충족:
+1. `chronos.median > 0`
+2. **`chronos.q90 - chronos.q10 < 0.6`** (absolute spread — 2026-05-17 spec amend, 기존 relative formula 가 zero-shot median≈0 빈번에서 모든 신호 거부)
+3. `minute_momentum == strong_up` (env 로 조정 가능)
+4. `asking_price.bid_ratio >= 0.6`
+
+종합 confidence = `chronos_conf * 0.5 + minute_score * 0.3 + screener_norm * 0.2`. `> 0.7` 시 emit.
+
+### 매도 (portfolio only, 우선순위 stop_loss → anomaly → take_profit)
+- **stop_loss**: `pnl_pct < -7%` 즉시 (confidence=1.0)
+- **anomaly**: `chronos.median < -1%` + `strong_down` + `bid_ratio < 0.4` + 종합 conf > 0.7
+- **take_profit**: `pnl_pct > 15%` 검토 (confidence=0.6)
+
+---
+
+## 알려진 함정 / Phase 7 백로그
+
+1. **KIS rate limit (EGW00201)** — V1+V2 동시 실행 시 충돌. Phase 6 자연 해소
+2. **`.venv` 한글 경로 깨짐** — 시스템 Python 사용
+3. **Chronos FP16 overflow** — 한국 주가 5만+ 시 inf. FP32 강제 (`chronos_predictor.py:39-41`)
+4. **`predict_quantiles` positional `inputs`** — ChronosBolt API 새 변경. `try/except TypeError` fallback 처리됨
+5. **`state.signals` consumer-drain protocol 미정의** — Phase 5 prereq. dict 무한 누적 위험 (실제로는 bounded by unique ticker count)
+6. **integration test 가 poll_loop 실제 호출 안 함** — `test_pull_worker.py:test_poll_loop_calls_generate_signals_after_cycle` 가 `generate_signals` 직접 호출. Phase 7 hardening 시 mock-iteration 으로 강화
+7. **KIS WebSocket URL `ws://ops.koreainvestment.com:21000/31000`** — 첫 운영 시 실제 KIS API docs 와 대조 필요
+8. **`_parse_asking_price` 필드 인덱스** — 마지막 2 필드 가정. 실 운영 raw 메시지 캡처 후 매핑 검증 필요
+9. **`holidays.json` 자동 동기화 부재** — NAS stock 의 `holidays.json` 을 수동 copy
+10. **schema rename** — Phase 0 §5.2 의 `lstm_pred_*`, `news_top[]` 는 `chronos_pred_*`, `news_reason(string)` 으로 변경됨. Phase 5 prompt 작성 시 반영
+11. **6개 env 필드가 `.env` 에 미기재** — 기본값으로 동작 가능하나 discoverability 위해 `.env.example` 또는 commented block 추가 권장
+
+---
+
+## 다음 단계 (Phase 5 진입 시 brainstorming 주제)
+
+- `state.signals` consumer 패턴: pop vs leave + Phase 5 자체 dedup
+- agent-office 의 `/signal` endpoint 설계 — POST 페이로드 schema
+- Ollama Qwen3 14B Q4 로컬 호출 — 타임아웃, retry, VRAM 공존 (Chronos + Qwen3 동시 메모리 9.3GB / 15.5GB 가용)
+- 이중 텔레그램 (본인 풀 / 아내 lite) — context augmentation 단일 호출에서 양쪽 메시지 생성
+- LLM 비용: ₩0 목표 유지 (로컬)
+
+---
+
+## 양쪽 디렉토리 (web-ui ↔ web-ai) 작업 시 주의
+
+- **코드**: signal_v2 는 web-ai/, spec/plan/메모리는 web-ui/
+- **커밋**: `web-ai` 와 `web-ui` 는 **별도 Gitea 저장소**. 각각 경로에서만 `git add/commit/push`
+- **메모리**: Claude Code 의 auto-memory 는 디렉토리별 격리. 핵심 reference 는 양쪽에 미러됨 (`./memory-mirror/` 또는 `~/.claude/projects/C--Users-jaeoh-Desktop-workspace-web-ai/memory/`)
+- **spec amendment 발생 시**: 코드는 `web-ai` 에 commit, spec 갱신은 `web-ui/docs/superpowers/specs/` 에 commit (Phase 4 spread formula 변경 사례 = web-ui commit `534ded5`)
+
+자세한 V1 가이드는 `signal_v1/CLAUDE.md` 참조 (있다면).

requirements.txt

@@ -7,3 +7,7 @@ pytest>=8.0
 pytest-asyncio>=0.23
 respx>=0.21
 websockets>=12
+# Phase 3b dependencies (Chronos-2 + ML)
+transformers>=4.40
+chronos-forecasting>=1.4
+# torch: typically already installed via V1 venv; if not, install with CUDA support manually

signal_v2/chronos_predictor.py

@@ -0,0 +1,132 @@
+"""Chronos-2 zero-shot forecaster wrapper."""
+from __future__ import annotations
+import logging
+from dataclasses import dataclass
+from datetime import datetime
+from zoneinfo import ZoneInfo
+
+import numpy as np
+
+logger = logging.getLogger(__name__)
+KST = ZoneInfo("Asia/Seoul")
+
+
+@dataclass
+class ChronosPrediction:
+    median: float
+    q10: float
+    q90: float
+    conf: float
+    as_of: str
+
+
+class ChronosPredictor:
+    """HuggingFace Chronos-2 zero-shot forecaster."""
+
+    def __init__(self, model_name: str = "amazon/chronos-2", device: str | None = None):
+        # BaseChronosPipeline auto-detects model variant (Chronos / ChronosBolt / Chronos-2)
+        # and returns the appropriate sub-pipeline. ChronosPipeline only supports legacy T5.
+        import torch
+        try:
+            from chronos import BaseChronosPipeline
+            pipeline_cls = BaseChronosPipeline
+        except ImportError:
+            from chronos import ChronosPipeline
+            pipeline_cls = ChronosPipeline
+
+        self._device = device or ("cuda" if torch.cuda.is_available() else "cpu")
+        # Always use float32 — Korean stock prices (e.g. 280,000원) exceed FP16 max (~65,504)
+        # causing inf in quantile output. FP32 is safe for typical price magnitudes.
+        dtype = torch.float32
+        logger.info("Loading Chronos pipeline: %s on %s (cls=%s)",
+                    model_name, self._device, pipeline_cls.__name__)
+        # Try `dtype` (newer API) first, fall back to `torch_dtype` (older)
+        try:
+            self._pipeline = pipeline_cls.from_pretrained(
+                model_name, device_map=self._device, dtype=dtype,
+            )
+        except TypeError:
+            self._pipeline = pipeline_cls.from_pretrained(
+                model_name, device_map=self._device, torch_dtype=dtype,
+            )
+        logger.info("Chronos pipeline loaded.")
+
+    def predict_batch(
+        self,
+        daily_ohlcv_dict: dict[str, list[dict]],
+        prediction_length: int = 1,
+        num_samples: int = 100,
+    ) -> dict[str, ChronosPrediction]:
+        """종목별 1-day return 분포 예측.
+
+        ChronosBolt / Chronos-2 등 신모델은 predict_quantiles 사용 (deterministic).
+        Legacy ChronosPipeline (T5) 는 sample-based predict.
+        """
+        import torch
+
+        tickers = list(daily_ohlcv_dict.keys())
+        if not tickers:
+            return {}
+
+        contexts = [
+            torch.tensor([bar["close"] for bar in daily_ohlcv_dict[t]], dtype=torch.float32)
+            for t in tickers
+        ]
+        now_iso = datetime.now(KST).isoformat()
+        results: dict[str, ChronosPrediction] = {}
+
+        # Modern API: predict_quantiles (ChronosBolt / Chronos-2)
+        if hasattr(self._pipeline, "predict_quantiles"):
+            quantile_levels = [0.1, 0.5, 0.9]
+            # ChronosBolt API: positional `inputs` (first arg). Older variants use `context`.
+            try:
+                quantiles_tensor, _ = self._pipeline.predict_quantiles(
+                    contexts,
+                    prediction_length=prediction_length,
+                    quantile_levels=quantile_levels,
+                )
+            except TypeError:
+                quantiles_tensor, _ = self._pipeline.predict_quantiles(
+                    context=contexts,
+                    prediction_length=prediction_length,
+                    quantile_levels=quantile_levels,
+                )
+            quantiles_np = (
+                quantiles_tensor.cpu().numpy()
+                if hasattr(quantiles_tensor, "cpu")
+                else np.asarray(quantiles_tensor)
+            )
+            # shape: [num_series, prediction_length, 3]
+            for i, ticker in enumerate(tickers):
+                q10_price, q50_price, q90_price = quantiles_np[i, 0, :]
+                last_close = daily_ohlcv_dict[ticker][-1]["close"]
+                median = float((q50_price - last_close) / last_close)
+                q10 = float((q10_price - last_close) / last_close)
+                q90 = float((q90_price - last_close) / last_close)
+                spread = (q90 - q10) / max(abs(median), 0.001)
+                conf = float(max(0.0, min(1.0, 1.0 - spread / 2.0)))
+                results[ticker] = ChronosPrediction(
+                    median=median, q10=q10, q90=q90, conf=conf, as_of=now_iso,
+                )
+            return results
+
+        # Legacy API: sample-based predict (ChronosPipeline T5)
+        forecasts = self._pipeline.predict(
+            context=contexts,
+            prediction_length=prediction_length,
+            num_samples=num_samples,
+        )
+        forecasts_np = forecasts.numpy() if hasattr(forecasts, "numpy") else np.asarray(forecasts)
+        for i, ticker in enumerate(tickers):
+            samples = forecasts_np[i, :, 0]
+            last_close = daily_ohlcv_dict[ticker][-1]["close"]
+            returns = (samples - last_close) / last_close
+            median = float(np.quantile(returns, 0.5))
+            q10 = float(np.quantile(returns, 0.1))
+            q90 = float(np.quantile(returns, 0.9))
+            spread = (q90 - q10) / max(abs(median), 0.001)
+            conf = float(max(0.0, min(1.0, 1.0 - spread / 2.0)))
+            results[ticker] = ChronosPrediction(
+                median=median, q10=q10, q90=q90, conf=conf, as_of=now_iso,
+            )
+        return results

signal_v2/config.py

@@ -34,6 +34,25 @@ class Settings:
                       str(Path(__file__).parent.parent / "signal_v1" / "data" / "kis_token.json"))
         )
     )
+    chronos_model: str = field(default_factory=lambda: os.getenv("CHRONOS_MODEL", "amazon/chronos-2"))
+    stop_loss_pct: float = field(
+        default_factory=lambda: float(os.getenv("STOP_LOSS_PCT", "-0.07"))
+    )
+    take_profit_pct: float = field(
+        default_factory=lambda: float(os.getenv("TAKE_PROFIT_PCT", "0.15"))
+    )
+    chronos_spread_threshold: float = field(
+        default_factory=lambda: float(os.getenv("CHRONOS_SPREAD_THRESHOLD", "0.6"))
+    )
+    asking_bid_ratio_threshold: float = field(
+        default_factory=lambda: float(os.getenv("ASKING_BID_RATIO_THRESHOLD", "0.6"))
+    )
+    confidence_threshold: float = field(
+        default_factory=lambda: float(os.getenv("CONFIDENCE_THRESHOLD", "0.7"))
+    )
+    min_momentum_for_buy: str = field(
+        default_factory=lambda: os.getenv("MIN_MOMENTUM_FOR_BUY", "strong_up")
+    )
 
     @property
     def kis_is_virtual(self) -> bool:

signal_v2/kis_client.py

@@ -4,7 +4,7 @@ import asyncio
 import json
 import logging
 import time
-from datetime import datetime
+from datetime import datetime, timedelta
 from pathlib import Path
 from zoneinfo import ZoneInfo
 
@@ -153,3 +153,41 @@ class KISClient:
             "current_price": current_price,
             "as_of": datetime.now(KST).isoformat(),
         }
+
+    async def get_daily_ohlcv(self, ticker: str, days: int = 60) -> list[dict]:
+        """KRX 일봉 OHLCV (TR_ID FHKST03010100).
+
+        Returns: [{"datetime", "open", "high", "low", "close", "volume"}, ...]
+                 시간 오름차순.
+        """
+        path = "/uapi/domestic-stock/v1/quotations/inquire-daily-itemchartprice"
+        today = datetime.now(KST).strftime("%Y%m%d")
+        start_date = (datetime.now(KST) - timedelta(days=days * 2)).strftime("%Y%m%d")
+        params = {
+            "FID_COND_MRKT_DIV_CODE": "J",
+            "FID_INPUT_ISCD": ticker,
+            "FID_INPUT_DATE_1": start_date,
+            "FID_INPUT_DATE_2": today,
+            "FID_PERIOD_DIV_CODE": "D",
+            "FID_ORG_ADJ_PRC": "1",
+        }
+        raw = await self._request_with_retry(
+            "GET", path, tr_id="FHKST03010100", params=params,
+        )
+        output2 = raw.get("output2", [])
+        bars = []
+        for row in output2:
+            try:
+                date = row["stck_bsop_date"]
+                bars.append({
+                    "datetime": f"{date[:4]}-{date[4:6]}-{date[6:]}",
+                    "open": int(row["stck_oprc"]),
+                    "high": int(row["stck_hgpr"]),
+                    "low": int(row["stck_lwpr"]),
+                    "close": int(row["stck_clpr"]),
+                    "volume": int(row["acml_vol"]),
+                })
+            except (KeyError, ValueError):
+                continue
+        bars.reverse()
+        return bars[-days:]

signal_v2/main.py

@@ -7,6 +7,7 @@ from contextlib import asynccontextmanager
 from fastapi import FastAPI
 
 from signal_v2 import state as state_mod
+from signal_v2.chronos_predictor import ChronosPredictor
 from signal_v2.config import get_settings
 from signal_v2.kis_client import KISClient
 from signal_v2.kis_websocket import KISWebSocket
@@ -24,6 +25,7 @@ class AppContext:
     poll_task: asyncio.Task | None = None
     kis_client: KISClient | None = None
     kis_ws: KISWebSocket | None = None
+    chronos: ChronosPredictor | None = None
 
 
 _ctx = AppContext()
@@ -69,10 +71,19 @@ async def lifespan(app: FastAPI):
         except Exception:
             logger.exception("KIS WebSocket startup failed — continuing without realtime asking_price")
 
+        # Load Chronos (heavy: ~1GB model download first time)
+        try:
+            _ctx.chronos = ChronosPredictor(model_name=settings.chronos_model)
+        except Exception:
+            logger.exception("ChronosPredictor load failed — continuing without chronos predictions")
+
     _ctx.poll_task = asyncio.create_task(
         poll_loop(
             _ctx.client, state_mod.state, _ctx.shutdown,
             kis_client=_ctx.kis_client,
+            chronos=_ctx.chronos,
+            dedup=_ctx.dedup,
+            settings=settings,
         )
     )
 

signal_v2/momentum_classifier.py

@@ -0,0 +1,69 @@
+"""분봉 OHLCV → 5-level 모멘텀 분류."""
+from __future__ import annotations
+from collections import deque
+
+# 분류 카테고리
+STRONG_UP = "strong_up"
+WEAK_UP = "weak_up"
+NEUTRAL = "neutral"
+WEAK_DOWN = "weak_down"
+STRONG_DOWN = "strong_down"
+
+_BARS_PER_5MIN = 5
+_LOOKBACK_5MIN_BARS = 5
+_VOLUME_AVG_WINDOW = 12  # 60분 = 5분봉 12개
+
+
+def aggregate_1min_to_5min(minute_bars: list[dict]) -> list[dict]:
+    """1분봉 N개 → 5분봉 floor(N/5) 개. 시간 오름차순.
+
+    각 5분봉: open=첫 1분봉 open, high=max, low=min, close=마지막 close, volume=sum.
+    """
+    bars_5min = []
+    chunks = len(minute_bars) // _BARS_PER_5MIN
+    for i in range(chunks):
+        chunk = minute_bars[i * _BARS_PER_5MIN : (i + 1) * _BARS_PER_5MIN]
+        bars_5min.append({
+            "datetime": chunk[0]["datetime"],
+            "open": chunk[0]["open"],
+            "high": max(b["high"] for b in chunk),
+            "low": min(b["low"] for b in chunk),
+            "close": chunk[-1]["close"],
+            "volume": sum(b["volume"] for b in chunk),
+        })
+    return bars_5min
+
+
+def classify_minute_momentum(minute_bars: deque) -> str:
+    """1분봉 deque → 5-level 모멘텀 분류.
+
+    Returns: STRONG_UP / WEAK_UP / NEUTRAL / WEAK_DOWN / STRONG_DOWN
+    """
+    minute_list = list(minute_bars)
+    if len(minute_list) < _BARS_PER_5MIN * _LOOKBACK_5MIN_BARS:
+        return NEUTRAL  # 데이터 부족
+
+    bars_5min = aggregate_1min_to_5min(minute_list)
+    if len(bars_5min) < _LOOKBACK_5MIN_BARS:
+        return NEUTRAL
+
+    recent = bars_5min[-_LOOKBACK_5MIN_BARS:]
+    up_count = sum(1 for b in recent if b["close"] > b["open"])
+
+    # 거래량 multiplier: recent 5 avg vs 60분 avg
+    recent_vol_avg = sum(b["volume"] for b in recent) / len(recent)
+    long_window = bars_5min[-_VOLUME_AVG_WINDOW:]
+    long_vol_avg = sum(b["volume"] for b in long_window) / len(long_window)
+    vol_mult = recent_vol_avg / long_vol_avg if long_vol_avg > 0 else 1.0
+
+    # 5-level 분류
+    if up_count == 5 and vol_mult >= 1.5:
+        return STRONG_UP
+    elif up_count >= 3 and vol_mult >= 1.0:
+        return WEAK_UP
+    elif up_count == 0 and vol_mult >= 1.5:
+        return STRONG_DOWN
+    elif up_count <= 2 and vol_mult < 1.0:
+        return WEAK_DOWN
+    else:
+        return NEUTRAL

signal_v2/pull_worker.py

@@ -7,7 +7,7 @@ from datetime import datetime
 
 from signal_v2.kis_client import KISClient
 from signal_v2.scheduler import (
-    KST, _is_market_day, _is_polling_window, _next_interval,
+    KST, _is_market_day, _is_polling_window, _next_interval, _is_post_close_trigger,
 )
 from signal_v2.state import PollState
 from signal_v2.stock_client import StockClient
@@ -18,6 +18,9 @@ logger = logging.getLogger(__name__)
 async def poll_loop(
     client: StockClient, state: PollState, shutdown: asyncio.Event,
     kis_client: KISClient | None = None,
+    chronos=None,
+    dedup=None,
+    settings=None,
 ) -> None:
     """FastAPI lifespan 에서 asyncio.create_task 로 시작."""
     logger.info("poll_loop started")
@@ -28,6 +31,24 @@ async def poll_loop(
                 await _run_polling_cycle(client, state, kis_client=kis_client)
             except Exception:
                 logger.exception("poll cycle failed")
+            # Minute momentum 갱신 (매 cycle)
+            try:
+                update_minute_momentum_for_all(state)
+            except Exception:
+                logger.exception("minute momentum update failed")
+            # Post-close trigger (16:00 KST)
+            if _is_post_close_trigger(now) and chronos is not None and kis_client is not None:
+                try:
+                    await _run_post_close_cycle(kis_client, chronos, state)
+                except Exception:
+                    logger.exception("post-close cycle failed")
+            # Phase 4: generate signals
+            if dedup is not None and settings is not None:
+                try:
+                    from signal_v2.signal_generator import generate_signals
+                    generate_signals(state, dedup, settings)
+                except Exception:
+                    logger.exception("generate_signals failed")
         interval = _next_interval(now)
         try:
             await asyncio.wait_for(shutdown.wait(), timeout=interval)
@@ -125,3 +146,48 @@ def _screener_tickers(state: PollState) -> list[str]:
     if state.screener_preview is None:
         return []
     return [i["ticker"] for i in state.screener_preview.get("items", []) if "ticker" in i]
+
+
+async def _run_post_close_cycle(kis_client, chronos, state) -> None:
+    """16:00 KST 종가 후 1회: daily fetch + chronos predict."""
+    tickers = list(set(_portfolio_tickers(state)) | set(_screener_tickers(state)))
+    if not tickers:
+        return
+
+    daily_results = await asyncio.gather(*[
+        kis_client.get_daily_ohlcv(t, days=60) for t in tickers
+    ], return_exceptions=True)
+    daily_dict = {}
+    for ticker, result in zip(tickers, daily_results):
+        if isinstance(result, list) and len(result) >= 30:
+            daily_dict[ticker] = result
+            state.daily_ohlcv[ticker] = result
+        elif isinstance(result, Exception):
+            state.fetch_errors[f"daily_ohlcv/{ticker}"] = (
+                state.fetch_errors.get(f"daily_ohlcv/{ticker}", 0) + 1
+            )
+
+    if daily_dict and chronos is not None:
+        try:
+            predictions = chronos.predict_batch(daily_dict)
+        except Exception:
+            logger.exception("chronos predict_batch failed")
+            return
+        for ticker, pred in predictions.items():
+            state.chronos_predictions[ticker] = {
+                "median": pred.median,
+                "q10": pred.q10,
+                "q90": pred.q90,
+                "conf": pred.conf,
+                "as_of": pred.as_of,
+            }
+            state.last_updated[f"chronos/{ticker}"] = pred.as_of
+
+
+def update_minute_momentum_for_all(state) -> None:
+    """매 분봉 cycle 후 호출 — 모든 종목 모멘텀 갱신."""
+    from signal_v2.momentum_classifier import classify_minute_momentum
+    now_iso = datetime.now(KST).isoformat()
+    for ticker, bars in state.minute_bars.items():
+        state.minute_momentum[ticker] = classify_minute_momentum(bars)
+        state.last_updated[f"momentum/{ticker}"] = now_iso

signal_v2/scheduler.py

@@ -76,6 +76,14 @@ def _seconds_until_nxt_or_market_open(now: datetime) -> float:
     return 86400.0
 
 
+def _is_post_close_trigger(now: datetime) -> bool:
+    """16:00 KST ±1분 (post-close cycle 트리거). 평일/영업일만."""
+    if not _is_market_day(now):
+        return False
+    t = now.time()
+    return time(16, 0) <= t < time(16, 1)
+
+
 def _seconds_until_next_market_open(now: datetime) -> float:
     """다음 영업일의 07:00 KST 까지 초수 (휴장일/주말용)."""
     candidate = now.replace(hour=7, minute=0, second=0, microsecond=0)

signal_v2/signal_generator.py

@@ -0,0 +1,228 @@
+"""Phase 4 — 매수/매도 신호 생성.
+
+순수 함수 generate_signals(state, dedup, settings). state 를 mutate.
+"""
+from __future__ import annotations
+import logging
+from datetime import datetime
+from zoneinfo import ZoneInfo
+
+logger = logging.getLogger(__name__)
+KST = ZoneInfo("Asia/Seoul")
+
+MOMENTUM_SCORES = {
+    "strong_up": 1.0,
+    "weak_up": 0.7,
+    "neutral": 0.5,
+    "weak_down": 0.3,
+    "strong_down": 0.0,
+}
+
+
+def generate_signals(state, dedup, settings) -> None:
+    """Phase 4 entry — state-mutating. Evaluation order: sell first (priority), then buy. A ticker receiving a sell signal in this cycle is excluded from buy evaluation to avoid silent overwrite."""
+    _evaluate_sell_signals(state, dedup, settings)
+    _evaluate_buy_signals(state, dedup, settings)
+
+
+# ----- 매수 -----
+
+def _evaluate_buy_signals(state, dedup, settings) -> None:
+    candidates = _buy_candidates(state)
+    for ticker, name, rank in candidates:
+        existing = state.signals.get(ticker)
+        if existing is not None and existing.get("action") == "sell":
+            logger.debug("buy %s skipped: same-cycle sell precedence", ticker)
+            continue
+        if not _check_buy_hard_gate(state, ticker, settings):
+            logger.debug("buy %s skipped: hard gate failed", ticker)
+            continue
+        confidence = _compute_buy_confidence(state, ticker, rank)
+        if confidence <= settings.confidence_threshold:
+            logger.debug("buy %s skipped: confidence %.3f <= %.3f",
+                         ticker, confidence, settings.confidence_threshold)
+            continue
+        if dedup.is_recent(ticker, "buy", within_hours=24):
+            logger.debug("buy %s skipped: dedup 24h", ticker)
+            continue
+        state.signals[ticker] = _build_buy_signal(state, ticker, name, rank, confidence)
+        dedup.record(ticker, "buy", confidence=confidence)
+        logger.info("signal emit %s buy conf=%.3f rank=%s", ticker, confidence, rank)
+
+
+def _buy_candidates(state) -> list[tuple[str, str, int | None]]:
+    """screener Top-N (rank 1..N) + portfolio (rank=None)."""
+    candidates: list[tuple[str, str, int | None]] = []
+    seen: set[str] = set()
+    if state.screener_preview is not None:
+        for i, item in enumerate(state.screener_preview.get("items", [])):
+            ticker = item.get("ticker")
+            if not ticker or ticker in seen:
+                continue
+            seen.add(ticker)
+            name = item.get("name", ticker)
+            candidates.append((ticker, name, i + 1))
+    if state.portfolio is not None:
+        for h in state.portfolio.get("holdings", []):
+            ticker = h.get("ticker")
+            if not ticker or ticker in seen:
+                continue
+            seen.add(ticker)
+            candidates.append((ticker, h.get("name", ticker), None))
+    return candidates
+
+
+def _check_buy_hard_gate(state, ticker: str, settings) -> bool:
+    pred = state.chronos_predictions.get(ticker)
+    if pred is None or pred.get("median", 0) <= 0:
+        return False
+    spread = pred.get("q90", 0) - pred.get("q10", 0)
+    if spread >= settings.chronos_spread_threshold:
+        return False
+    momentum = state.minute_momentum.get(ticker)
+    if momentum != settings.min_momentum_for_buy:
+        return False
+    ap = state.asking_price.get(ticker)
+    if ap is None or ap.get("bid_ratio", 0) < settings.asking_bid_ratio_threshold:
+        return False
+    return True
+
+
+def _compute_buy_confidence(state, ticker: str, rank: int | None) -> float:
+    pred = state.chronos_predictions[ticker]
+    chronos_conf = pred["conf"]
+    minute_score = MOMENTUM_SCORES.get(state.minute_momentum.get(ticker, "neutral"), 0.5)
+    screener_norm = max(0.0, 1 - (rank - 1) / 20) if rank is not None else 0.0
+    return chronos_conf * 0.5 + minute_score * 0.3 + screener_norm * 0.2
+
+
+def _build_buy_signal(state, ticker: str, name: str, rank: int | None, confidence: float) -> dict:
+    ap = state.asking_price[ticker]
+    return {
+        "ticker": ticker,
+        "name": name,
+        "action": "buy",
+        "confidence_webai": confidence,
+        "current_price": ap["current_price"],
+        "avg_price": None,
+        "pnl_pct": None,
+        "context": _build_context(state, ticker, rank),
+        "as_of": datetime.now(KST).isoformat(),
+    }
+
+
+# ----- 매도 -----
+
+def _evaluate_sell_signals(state, dedup, settings) -> None:
+    if state.portfolio is None:
+        return
+    for holding in state.portfolio.get("holdings", []):
+        ticker = holding.get("ticker")
+        if not ticker:
+            continue
+        sell = _try_stop_loss(state, holding, settings)
+        if sell is None:
+            sell = _try_anomaly(state, holding, settings)
+        if sell is None:
+            sell = _try_take_profit(state, holding, settings)
+        if sell is None:
+            continue
+        if dedup.is_recent(ticker, "sell", within_hours=24):
+            logger.debug("sell %s skipped: dedup 24h", ticker)
+            continue
+        state.signals[ticker] = sell
+        dedup.record(ticker, "sell", confidence=sell["confidence_webai"])
+        logger.info("signal emit %s sell conf=%.3f reason=%s",
+                    ticker, sell["confidence_webai"],
+                    sell.get("context", {}).get("sell_reason"))
+
+
+def _try_stop_loss(state, holding: dict, settings) -> dict | None:
+    pnl = holding.get("pnl_pct")
+    if pnl is None or pnl >= settings.stop_loss_pct:
+        return None
+    return _build_sell_signal(state, holding, confidence=1.0, reason="stop_loss")
+
+
+def _try_take_profit(state, holding: dict, settings) -> dict | None:
+    pnl = holding.get("pnl_pct")
+    if pnl is None or pnl <= settings.take_profit_pct:
+        return None
+    return _build_sell_signal(state, holding, confidence=0.6, reason="take_profit")
+
+
+def _try_anomaly(state, holding: dict, settings) -> dict | None:
+    ticker = holding["ticker"]
+    pred = state.chronos_predictions.get(ticker)
+    if pred is None or pred["median"] >= -0.01:
+        return None
+    momentum = state.minute_momentum.get(ticker)
+    if momentum != "strong_down":
+        return None
+    ap = state.asking_price.get(ticker)
+    if ap is None:
+        return None
+    if ap["bid_ratio"] > (1 - settings.asking_bid_ratio_threshold):
+        return None
+    minute_score = 1.0 - MOMENTUM_SCORES.get(momentum, 0.5)
+    confidence = pred["conf"] * 0.5 + minute_score * 0.3 + 1.0 * 0.2
+    if confidence <= settings.confidence_threshold:
+        return None
+    return _build_sell_signal(state, holding, confidence=confidence, reason="anomaly")
+
+
+def _build_sell_signal(state, holding: dict, confidence: float, reason: str) -> dict:
+    ticker = holding["ticker"]
+    return {
+        "ticker": ticker,
+        "name": holding.get("name", ticker),
+        "action": "sell",
+        "confidence_webai": confidence,
+        "current_price": holding.get("current_price"),
+        "avg_price": holding.get("avg_price"),
+        "pnl_pct": holding.get("pnl_pct"),
+        "context": _build_context(state, ticker, rank=None, sell_reason=reason),
+        "as_of": datetime.now(KST).isoformat(),
+    }
+
+
+# ----- Context -----
+
+def _build_context(state, ticker: str, rank: int | None, sell_reason: str | None = None) -> dict:
+    pred = state.chronos_predictions.get(ticker) or {}
+    ap = state.asking_price.get(ticker) or {}
+    news_item = _find_news_sentiment(state, ticker)
+    screener_scores = _find_screener_scores(state, ticker)
+    context: dict = {
+        "chronos_pred_1d": pred.get("median"),
+        "chronos_pred_conf": pred.get("conf"),
+        "chronos_q10": pred.get("q10"),
+        "chronos_q90": pred.get("q90"),
+        "screener_rank": rank,
+        "screener_scores": screener_scores,
+        "minute_momentum": state.minute_momentum.get(ticker),
+        "asking_bid_ratio": ap.get("bid_ratio"),
+        "news_sentiment": news_item.get("score") if news_item else None,
+        "news_reason": news_item.get("reason") if news_item else None,
+    }
+    if sell_reason is not None:
+        context["sell_reason"] = sell_reason
+    return context
+
+
+def _find_news_sentiment(state, ticker: str) -> dict | None:
+    if state.news_sentiment is None:
+        return None
+    for item in state.news_sentiment.get("items", []):
+        if item.get("ticker") == ticker:
+            return item
+    return None
+
+
+def _find_screener_scores(state, ticker: str) -> dict | None:
+    if state.screener_preview is None:
+        return None
+    for item in state.screener_preview.get("items", []):
+        if item.get("ticker") == ticker:
+            return item.get("scores")
+    return None

signal_v2/state.py

@@ -8,9 +8,13 @@ class PollState:
     portfolio: dict | None = None
     news_sentiment: dict | None = None
     screener_preview: dict | None = None
-    # Phase 3a additions
     minute_bars: dict[str, deque] = field(default_factory=dict)
     asking_price: dict[str, dict] = field(default_factory=dict)
+    # Phase 3b additions
+    daily_ohlcv: dict[str, list[dict]] = field(default_factory=dict)
+    chronos_predictions: dict[str, dict] = field(default_factory=dict)
+    minute_momentum: dict[str, str] = field(default_factory=dict)
+    signals: dict[str, dict] = field(default_factory=dict)
     last_updated: dict[str, str] = field(default_factory=dict)
     fetch_errors: dict[str, int] = field(default_factory=dict)
 

signal_v2/tests/test_chronos_predictor.py

@@ -0,0 +1,92 @@
+"""Tests for ChronosPredictor (model mock)."""
+from unittest.mock import MagicMock, patch
+
+import numpy as np
+import pytest
+
+
+@pytest.fixture
+def mock_pipeline():
+    """Mock BaseChronosPipeline.from_pretrained returning a mock pipeline object."""
+    with patch("chronos.BaseChronosPipeline") as cls:
+        cls.__name__ = "BaseChronosPipeline"
+        instance = MagicMock()
+        # ChronosBolt API: predict_quantiles returns (quantiles_tensor, mean_tensor)
+        # Modern (predict_quantiles) branch will be used since hasattr(MagicMock, "predict_quantiles") is True.
+        cls.from_pretrained.return_value = instance
+        yield instance
+
+
+@pytest.fixture
+def mock_torch_cpu():
+    with patch("torch.cuda.is_available", return_value=False):
+        yield
+
+
+def _daily_ohlcv(close_seq):
+    return [{"datetime": f"2026-05-{i+1:02d}", "open": c, "high": c, "low": c,
+             "close": c, "volume": 1000} for i, c in enumerate(close_seq)]
+
+
+def _mk_quantiles_tensor(q10_price: float, q50_price: float, q90_price: float):
+    """Helper: build predict_quantiles return tensor shape [1, 1, 3]."""
+    import torch
+    return torch.tensor([[[q10_price, q50_price, q90_price]]], dtype=torch.float32)
+
+
+def test_predict_batch_returns_prediction_dict(mock_pipeline, mock_torch_cpu):
+    """mock predict_quantiles → dict[ticker, ChronosPrediction]. last_close=100, q50=102 → median≈+2%."""
+    quantiles = _mk_quantiles_tensor(101.5, 102.0, 102.5)  # narrow around 102
+    mock_pipeline.predict_quantiles.return_value = (quantiles, None)
+
+    from signal_v2.chronos_predictor import ChronosPredictor, ChronosPrediction
+    predictor = ChronosPredictor(model_name="mock-model")
+    daily = {"005930": _daily_ohlcv([100] * 60)}
+    result = predictor.predict_batch(daily)
+    assert "005930" in result
+    pred = result["005930"]
+    assert isinstance(pred, ChronosPrediction)
+    assert abs(pred.median - 0.02) < 0.001
+
+
+def test_conf_high_when_distribution_narrow(mock_pipeline, mock_torch_cpu):
+    """좁은 distribution (q90-q10 작음, median 0 아님) → conf ≈ 1."""
+    # last_close=100, q10=101.99, q50=102.00, q90=102.01
+    # returns: q10=0.0199, q50=0.02, q90=0.0201
+    # spread = (0.0201 - 0.0199) / max(0.02, 0.001) = 0.0002/0.02 = 0.01 → conf = 1 - 0.005 = 0.995
+    quantiles = _mk_quantiles_tensor(101.99, 102.0, 102.01)
+    mock_pipeline.predict_quantiles.return_value = (quantiles, None)
+
+    from signal_v2.chronos_predictor import ChronosPredictor
+    predictor = ChronosPredictor(model_name="mock-model")
+    daily = {"005930": _daily_ohlcv([100] * 60)}
+    result = predictor.predict_batch(daily)
+    assert result["005930"].conf > 0.8
+
+
+def test_conf_low_when_distribution_wide(mock_pipeline, mock_torch_cpu):
+    """넓은 distribution → conf ≈ 0."""
+    # last_close=100, q10=70, q50=100, q90=130
+    # returns: q10=-0.3, q50=0.0, q90=0.3
+    # spread = (0.3 - (-0.3)) / max(0.0, 0.001) = 0.6 / 0.001 = 600 → conf = max(0, 1 - 300) = 0
+    quantiles = _mk_quantiles_tensor(70.0, 100.0, 130.0)
+    mock_pipeline.predict_quantiles.return_value = (quantiles, None)
+
+    from signal_v2.chronos_predictor import ChronosPredictor
+    predictor = ChronosPredictor(model_name="mock-model")
+    daily = {"005930": _daily_ohlcv([100] * 60)}
+    result = predictor.predict_batch(daily)
+    assert result["005930"].conf < 0.3
+
+
+def test_return_computed_from_price_relative_to_last_close(mock_pipeline, mock_torch_cpu):
+    """price 예측 → last_close 대비 return 변환. last_close=100, q50=110 → return ≈ +10%."""
+    quantiles = _mk_quantiles_tensor(109.0, 110.0, 111.0)
+    mock_pipeline.predict_quantiles.return_value = (quantiles, None)
+
+    from signal_v2.chronos_predictor import ChronosPredictor
+    predictor = ChronosPredictor(model_name="mock-model")
+    # last close = 100
+    daily = {"005930": _daily_ohlcv(list(range(41, 101)))}  # last = 100
+    result = predictor.predict_batch(daily)
+    assert abs(result["005930"].median - 0.10) < 0.001

signal_v2/tests/test_kis_client.py

@@ -126,3 +126,36 @@ async def test_get_asking_price_computes_bid_ratio(kis_client_factory):
         assert "as_of" in data
     finally:
         await client.close()
+
+
+@respx.mock
+async def test_get_daily_ohlcv_returns_60_bars(kis_client_factory):
+    """KIS daily endpoint returns 60 ascending bars after parsing."""
+    # Build 60 KIS-format daily bars (descending dates as KIS does)
+    sample_output2 = []
+    for i in range(60):
+        # Generate a fake date 60 days ago, descending
+        day = 60 - i
+        sample_output2.append({
+            "stck_bsop_date": f"2026{(((day-1)//30)+1):02d}{(((day-1)%30)+1):02d}",
+            "stck_oprc": "78000", "stck_hgpr": "78500",
+            "stck_lwpr": "77800", "stck_clpr": str(78000 + i),
+            "acml_vol": "12345",
+        })
+
+    respx.get(
+        "https://openapivts.koreainvestment.com:29443/uapi/domestic-stock/v1/quotations/inquire-daily-itemchartprice"
+    ).mock(return_value=httpx.Response(200, json={"output2": sample_output2}))
+
+    client = kis_client_factory()
+    try:
+        bars = await client.get_daily_ohlcv("005930", days=60)
+        # KIS returns descending; client reverses to ascending
+        assert len(bars) == 60
+        # Ascending order: first item has smaller datetime than last
+        assert bars[0]["datetime"] < bars[-1]["datetime"]
+        assert isinstance(bars[0]["open"], int)
+        assert isinstance(bars[0]["close"], int)
+        assert "datetime" in bars[0]
+    finally:
+        await client.close()

signal_v2/tests/test_momentum_classifier.py

@@ -0,0 +1,92 @@
+"""Tests for minute momentum classifier."""
+from collections import deque
+
+from signal_v2.momentum_classifier import (
+    aggregate_1min_to_5min, classify_minute_momentum,
+    STRONG_UP, WEAK_UP, NEUTRAL, WEAK_DOWN, STRONG_DOWN,
+)
+
+
+def _bar(open_, high, low, close, volume):
+    return {
+        "datetime": "2026-05-18T09:00:00+09:00",
+        "open": open_, "high": high, "low": low, "close": close, "volume": volume,
+    }
+
+
+def _make_chunks(num_chunks_up: int, num_chunks_total: int, base_vol: int = 1000):
+    """num_chunks_total 개의 5-bar 청크. num_chunks_up 청크는 양봉, 나머지는 음봉.
+    각 청크는 5개 1분봉. 거래량 = base_vol per bar.
+    """
+    bars = []
+    for i in range(num_chunks_total):
+        is_up = i < num_chunks_up
+        o, c = (100, 110) if is_up else (110, 100)
+        for j in range(5):
+            bars.append(_bar(o, max(o, c) + 5, min(o, c) - 5, c, base_vol))
+    return bars
+
+
+def test_strong_up_5_consecutive_green_with_high_volume():
+    """직전 5개 5분봉 모두 양봉 + 거래량 1.5x → STRONG_UP."""
+    # 60분 (12 5분봉) 데이터: 7 normal + 5 high-vol up
+    older = _make_chunks(num_chunks_up=3, num_chunks_total=7, base_vol=1000)
+    recent = _make_chunks(num_chunks_up=5, num_chunks_total=5, base_vol=2500)
+    minute_bars = deque(older + recent, maxlen=60)
+    assert classify_minute_momentum(minute_bars) == STRONG_UP
+
+
+def test_weak_up_3of5_green_normal_volume():
+    """직전 5개 5분봉 중 3-4개 양봉 + 거래량 ≥ 1.0x → WEAK_UP."""
+    older = _make_chunks(num_chunks_up=3, num_chunks_total=7, base_vol=1000)
+    # 5 chunks: 3 up + 2 down, normal vol
+    recent_up = _make_chunks(num_chunks_up=3, num_chunks_total=3, base_vol=1000)
+    recent_down = _make_chunks(num_chunks_up=0, num_chunks_total=2, base_vol=1000)
+    minute_bars = deque(older + recent_up + recent_down, maxlen=60)
+    assert classify_minute_momentum(minute_bars) == WEAK_UP
+
+
+def test_neutral_mixed():
+    """up_count=2, vol normal → NEUTRAL (rule 미해당)."""
+    older = _make_chunks(num_chunks_up=3, num_chunks_total=7, base_vol=1000)
+    recent_up = _make_chunks(num_chunks_up=2, num_chunks_total=2, base_vol=1000)
+    recent_down = _make_chunks(num_chunks_up=0, num_chunks_total=3, base_vol=1000)
+    minute_bars = deque(older + recent_up + recent_down, maxlen=60)
+    # up_count=2, vol_mult=1.0 → 어느 분기 조건도 만족 안 함 → NEUTRAL
+    assert classify_minute_momentum(minute_bars) == NEUTRAL
+
+
+def test_weak_down_low_green_low_volume():
+    """up_count <= 2 + vol < 1.0 → WEAK_DOWN."""
+    older = _make_chunks(num_chunks_up=3, num_chunks_total=7, base_vol=1000)
+    recent_up = _make_chunks(num_chunks_up=1, num_chunks_total=1, base_vol=500)
+    recent_down = _make_chunks(num_chunks_up=0, num_chunks_total=4, base_vol=500)
+    minute_bars = deque(older + recent_up + recent_down, maxlen=60)
+    # recent 5 chunks avg vol = 500, long 12 avg ≈ (7*1000 + 5*500) / 12 ≈ 791 → vol_mult ≈ 0.63
+    assert classify_minute_momentum(minute_bars) == WEAK_DOWN
+
+
+def test_strong_down_5_consecutive_red_high_volume():
+    """직전 5개 5분봉 모두 음봉 + 거래량 1.5x → STRONG_DOWN."""
+    older = _make_chunks(num_chunks_up=3, num_chunks_total=7, base_vol=1000)
+    recent = _make_chunks(num_chunks_up=0, num_chunks_total=5, base_vol=2500)
+    minute_bars = deque(older + recent, maxlen=60)
+    assert classify_minute_momentum(minute_bars) == STRONG_DOWN
+
+
+def test_aggregate_1min_to_5min_correctness():
+    """5 1분봉 → 1개 5분봉 — open/close/high/low/volume 정확."""
+    bars = [
+        _bar(100, 105, 99, 102, 1000),
+        _bar(102, 108, 101, 107, 1500),
+        _bar(107, 110, 105, 106, 800),
+        _bar(106, 109, 104, 108, 1200),
+        _bar(108, 112, 107, 111, 900),
+    ]
+    result = aggregate_1min_to_5min(bars)
+    assert len(result) == 1
+    assert result[0]["open"] == 100  # 첫 bar
+    assert result[0]["close"] == 111  # 마지막 bar
+    assert result[0]["high"] == 112  # max
+    assert result[0]["low"] == 99    # min
+    assert result[0]["volume"] == 5400  # sum

signal_v2/tests/test_pull_worker.py

@@ -53,3 +53,79 @@ def test_websocket_message_updates_state_asking_price():
                   "current_price": 78500, "as_of": "2026-05-18T10:00:00+09:00"})
     assert state.asking_price["005930"]["bid_total"] == 1000
     assert "asking_price/005930" in state.last_updated
+
+
+async def test_post_close_cycle_updates_chronos_predictions():
+    """mock kis + mock chronos → state.chronos_predictions + state.daily_ohlcv 갱신."""
+    from unittest.mock import AsyncMock, MagicMock
+    from signal_v2.pull_worker import _run_post_close_cycle
+    from signal_v2.chronos_predictor import ChronosPrediction
+    from signal_v2.state import PollState
+
+    state = PollState()
+    state.portfolio = {"holdings": [{"ticker": "005930"}]}
+    state.screener_preview = {"items": [{"ticker": "000660"}]}
+
+    kis_mock = MagicMock()
+    daily_005930 = [{"datetime": f"2026-05-{i+1:02d}", "open": 100, "high": 105,
+                     "low": 95, "close": 100 + i, "volume": 1000} for i in range(60)]
+    daily_000660 = [{"datetime": f"2026-05-{i+1:02d}", "open": 200, "high": 210,
+                     "low": 190, "close": 200 + i, "volume": 2000} for i in range(60)]
+    # _run_post_close_cycle iterates tickers and calls get_daily_ohlcv per ticker.
+    # Order depends on set() so use side_effect mapping if possible, otherwise list.
+    async def fake_daily(ticker, days=60):
+        if ticker == "005930":
+            return daily_005930
+        if ticker == "000660":
+            return daily_000660
+        return []
+    kis_mock.get_daily_ohlcv = AsyncMock(side_effect=fake_daily)
+
+    chronos_mock = MagicMock()
+    chronos_mock.predict_batch = MagicMock(return_value={
+        "005930": ChronosPrediction(0.02, -0.01, 0.04, 0.85, "2026-05-18T16:00:00+09:00"),
+        "000660": ChronosPrediction(0.03, -0.02, 0.06, 0.75, "2026-05-18T16:00:00+09:00"),
+    })
+
+    await _run_post_close_cycle(kis_mock, chronos_mock, state)
+
+    assert "005930" in state.chronos_predictions
+    assert "000660" in state.chronos_predictions
+    assert state.chronos_predictions["005930"]["median"] == 0.02
+    assert state.chronos_predictions["005930"]["conf"] == 0.85
+    assert "005930" in state.daily_ohlcv
+    assert "chronos/005930" in state.last_updated
+
+
+def test_poll_loop_calls_generate_signals_after_cycle(monkeypatch):
+    """Phase 4: generate_signals 가 cycle 후 state.signals 를 갱신한다."""
+    from unittest.mock import MagicMock
+    from signal_v2.state import PollState
+    from signal_v2.signal_generator import generate_signals
+
+    state = PollState()
+    state.portfolio = {"holdings": [{
+        "ticker": "005930", "name": "삼성전자",
+        "avg_price": 75000, "current_price": 69000,
+        "pnl_pct": -0.08, "profit_rate": -8.0,
+        "quantity": 100, "broker": "키움",
+    }]}
+    state.screener_preview = {"items": []}
+
+    dedup = MagicMock()
+    dedup.is_recent.return_value = False
+
+    settings = MagicMock()
+    settings.stop_loss_pct = -0.07
+    settings.take_profit_pct = 0.15
+    settings.chronos_spread_threshold = 0.6
+    settings.asking_bid_ratio_threshold = 0.6
+    settings.confidence_threshold = 0.7
+    settings.min_momentum_for_buy = "strong_up"
+
+    generate_signals(state, dedup, settings)
+
+    assert "005930" in state.signals
+    assert state.signals["005930"]["action"] == "sell"
+    assert state.signals["005930"]["confidence_webai"] == 1.0
+    dedup.record.assert_called_with("005930", "sell", confidence=1.0)

signal_v2/tests/test_signal_generator.py

@@ -0,0 +1,172 @@
+"""Tests for signal_generator."""
+from unittest.mock import MagicMock
+
+import pytest
+
+from signal_v2.signal_generator import generate_signals
+from signal_v2.state import PollState
+
+
+def _settings(**overrides):
+    """Build a Settings-like object for tests (avoid env)."""
+    defaults = dict(
+        stop_loss_pct=-0.07,
+        take_profit_pct=0.15,
+        chronos_spread_threshold=0.6,
+        asking_bid_ratio_threshold=0.6,
+        confidence_threshold=0.7,
+        min_momentum_for_buy="strong_up",
+    )
+    defaults.update(overrides)
+    m = MagicMock()
+    for k, v in defaults.items():
+        setattr(m, k, v)
+    return m
+
+
+def _make_state_with_buy_candidate(
+    ticker="005930", name="삼성전자",
+    chronos_median=0.02, chronos_q10=-0.01, chronos_q90=0.04, chronos_conf=0.85,
+    momentum="strong_up", bid_ratio=0.7, current_price=78500,
+):
+    state = PollState()
+    state.screener_preview = {"items": [{"ticker": ticker, "name": name}]}
+    state.chronos_predictions[ticker] = {
+        "median": chronos_median, "q10": chronos_q10, "q90": chronos_q90,
+        "conf": chronos_conf, "as_of": "2026-05-17T16:00:00+09:00",
+    }
+    state.minute_momentum[ticker] = momentum
+    state.asking_price[ticker] = {
+        "bid_total": int(bid_ratio * 1000),
+        "ask_total": int((1 - bid_ratio) * 1000),
+        "bid_ratio": bid_ratio,
+        "current_price": current_price,
+        "as_of": "2026-05-17T16:00:01+09:00",
+    }
+    return state
+
+
+def _make_state_with_holding(
+    ticker="005930", name="삼성전자",
+    pnl_pct=0.0, avg_price=75000, current_price=75000,
+):
+    state = PollState()
+    state.portfolio = {"holdings": [{
+        "ticker": ticker, "name": name,
+        "avg_price": avg_price, "current_price": current_price,
+        "pnl_pct": pnl_pct, "profit_rate": pnl_pct * 100,
+        "quantity": 100, "broker": "키움",
+    }]}
+    state.screener_preview = {"items": []}
+    return state
+
+
+@pytest.fixture
+def dedup_mock():
+    d = MagicMock()
+    d.is_recent.return_value = False
+    return d
+
+
+def test_buy_signal_when_all_conditions_pass_and_confidence_high(dedup_mock):
+    state = _make_state_with_buy_candidate()
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" in state.signals
+    sig = state.signals["005930"]
+    assert sig["action"] == "buy"
+    assert sig["confidence_webai"] > 0.7
+    dedup_mock.record.assert_called()
+
+
+def test_silent_when_chronos_median_negative(dedup_mock):
+    state = _make_state_with_buy_candidate(chronos_median=-0.01)
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" not in state.signals
+
+
+def test_silent_when_distribution_spread_too_wide(dedup_mock):
+    # spread = q90 - q10 = 0.5 - (-0.5) = 1.0 > 0.6 → hard gate fails
+    state = _make_state_with_buy_candidate(
+        chronos_median=0.001, chronos_q10=-0.5, chronos_q90=0.5,
+    )
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" not in state.signals
+
+
+def test_silent_when_momentum_not_strong_up(dedup_mock):
+    state = _make_state_with_buy_candidate(momentum="weak_up")
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" not in state.signals
+
+
+def test_silent_when_bid_ratio_below_threshold(dedup_mock):
+    state = _make_state_with_buy_candidate(bid_ratio=0.5)
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" not in state.signals
+
+
+def test_silent_when_confidence_below_threshold(dedup_mock):
+    # chronos_conf low + rank=20 → confidence < 0.7
+    state = _make_state_with_buy_candidate(chronos_conf=0.3)
+    # add 19 fake items to push 005930 rank to 20
+    state.screener_preview["items"] = (
+        [{"ticker": f"FAKE{i:03d}"} for i in range(19)]
+        + [{"ticker": "005930", "name": "삼성전자"}]
+    )
+    generate_signals(state, dedup_mock, _settings())
+    # confidence_webai = 0.3*0.5 + 1.0*0.3 + 0.05*0.2 = 0.46 < 0.7
+    assert "005930" not in state.signals
+
+
+def test_sell_signal_when_stop_loss_triggered(dedup_mock):
+    state = _make_state_with_holding(pnl_pct=-0.08, current_price=69000, avg_price=75000)
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" in state.signals
+    sig = state.signals["005930"]
+    assert sig["action"] == "sell"
+    assert sig["confidence_webai"] == 1.0
+    assert sig["pnl_pct"] == -0.08
+
+
+def test_sell_signal_when_take_profit_triggered(dedup_mock):
+    state = _make_state_with_holding(pnl_pct=0.16, current_price=87000, avg_price=75000)
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" in state.signals
+    sig = state.signals["005930"]
+    assert sig["action"] == "sell"
+    assert sig["confidence_webai"] == 0.6
+
+
+def test_silent_when_dedup_recently_sent(dedup_mock):
+    state = _make_state_with_buy_candidate()
+    dedup_mock.is_recent.return_value = True
+    generate_signals(state, dedup_mock, _settings())
+    assert "005930" not in state.signals
+    dedup_mock.record.assert_not_called()
+
+
+def test_sell_signal_triggers_on_anomaly_path(dedup_mock):
+    """Anomaly sell: median < -1%, momentum strong_down, low bid_ratio, confidence > threshold."""
+    state = PollState()
+    state.portfolio = {"holdings": [{
+        "ticker": "005930", "name": "삼성전자",
+        "avg_price": 75000, "current_price": 70000,
+        "pnl_pct": -0.067,  # within stop_loss tolerance (default -0.07): NOT triggering stop_loss
+        "quantity": 100, "broker": "키움",
+    }]}
+    state.screener_preview = {"items": []}
+    state.chronos_predictions["005930"] = {
+        "median": -0.025, "q10": -0.05, "q90": 0.005, "conf": 0.85,
+    }
+    state.minute_momentum["005930"] = "strong_down"
+    state.asking_price["005930"] = {"current_price": 70000, "bid_ratio": 0.30}
+    # bid_ratio 0.30 < (1 - 0.6) = 0.4 → anomaly bid_ratio gate passes
+    # confidence = 0.85*0.5 + 1.0*0.3 + 1.0*0.2 = 0.425 + 0.3 + 0.2 = 0.925 > 0.7
+
+    generate_signals(state, dedup_mock, _settings())
+
+    assert "005930" in state.signals
+    sig = state.signals["005930"]
+    assert sig["action"] == "sell"
+    assert sig["context"]["sell_reason"] == "anomaly"
+    assert sig["confidence_webai"] > 0.7