Initial commit: stock analysis backend and prototype UI.

Co-authored-by: Cursor <cursoragent@cursor.com>

.gitignore

@@ -0,0 +1,10 @@
+# 密钥与本地配置
+backend/.env
+
+# Python
+backend/.venv/
+__pycache__/
+*.pyc
+
+# 临时/日志
+*.log

backend/.env.example

@@ -0,0 +1,29 @@
+# 复制本文件为 .env 并填入你的密钥，然后重启后端（python main.py）生效。
+
+# ===== 大模型（OpenAI 兼容；任选其一）=====
+# DeepSeek
+LLM_API_KEY=
+LLM_BASE_URL=https://api.deepseek.com/v1
+LLM_MODEL=deepseek-chat
+# 通义千问（示例）
+# LLM_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
+# LLM_MODEL=qwen-plus
+# Kimi（示例）
+# LLM_BASE_URL=https://api.moonshot.cn/v1
+# LLM_MODEL=moonshot-v1-8k
+
+# ===== 邮件推送（SMTP）=====
+# 以QQ邮箱为例：SMTP_HOST=smtp.qq.com  SMTP_PORT=465  SMTP_PASSWORD填“授权码”而非登录密码
+SMTP_HOST=
+SMTP_PORT=465
+SMTP_USER=
+SMTP_PASSWORD=
+SMTP_TO=
+
+# ===== 微信推送（任选其一）=====
+# Server酱 Turbo：https://sct.ftqq.com 获取 SendKey
+SERVERCHAN_KEY=
+# 企业微信群机器人 webhook 完整地址
+WECOM_WEBHOOK=
+# PushPlus：https://www.pushplus.plus 获取 token
+PUSHPLUS_TOKEN=

backend/ai.py

@@ -0,0 +1,232 @@
+"""AI 分析层：基于中台数据构造上下文，调用大模型生成点评；无 key 时规则降级。"""
+from __future__ import annotations
+
+import datetime as dt
+
+from sqlalchemy import select, func, desc
+
+import llm
+import signals
+import rag
+from db import get_session
+from models import (SectorDaily, FundFlowDaily, SentimentDaily, DragonTiger,
+                    StockMetric, DailyQuote, Security)
+
+DISCLAIMER = "（注：以上为基于数据的程序化分析，不构成投资建议，市场有风险。）"
+
+
+# ============ 每日复盘点评 ============
+def _daily_context(date=None):
+    with get_session() as s:
+        d = dt.date.fromisoformat(date) if date else s.execute(select(func.max(SectorDaily.date))).scalar()
+        if not d:
+            return None
+        secs = s.execute(select(SectorDaily).where(SectorDaily.date == d).order_by(SectorDaily.pct.desc())).scalars().all()
+        flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d).order_by(FundFlowDaily.net.desc())).scalars().all()
+        senti = s.execute(select(SentimentDaily).where(SentimentDaily.date == d)).scalar_one_or_none()
+        lhb = s.execute(select(DragonTiger).where(DragonTiger.date == d).order_by(DragonTiger.net.desc()).limit(6)).scalars().all()
+    return {
+        "date": d, "senti": senti,
+        "top": [(x.name, x.pct) for x in secs[:6]],
+        "bot": [(x.name, x.pct) for x in secs[-4:]],
+        "inflow": [(x.name, x.net) for x in flows[:6]],
+        "outflow": [(x.name, x.net) for x in flows[-4:][::-1]],
+        "lhb": [(x.name, x.net) for x in lhb],
+    }
+
+
+def review_daily_comment(date=None):
+    ctx = _daily_context(date)
+    if not ctx:
+        return {"ok": False, "msg": "暂无入库数据，请先到数据中台入库"}
+    s = ctx["senti"]
+    senti_line = (f"上涨{s.up}家/下跌{s.down}家，涨停{s.limit_up}家/跌停{s.limit_down}家" if s else "无情绪数据")
+    facts = (
+        f"日期：{ctx['date']}\n"
+        f"市场情绪：{senti_line}\n"
+        f"领涨板块：{', '.join(f'{n}({p:+.2f}%)' for n,p in ctx['top'])}\n"
+        f"领跌板块：{', '.join(f'{n}({p:+.2f}%)' for n,p in ctx['bot'])}\n"
+        f"主力净流入前列：{', '.join(f'{n}({v:+.1f}亿)' for n,v in ctx['inflow'])}\n"
+        f"主力净流出前列：{', '.join(f'{n}({v:+.1f}亿)' for n,v in ctx['outflow'])}\n"
+        f"龙虎榜净买额居前：{', '.join(f'{n}({v:+.2f}亿)' for n,v in ctx['lhb'])}\n"
+    )
+    if llm.enabled():
+        try:
+            prompt = ("请根据以下当日A股盘面数据，撰写一篇 250 字以内的收盘复盘点评，"
+                      "包含：① 今日市场情绪与赚钱效应判断；② 资金与题材主线；③ 明日需关注的方向与风险。"
+                      "分段清晰。\n\n" + facts)
+            text = llm.ask(prompt, temperature=0.6)
+            return {"ok": True, "source": "llm", "date": ctx["date"].isoformat(), "facts": facts, "text": text}
+        except Exception as e:
+            pass
+    # 规则降级
+    tone = "情绪偏暖、赚钱效应尚可" if (s and s.up > s.down) else "情绪偏弱、需控制仓位"
+    text = (f"今日{tone}。{senti_line}。\n"
+            f"领涨主线集中在 {', '.join(n for n,_ in ctx['top'][:3])}，"
+            f"主力资金净流入居前的是 {', '.join(n for n,_ in ctx['inflow'][:3])}；"
+            f"而 {', '.join(n for n,_ in ctx['outflow'][:2])} 遭资金流出，需回避。\n"
+            f"明日关注领涨板块的持续性与量能配合，注意高位股退潮风险。\n{DISCLAIMER}")
+    return {"ok": True, "source": "rule", "date": ctx["date"].isoformat(), "facts": facts, "text": text}
+
+
+# ============ 个股诊断 ============
+def _stock_context(symbol):
+    with get_session() as s:
+        m = s.get(StockMetric, symbol)
+        sec = s.get(Security, symbol)
+        rows = s.execute(select(DailyQuote).where(DailyQuote.code == symbol)
+                         .order_by(DailyQuote.date.desc()).limit(60)).scalars().all()
+    return m, sec, list(reversed(rows))
+
+
+def _build_evidence(m, stats, news_tone):
+    """构造证据链：每条含 维度/事实/方向/历史命中率。"""
+    def hr(key):
+        s = stats.get(key)
+        return (s["win_rate"], s["avg_ret"], s["samples"]) if s else (None, None, None)
+
+    ev = []
+    # 趋势
+    wr, ar, ns = hr("ma_bull")
+    ev.append({"dim": "趋势", "fact": f"均线{'多头排列(MA5>MA10>MA20)' if m.ma_bull else '未呈多头'}，20日{m.ret20:+.1f}%",
+               "signal": "bull" if m.ma_bull else ("bear" if m.ret20 < -3 else "neutral"),
+               "win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 1.2})
+    # 技术(MACD)
+    wr, ar, ns = hr("macd_gold")
+    ev.append({"dim": "技术", "fact": f"{'MACD金叉' if m.macd_gold else 'MACD未金叉'}，RSI14={m.rsi14:.0f}"
+               + ("（超买）" if m.rsi14 >= 80 else ("（超卖）" if m.rsi14 < 30 else "")),
+               "signal": "bull" if m.macd_gold else ("bear" if m.rsi14 >= 80 else "neutral"),
+               "win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 1.0})
+    # 动量
+    wr, ar, ns = hr("up_streak3")
+    streak_sig = "bull" if m.up_streak >= 3 else ("bear" if m.ret5 < -5 else "neutral")
+    ev.append({"dim": "动量", "fact": f"连涨{m.up_streak}日，5日{m.ret5:+.1f}%",
+               "signal": streak_sig, "win_rate": wr if m.up_streak >= 3 else None,
+               "avg_ret": ar if m.up_streak >= 3 else None, "samples": ns if m.up_streak >= 3 else None, "weight": 1.0})
+    # 资金
+    wr, ar, ns = hr("vol_breakout")
+    vol_sig = "bull" if (m.vol_ratio > 2 and m.pct > 0) else ("bear" if (m.vol_ratio > 2 and m.pct < 0) else "neutral")
+    ev.append({"dim": "资金", "fact": f"量比{m.vol_ratio}，当日{m.pct:+.2f}%，成交额{m.amount}亿",
+               "signal": vol_sig, "win_rate": wr if vol_sig != "neutral" else None,
+               "avg_ret": ar if vol_sig != "neutral" else None, "samples": ns if vol_sig != "neutral" else None, "weight": 1.1})
+    # 位置
+    key = "new_high60" if m.pos60 >= 0.99 else ("rsi_oversold" if m.rsi14 < 30 else None)
+    wr, ar, ns = hr(key) if key else (None, None, None)
+    pos_sig = "bear" if m.pos60 > 0.92 else ("bull" if m.pos60 < 0.2 else "neutral")
+    ev.append({"dim": "位置", "fact": f"处于60日 {m.pos60*100:.0f}% 分位" + ("（创新高）" if m.pos60 >= 0.99 else ("（低位）" if m.pos60 < 0.2 else "")),
+               "signal": pos_sig, "win_rate": wr, "avg_ret": ar, "samples": ns, "weight": 0.8})
+    # 消息面（RAG）
+    ev.append({"dim": "消息", "fact": f"近期资讯情绪：{news_tone}",
+               "signal": {"利好": "bull", "利空": "bear"}.get(news_tone, "neutral"),
+               "win_rate": None, "avg_ret": None, "samples": None, "weight": 0.9})
+    return ev
+
+
+def _confidence_direction(ev):
+    sval = {"bull": 1, "bear": -1, "neutral": 0}
+    net = sum(sval[e["signal"]] * e["weight"] for e in ev)
+    tw = sum(e["weight"] for e in ev if e["signal"] != "neutral")
+    agreement = abs(net) / tw if tw else 0.0
+    conf = int(max(5, min(96, 35 + agreement * 55 + (8 if tw >= 3 else 0))))
+    thr = 0.15 * sum(e["weight"] for e in ev)
+    direction = "up" if net > thr else ("down" if net < -thr else "flat")
+    return conf, direction, round(net, 2)
+
+
+def diagnose(symbol):
+    m, sec, rows = _stock_context(symbol)
+    if not m:
+        return {"ok": False, "msg": "该股票无因子数据，请先入库（数据中台→全市场回填或指定入库）"}
+    name = (sec.name if sec else m.name) or symbol
+
+    # 规则打分（0~100）
+    trend = 50 + (15 if m.ma_bull else -10) + min(20, max(-20, m.ret20))
+    momentum = 50 + min(25, max(-25, m.ret5)) + (10 if 50 < m.rsi14 < 75 else (-10 if m.rsi14 >= 80 else 0))
+    capital = 50 + min(30, (m.vol_ratio - 1) * 20) * (1 if m.pct > 0 else -1)
+    position = 100 - abs(m.pos60 * 100 - 45)  # 价格分位适中得分高
+    clamp = lambda x: int(max(5, min(98, x)))
+    scores = {"趋势": clamp(trend), "动量": clamp(momentum), "资金": clamp(capital), "位置": clamp(position)}
+    total = clamp(sum(scores.values()) / 4)
+
+    # 证据链 + 历史命中率 + RAG 资讯
+    stats = signals.get_stats(horizon=5)
+    rctx = rag.stock_context(symbol, limit=5)
+    evidence = _build_evidence(m, stats, rctx["tone"])
+    confidence, direction, net = _confidence_direction(evidence)
+
+    # 留痕：本次诊断写入预测，N 日后核验
+    try:
+        signals.record_prediction(symbol, name, m.date, total, confidence, direction, m.close, horizon=5)
+    except Exception:
+        pass
+
+    facts = (
+        f"股票：{name}({symbol})\n"
+        f"最新价：{m.close}，当日{m.pct:+.2f}%\n"
+        f"均线：MA5={m.ma5} MA10={m.ma10} MA20={m.ma20} MA60={m.ma60}，{'多头排列' if m.ma_bull else '非多头'}\n"
+        f"涨幅：5日{m.ret5:+.2f}% / 20日{m.ret20:+.2f}% / 60日{m.ret60:+.2f}%\n"
+        f"量比：{m.vol_ratio}，RSI14：{m.rsi14}，60日价格分位：{m.pos60*100:.0f}%，"
+        f"{'MACD金叉' if m.macd_gold else 'MACD未金叉'}，连涨{m.up_streak}日，成交额{m.amount}亿\n"
+    )
+    hit_lines = "\n".join(
+        f"- {e['dim']}信号「{e['fact']}」历史5日上涨概率 {e['win_rate']}%、平均{e['avg_ret']:+.2f}%（样本{e['samples']}）"
+        for e in evidence if e.get("win_rate") is not None)
+    base = {"ok": True, "symbol": symbol, "name": name, "scores": scores, "total": total,
+            "confidence": confidence, "direction": direction, "evidence": evidence,
+            "news": rctx["items"], "news_tone": rctx["tone"], "facts": facts}
+
+    if llm.enabled():
+        try:
+            prompt = ("请基于以下个股量化数据、信号历史命中率与相关资讯，输出结构化诊断：\n"
+                      "1) 一句话结论（含方向与置信判断）；2) 技术面/资金面/趋势面分别点评（引用历史命中率）；"
+                      "3) 操作建议（含参考性关注位与止损思路）；4) 主要风险。200~320字。\n\n"
+                      f"{facts}\n历史命中率：\n{hit_lines or '（暂无回测数据，请先在AI准确率页计算）'}\n\n"
+                      f"{rctx['block'] or '近期无相关资讯。'}\n")
+            text = llm.ask(prompt, temperature=0.4)
+            base.update({"source": "llm", "text": text})
+            return base
+        except Exception:
+            pass
+
+    judge = "偏强" if total >= 60 else ("偏弱" if total < 45 else "中性")
+    dir_cn = {"up": "看多", "down": "看空", "flat": "中性观望"}[direction]
+    text = (f"综合评分 {total}（{judge}），方向{dir_cn}，置信度 {confidence}%。"
+            f"趋势{'多头向上' if m.ma_bull else '尚未走强'}，20日{m.ret20:+.1f}%；"
+            f"动量 RSI {m.rsi14:.0f}{'，超买防回调' if m.rsi14>=80 else ''}；"
+            f"资金{'放量' if m.vol_ratio>=1.5 else '量能一般'}（量比{m.vol_ratio}）；"
+            f"价格处于60日 {m.pos60*100:.0f}% 分位；消息面{rctx['tone']}。\n"
+            + (f"依据信号历史表现：\n{hit_lines}\n" if hit_lines else "")
+            + f"操作：{'回踩不破MA20可关注' if m.ma_bull else '等待站稳均线再观察'}，跌破MA20或前低则减仓。"
+            f"风险：{'高位放量谨防见顶' if m.pos60>0.9 else '大盘系统性波动'}。\n{DISCLAIMER}")
+    base.update({"source": "rule", "text": text})
+    return base
+
+
+# ============ 今日策略 ============
+def today_strategy():
+    with get_session() as s:
+        d = s.execute(select(func.max(SectorDaily.date))).scalar()
+        secs = s.execute(select(SectorDaily).where(SectorDaily.date == d).order_by(SectorDaily.pct.desc()).limit(5)).scalars().all() if d else []
+        flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d).order_by(FundFlowDaily.net.desc()).limit(5)).scalars().all() if d else []
+        strong = s.execute(select(StockMetric).where(StockMetric.up_streak >= 3).order_by(StockMetric.up_streak.desc()).limit(8)).scalars().all()
+        macd = s.execute(select(StockMetric).where(StockMetric.macd_gold.is_(True)).order_by(StockMetric.ret5.desc()).limit(8)).scalars().all()
+    facts = (
+        f"领涨板块：{', '.join(f'{x.name}({x.pct:+.2f}%)' for x in secs)}\n"
+        f"主力净流入板块：{', '.join(f'{x.name}({x.net:+.1f}亿)' for x in flows)}\n"
+        f"强势连涨个股：{', '.join(f'{x.name}({x.up_streak}连阳)' for x in strong)}\n"
+        f"MACD金叉个股：{', '.join(x.name for x in macd)}\n"
+    )
+    if llm.enabled():
+        try:
+            prompt = ("请基于以下盘面数据，给出『今天/接下来怎么做』的策略观点："
+                      "① 当前资金与题材主线；② 值得关注的方向（给理由）；③ 需要回避的风险。"
+                      "220字以内。\n\n" + facts)
+            text = llm.ask(prompt, temperature=0.6)
+            return {"ok": True, "source": "llm", "facts": facts, "text": text}
+        except Exception:
+            pass
+    main = "、".join(x.name for x in secs[:3]) or "暂无数据"
+    text = (f"当前资金主线集中在 {main}。"
+            f"可重点跟踪净流入居前的 {', '.join(x.name for x in flows[:3])} 板块内的强势个股；"
+            f"强势连涨股需注意分歧与退潮节奏，避免追高。\n{DISCLAIMER}")
+    return {"ok": True, "source": "rule", "facts": facts, "text": text}

backend/akshare_service.py

@@ -0,0 +1,444 @@
+"""AkShare 数据服务层。
+
+每个函数都做了 try/except 降级：真实数据拿不到时返回 Python 端生成的模拟数据，
+并通过 `source` 字段标注来源（akshare / mock），保证前端任何情况下都有数据可渲染。
+"""
+from __future__ import annotations
+
+import random
+import datetime as dt
+from functools import wraps
+
+from cachetools import TTLCache
+
+import requests
+
+try:
+    import akshare as ak
+    AK_OK = True
+except Exception:  # akshare 未安装也能跑（全部走 mock）
+    ak = None
+    AK_OK = False
+
+# ---- 简单 TTL 缓存（按函数+参数） ----
+_cache = TTLCache(maxsize=256, ttl=30)
+
+
+def cached(ttl: int):
+    def deco(fn):
+        local = TTLCache(maxsize=64, ttl=ttl)
+
+        @wraps(fn)
+        def wrapper(*args, **kwargs):
+            key = (fn.__name__, args, tuple(sorted(kwargs.items())))
+            if key in local:
+                return local[key]
+            val = fn(*args, **kwargs)
+            local[key] = val
+            return val
+
+        return wrapper
+
+    return deco
+
+
+def _rnd(a, b):
+    return round(random.uniform(a, b), 2)
+
+
+# ============================================================
+# 指数
+# ============================================================
+MAJOR_INDEX = {
+    "sh000001": ("上证指数", 3210),
+    "sz399001": ("深证成指", 10180),
+    "sz399006": ("创业板指", 2105),
+    "sh000300": ("沪深300", 3760),
+    "bj899050": ("北证50", 1080),
+}
+
+
+@cached(10)
+def get_indices():
+    if AK_OK:
+        try:
+            df = ak.stock_zh_index_spot_sina()
+            rows = []
+            for code, (name, _base) in MAJOR_INDEX.items():
+                r = df[df["代码"] == code]
+                if r.empty:
+                    continue
+                r = r.iloc[0]
+                rows.append({
+                    "code": code, "name": name,
+                    "price": float(r["最新价"]),
+                    "change": float(r["涨跌额"]),
+                    "pct": float(r["涨跌幅"]),
+                })
+            if rows:
+                return {"source": "akshare", "list": rows}
+        except Exception as e:  # noqa
+            pass
+    # mock
+    rows = []
+    for code, (name, base) in MAJOR_INDEX.items():
+        pct = _rnd(-2.5, 2.5)
+        price = round(base * (1 + pct / 100), 2)
+        rows.append({"code": code, "name": name, "price": price,
+                     "change": round(price - base, 2), "pct": pct})
+    return {"source": "mock", "list": rows}
+
+
+# ============================================================
+# K线
+# ============================================================
+# ============================================================
+# 实时报价（新浪 hq，速度快且稳定，用于盯盘预警）
+# ============================================================
+def realtime_quotes(codes):
+    """返回 {code: {name, price, prev_close, pct, open, high, low}}。失败返回 {}。"""
+    if not codes:
+        return {}
+    syms = ",".join(_sina_symbol(c) for c in codes)
+    try:
+        r = requests.get("https://hq.sinajs.cn/list=" + syms,
+                         headers={"Referer": "https://finance.sina.com.cn"}, timeout=6)
+        out = {}
+        for line in r.text.split(";\n"):
+            if "hq_str_" not in line or '="' not in line:
+                continue
+            head, body = line.split('="', 1)
+            sym = head.split("hq_str_")[1].strip()
+            code = sym[2:]
+            f = body.strip('"').split(",")
+            if len(f) < 6 or not f[3]:
+                continue
+            price = float(f[3]); prev = float(f[2]) if f[2] else 0.0
+            out[code] = {"name": f[0], "open": float(f[1] or 0), "prev_close": prev,
+                         "price": price, "high": float(f[4] or 0), "low": float(f[5] or 0),
+                         "pct": round((price - prev) / prev * 100, 2) if prev else 0.0}
+        return out
+    except Exception:
+        return {}
+
+
+# ============================================================
+# 资讯新闻
+# ============================================================
+_BULL = ["涨停", "利好", "增长", "大涨", "突破", "中标", "签约", "回购", "增持", "扭亏",
+         "超预期", "新高", "提价", "涨价", "订单", "合作", "获批", "盈利", "分红", "重组",
+         "并购", "补贴", "减税", "降准", "降息", "刺激", "国产替代", "放量", "净流入"]
+_BEAR = ["跌停", "利空", "下滑", "大跌", "亏损", "减持", "处罚", "退市", "违规", "下调",
+         "不及预期", "新低", "停牌", "质押", "爆雷", "诉讼", "解禁", "商誉", "预亏", "降价",
+         "裁员", "债务", "暴跌", "净流出", "风险警示"]
+
+
+def judge_sentiment(text: str):
+    t = text or ""
+    pos = [w for w in _BULL if w in t]
+    neg = [w for w in _BEAR if w in t]
+    if len(pos) > len(neg):
+        return "利好", pos[:4]
+    if len(neg) > len(pos):
+        return "利空", neg[:4]
+    return "中性", (pos or neg)[:4]
+
+
+@cached(120)
+def get_news(limit: int = 40):
+    if AK_OK:
+        try:
+            df = ak.stock_info_global_em()
+            rows = []
+            for _, r in df.head(limit).iterrows():
+                title = str(r["标题"]); summary = str(r.get("摘要", ""))
+                senti, kw = judge_sentiment(title + summary)
+                rows.append({"time": str(r["发布时间"]), "title": title, "summary": summary,
+                             "url": str(r.get("链接", "")), "sentiment": senti, "keywords": kw})
+            if rows:
+                return {"source": "akshare", "list": rows}
+        except Exception:
+            pass
+    return {"source": "mock", "list": [
+        {"time": "—", "title": "示例资讯：市场情绪回暖，多板块走强", "summary": "（演示数据）",
+         "sentiment": "利好", "keywords": ["利好"], "url": ""}]}
+
+
+@cached(180)
+def get_stock_news(code: str, limit: int = 12):
+    if AK_OK:
+        try:
+            df = ak.stock_news_em(symbol=code)
+            rows = []
+            for _, r in df.head(limit).iterrows():
+                title = str(r["新闻标题"]); content = str(r.get("新闻内容", ""))
+                senti, kw = judge_sentiment(title + content)
+                rows.append({"time": str(r["发布时间"]), "title": title,
+                             "summary": content[:120], "source": str(r.get("文章来源", "")),
+                             "url": str(r.get("新闻链接", "")), "sentiment": senti, "keywords": kw})
+            if rows:
+                return {"source": "akshare", "list": rows}
+        except Exception:
+            pass
+    return {"source": "mock", "list": []}
+
+
+def _sina_symbol(code: str) -> str:
+    if code.startswith("6"):
+        return "sh" + code
+    if code.startswith(("0", "3")):
+        return "sz" + code
+    if code.startswith(("8", "4")):
+        return "bj" + code
+    return "sh" + code
+
+
+@cached(60)
+def get_kline(symbol: str = "600519", days: int = 120):
+    if AK_OK:
+        # 主源：新浪日线（更稳定）；备源：腾讯
+        for src in ("sina", "tx"):
+            try:
+                sym = _sina_symbol(symbol)
+                if src == "sina":
+                    df = ak.stock_zh_a_daily(symbol=sym, adjust="qfq")
+                else:
+                    df = ak.stock_zh_a_hist_tx(symbol=sym)
+                if df is not None and not df.empty:
+                    df = df.tail(days)
+                    dates = [str(d)[5:].replace("-", "/") for d in df["date"]]
+                    ohlc = [[float(o), float(c), float(l), float(h)] for o, c, l, h in
+                            zip(df["open"], df["close"], df["low"], df["high"])]
+                    vols = [int(v) for v in (df["volume"] if "volume" in df.columns else df["amount"])]
+                    return {"source": "akshare", "symbol": symbol, "dates": dates, "ohlc": ohlc, "vols": vols}
+            except Exception:
+                continue
+    # mock
+    dates, ohlc, vols = [], [], []
+    price = 1680.0
+    today = dt.date.today()
+    for i in range(days, 0, -1):
+        d = today - dt.timedelta(days=i)
+        dates.append(f"{d.month}/{d.day}")
+        o = price
+        c = round(o + _rnd(-o * 0.03, o * 0.03), 2)
+        h = round(max(o, c) + _rnd(0, o * 0.02), 2)
+        l = round(min(o, c) - _rnd(0, o * 0.02), 2)
+        ohlc.append([o, c, l, h])
+        vols.append(int(_rnd(2, 9) * 1e6))
+        price = c
+    return {"source": "mock", "symbol": symbol, "dates": dates, "ohlc": ohlc, "vols": vols}
+
+
+# ============================================================
+# 行业板块（云图 / 热门板块复用）—— 新浪行业（东财push2在部分网络被封）
+# ============================================================
+@cached(60)
+def get_industry_boards():
+    if AK_OK:
+        try:
+            df = ak.stock_sector_spot(indicator="新浪行业")
+            rows = []
+            for _, r in df.iterrows():
+                rows.append({
+                    "name": str(r["板块"]),
+                    "pct": float(r["涨跌幅"]),
+                    "amount": round(float(r["总成交额"]) / 1e8, 1),  # 亿
+                    "count": int(r.get("公司家数", 0) or 0),
+                    "leader": str(r.get("股票名称", "")),
+                })
+            if rows:
+                rows.sort(key=lambda x: x["pct"], reverse=True)
+                return {"source": "akshare", "list": rows}
+        except Exception:
+            pass
+    sectors = ["半导体", "新能源", "医药生物", "白酒食品", "军工", "证券", "银行",
+               "房地产", "人工智能", "消费电子", "光伏", "汽车", "有色金属", "煤炭", "传媒"]
+    return {"source": "mock", "list": [
+        {"name": s, "pct": _rnd(-3, 6), "amount": _rnd(50, 500),
+         "count": int(_rnd(10, 80)), "leader": "龙头股"} for s in sectors]}
+
+
+# ============================================================
+# 全市场快照（情绪 / 全市场云图）
+# ============================================================
+@cached(60)
+def _spot():
+    if AK_OK:
+        try:
+            df = ak.stock_zh_a_spot_em()
+            if df is not None and not df.empty:
+                return df
+        except Exception:
+            pass
+    return None
+
+
+@cached(30)
+def get_sentiment():
+    if AK_OK:
+        try:
+            df = ak.stock_market_activity_legu()
+            m = {}
+            for _, r in df.iterrows():
+                m[str(r["item"]).strip()] = r["value"]
+
+            def num(k):
+                try:
+                    return int(float(m.get(k, 0)))
+                except Exception:
+                    return 0
+            up, down, flat = num("上涨"), num("下跌"), num("平盘")
+            if up or down:
+                return {"source": "akshare", "up": up, "down": down, "flat": flat,
+                        "limit_up": num("涨停"), "limit_down": num("跌停"),
+                        "height": min(9, max(3, num("涨停") // 8))}
+        except Exception:
+            pass
+    up, down, flat = int(_rnd(1800, 3200)), int(_rnd(1200, 2600)), int(_rnd(80, 260))
+    return {"source": "mock", "up": up, "down": down, "flat": flat,
+            "limit_up": int(_rnd(20, 90)), "limit_down": int(_rnd(2, 30)), "height": int(_rnd(4, 9))}
+
+
+@cached(60)
+def get_treemap(mode: str = "sector"):
+    if mode == "all":
+        df = _spot()
+        if df is not None:
+            try:
+                top = df.sort_values("成交额", ascending=False).head(150)
+                items = [{"name": str(r["名称"]), "value": round(float(r["成交额"]) / 1e8, 2),
+                          "pct": float(r["涨跌幅"])} for _, r in top.iterrows()]
+                return {"source": "akshare", "mode": "all", "items": items}
+            except Exception:
+                pass
+        # mock flat
+        items = [{"name": f"个股{i}", "value": _rnd(2, 50), "pct": _rnd(-9, 9)} for i in range(60)]
+        return {"source": "mock", "mode": "all", "items": items}
+    # sector
+    boards = get_industry_boards()
+    items = [{"name": b["name"], "value": b.get("amount", 1), "pct": b["pct"]} for b in boards["list"]]
+    return {"source": boards["source"], "mode": "sector", "items": items}
+
+
+# ============================================================
+# 资金流向（行业）
+# ============================================================
+@cached(60)
+def get_fund_flow():
+    if AK_OK:
+        try:
+            df = ak.stock_fund_flow_industry(symbol="即时")
+            rows = []
+            for _, r in df.iterrows():
+                rows.append({"name": str(r["行业"]),
+                             "net": round(float(r["净额"]), 2),  # 同花顺已是亿元
+                             "pct": float(r["行业-涨跌幅"])})
+            if rows:
+                rows.sort(key=lambda x: x["net"])
+                # 取首尾各15条，突出流入流出两端
+                show = rows[:15] + rows[-15:] if len(rows) > 30 else rows
+                show.sort(key=lambda x: x["net"])
+                return {"source": "akshare", "list": show}
+        except Exception:
+            pass
+    sectors = ["半导体", "新能源", "医药生物", "白酒食品", "军工", "证券", "银行",
+               "房地产", "人工智能", "消费电子", "光伏", "汽车", "有色金属", "煤炭", "传媒"]
+    rows = [{"name": s, "net": _rnd(-40, 60), "pct": _rnd(-3, 6)} for s in sectors]
+    rows.sort(key=lambda x: x["net"])
+    return {"source": "mock", "list": rows}
+
+
+# ============================================================
+# 热门股票（人气榜）
+# ============================================================
+@cached(60)
+def get_hot_stocks():
+    if AK_OK:
+        try:
+            df = ak.stock_hot_rank_em()
+            rows = []
+            for _, r in df.head(20).iterrows():
+                rows.append({"rank": int(r["当前排名"]), "code": str(r["代码"]),
+                             "name": str(r["股票名称"]), "price": float(r["最新价"]),
+                             "pct": float(r["涨跌幅"])})
+            if rows:
+                return {"source": "akshare", "list": rows}
+        except Exception:
+            pass
+    pool = ["龙头A", "龙头B", "中军C", "黑马D", "次新E", "蓝筹F", "题材G", "妖股H"]
+    return {"source": "mock", "list": [
+        {"rank": i + 1, "code": f"60{1000+i}", "name": f"{pool[i % len(pool)]}{i}",
+         "price": _rnd(5, 200), "pct": _rnd(-5, 11)} for i in range(15)]}
+
+
+# ============================================================
+# 龙虎榜
+# ============================================================
+@cached(300)
+def get_dragon_tiger():
+    if AK_OK:
+        try:
+            for back in range(0, 7):
+                d = (dt.date.today() - dt.timedelta(days=back)).strftime("%Y%m%d")
+                try:
+                    df = ak.stock_lhb_detail_em(start_date=d, end_date=d)
+                except Exception:
+                    df = None
+                if df is not None and not df.empty:
+                    rows = []
+                    for _, r in df.head(20).iterrows():
+                        rows.append({
+                            "code": str(r.get("代码", "")), "name": str(r.get("名称", "")),
+                            "pct": float(r.get("涨跌幅", 0) or 0),
+                            "net": round(float(r.get("龙虎榜净买额", 0) or 0) / 1e8, 2),
+                            "reason": str(r.get("上榜原因", "")),
+                        })
+                    return {"source": "akshare", "date": d, "list": rows}
+        except Exception:
+            pass
+    pool = ["龙头A", "龙头B", "中军C", "黑马D"]
+    return {"source": "mock", "date": "", "list": [
+        {"code": f"60{1000+i}", "name": f"{pool[i % len(pool)]}{i}", "pct": _rnd(-3, 10),
+         "net": _rnd(-3, 5), "reason": ["日涨幅偏离", "换手率达20%", "连续三日涨停"][i % 3]} for i in range(12)]}
+
+
+# ============================================================
+# 自选股 —— 代码名称表 + 个股日线（push2 被封时的稳妥方案）
+# ============================================================
+@cached(3600)
+def _code_name_map():
+    if AK_OK:
+        try:
+            cn = ak.stock_info_a_code_name()
+            return {str(r["code"]): str(r["name"]) for _, r in cn.iterrows()}
+        except Exception:
+            pass
+    return {}
+
+
+def get_watchlist(symbols: list[str]):
+    names = {"600519": "贵州茅台", "300750": "宁德时代", "002594": "比亚迪",
+             "688981": "中芯国际", "300059": "东方财富", "601012": "隆基绿能"}
+    if AK_OK:
+        cmap = _code_name_map()
+        rows = []
+        for s in symbols:
+            try:
+                k = get_kline(s, 30)
+                if k["source"] != "akshare" or len(k["ohlc"]) < 2:
+                    continue
+                last, prev = k["ohlc"][-1], k["ohlc"][-2]
+                price, prev_close = last[1], prev[1]
+                change = round(price - prev_close, 2)
+                pct = round(change / prev_close * 100, 2) if prev_close else 0.0
+                rows.append({"code": s, "name": cmap.get(s, names.get(s, s)), "price": price,
+                             "pct": pct, "change": change,
+                             "amount": round(k["vols"][-1] * price / 1e8, 2)})
+            except Exception:
+                continue
+        if rows:
+            return {"source": "akshare", "list": rows}
+    return {"source": "mock", "list": [
+        {"code": c, "name": names.get(c, c), "price": _rnd(20, 1800), "pct": _rnd(-4, 5),
+         "change": _rnd(-30, 30), "amount": _rnd(3, 60)} for c in symbols]}

backend/alerts.py

@@ -0,0 +1,67 @@
+"""预警引擎：拉取实时价格，评估规则，触发即生成站内事件。"""
+from __future__ import annotations
+
+import datetime as dt
+
+from sqlalchemy import select
+
+import akshare_service as svc
+import notifier
+from db import get_session
+from models import AlertRule, AlertEvent
+
+KIND_LABEL = {
+    "price_above": "价格突破",
+    "price_below": "价格跌破",
+    "pct_above": "涨幅达到",
+    "pct_below": "跌幅达到",
+}
+
+
+def _hit(kind, threshold, q):
+    price, pct = q["price"], q["pct"]
+    if kind == "price_above":
+        return price >= threshold, price
+    if kind == "price_below":
+        return price <= threshold, price
+    if kind == "pct_above":
+        return pct >= threshold, pct
+    if kind == "pct_below":
+        return pct <= -abs(threshold), pct
+    return False, price
+
+
+def check_alerts():
+    with get_session() as s:
+        rules = s.execute(select(AlertRule).where(AlertRule.status == "active")).scalars().all()
+        if not rules:
+            return {"checked": 0, "triggered": 0}
+        codes = list({r.code for r in rules})
+        quotes = svc.realtime_quotes(codes)
+        triggered = 0
+        push_msgs = []
+        for r in rules:
+            q = quotes.get(r.code)
+            if not q:
+                continue
+            hit, val = _hit(r.kind, r.threshold, q)
+            r.last_value = q["price"]
+            if hit:
+                unit = "" if r.kind.startswith("price") else "%"
+                msg = (f"{q['name']} {KIND_LABEL.get(r.kind, r.kind)} {r.threshold}{unit}"
+                       f"（现价 {q['price']}，{q['pct']:+.2f}%）")
+                s.add(AlertEvent(rule_id=r.id, code=r.code, name=q["name"], message=msg, value=val))
+                r.status = "triggered"
+                r.triggered_at = dt.datetime.now()
+                triggered += 1
+                push_msgs.append(msg)
+        s.commit()
+
+    # 触发后向已配置渠道推送（站外）
+    if push_msgs and notifier.any_enabled():
+        try:
+            notifier.notify("【智策预警】" + (push_msgs[0] if len(push_msgs) == 1 else f"{len(push_msgs)} 条预警触发"),
+                            "\n".join(push_msgs))
+        except Exception:
+            pass
+    return {"checked": len(rules), "triggered": triggered}

backend/backtest.py

@@ -0,0 +1,78 @@
+"""基于中台日线的简易回测引擎（均线交叉策略）。
+
+读取 quotes_daily，金叉满仓 / 死叉空仓，输出资金曲线与核心指标。
+"""
+from __future__ import annotations
+
+from sqlalchemy import select
+
+from db import get_session
+from models import DailyQuote
+
+
+def _ma(arr, n):
+    out = [None] * len(arr)
+    s = 0.0
+    for i, v in enumerate(arr):
+        s += v
+        if i >= n:
+            s -= arr[i - n]
+        if i >= n - 1:
+            out[i] = s / n
+    return out
+
+
+def run_backtest(symbol: str, fast: int = 5, slow: int = 20, fee: float = 0.0005):
+    with get_session() as s:
+        rows = s.execute(
+            select(DailyQuote.date, DailyQuote.close)
+            .where(DailyQuote.code == symbol)
+            .order_by(DailyQuote.date)
+        ).all()
+
+    if len(rows) < slow + 5:
+        return {"ok": False, "msg": "该股票库内日线不足，请先在数据中台入库", "have": len(rows)}
+
+    dates = [r[0].isoformat() for r in rows]
+    close = [float(r[1]) for r in rows]
+    maf, mas = _ma(close, fast), _ma(close, slow)
+
+    equity, bench = [], []
+    cash, pos, shares = 1.0, 0, 0.0
+    trades, wins, entry = 0, 0, 0.0
+    peak, max_dd = 1.0, 0.0
+    base = close[0]
+
+    for i in range(len(close)):
+        if maf[i] is not None and mas[i] is not None:
+            if pos == 0 and maf[i] > mas[i]:
+                shares = cash * (1 - fee) / close[i]
+                cash, pos, entry = 0.0, 1, close[i]
+                trades += 1
+            elif pos == 1 and maf[i] < mas[i]:
+                cash = shares * close[i] * (1 - fee)
+                shares, pos = 0.0, 0
+                if close[i] > entry:
+                    wins += 1
+        nav = cash + shares * close[i]
+        equity.append(round(nav, 4))
+        bench.append(round(close[i] / base, 4))
+        peak = max(peak, nav)
+        max_dd = max(max_dd, (peak - nav) / peak)
+
+    total_ret = equity[-1] - 1
+    bench_ret = bench[-1] - 1
+    closed = trades - pos
+    win_rate = (wins / closed) if closed > 0 else 0.0
+    return {
+        "ok": True, "symbol": symbol, "fast": fast, "slow": slow,
+        "dates": dates, "equity": equity, "bench": bench,
+        "metrics": {
+            "total_return": round(total_ret * 100, 2),
+            "bench_return": round(bench_ret * 100, 2),
+            "excess": round((total_ret - bench_ret) * 100, 2),
+            "max_drawdown": round(max_dd * 100, 2),
+            "trades": trades,
+            "win_rate": round(win_rate * 100, 1),
+        },
+    }

backend/cli.py

@@ -0,0 +1,33 @@
+"""命令行入库工具。
+
+用法：
+  python cli.py init                 # 仅建库建表
+  python cli.py ingest               # 全量入库（默认股票池）
+  python cli.py ingest 600519 000001 # 指定股票入库（含快照）
+"""
+import sys
+
+from db import init_db
+import ingest
+
+
+def main():
+    init_db()
+    args = sys.argv[1:]
+    if not args or args[0] == "init":
+        print("init done")
+        return
+    if args[0] == "ingest":
+        codes = args[1:] or None
+        res = ingest.run_daily_ingest(universe=codes)
+        print(res)
+    elif args[0] == "ingest_all":
+        days = int(args[1]) if len(args) > 1 else 250
+        # 先抓快照类数据，再全市场日线
+        ingest.run_daily_ingest(with_quotes=False)
+        res = ingest.ingest_quotes_all(days=days)
+        print(res)
+
+
+if __name__ == "__main__":
+    main()

backend/config.py

@@ -0,0 +1,53 @@
+"""中台配置。优先读环境变量，便于以后切换部署环境。
+
+可在 backend/.env 写入密钥（大模型 key、邮箱、推送 token），无需改代码。
+"""
+import os
+
+try:
+    from dotenv import load_dotenv
+    load_dotenv(os.path.join(os.path.dirname(os.path.abspath(__file__)), ".env"))
+except Exception:
+    pass
+
+PG_USER = os.getenv("PG_USER", "postgres")
+PG_PASSWORD = os.getenv("PG_PASSWORD", "13142324")
+PG_HOST = os.getenv("PG_HOST", "localhost")
+PG_PORT = os.getenv("PG_PORT", "5432")
+PG_DB = os.getenv("PG_DB", "stock_cs")
+
+DB_URL = f"postgresql+psycopg2://{PG_USER}:{PG_PASSWORD}@{PG_HOST}:{PG_PORT}/{PG_DB}"
+DB_URL_DEFAULT = f"postgresql+psycopg2://{PG_USER}:{PG_PASSWORD}@{PG_HOST}:{PG_PORT}/postgres"
+
+# 默认入库股票池（沪深龙头 + 自选）。可在「数据中台」页或环境变量扩展。
+DEFAULT_UNIVERSE = [
+    "600519", "300750", "002594", "688981", "300059", "601012",
+    "600036", "601318", "000858", "002415", "600276", "002230",
+    "601899", "600030", "000333", "002475", "300760", "601166",
+    "688111", "600887",
+]
+
+# 收盘后定时任务时间（24h制，本地时区）
+INGEST_HOUR = int(os.getenv("INGEST_HOUR", "15"))
+INGEST_MINUTE = int(os.getenv("INGEST_MINUTE", "35"))
+
+# ---- 大模型（OpenAI 兼容，DeepSeek/通义/Kimi 等均可）----
+# 配置环境变量 LLM_API_KEY 后即启用真实大模型，否则走内置规则点评。
+LLM_API_KEY = os.getenv("LLM_API_KEY", "")
+LLM_BASE_URL = os.getenv("LLM_BASE_URL", "https://api.deepseek.com/v1")
+LLM_MODEL = os.getenv("LLM_MODEL", "deepseek-chat")
+LLM_TIMEOUT = int(os.getenv("LLM_TIMEOUT", "40"))
+
+# ---- 推送通知 ----
+# 邮件（SMTP）
+SMTP_HOST = os.getenv("SMTP_HOST", "")
+SMTP_PORT = int(os.getenv("SMTP_PORT", "465"))
+SMTP_USER = os.getenv("SMTP_USER", "")
+SMTP_PASSWORD = os.getenv("SMTP_PASSWORD", "")   # 邮箱授权码
+SMTP_TO = os.getenv("SMTP_TO", "")               # 收件人，多个用逗号
+# Server酱 Turbo（微信推送，最简单）
+SERVERCHAN_KEY = os.getenv("SERVERCHAN_KEY", "")
+# 企业微信群机器人 webhook
+WECOM_WEBHOOK = os.getenv("WECOM_WEBHOOK", "")
+# PushPlus（微信推送）
+PUSHPLUS_TOKEN = os.getenv("PUSHPLUS_TOKEN", "")

backend/db.py

@@ -0,0 +1,43 @@
+"""数据库引擎与初始化。
+
+init_db():
+- 若目标库不存在则自动创建（连到默认 postgres 库执行 CREATE DATABASE）
+- 建表
+"""
+from __future__ import annotations
+
+import psycopg2
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+import config
+from models import Base
+
+engine = create_engine(config.DB_URL, pool_pre_ping=True, future=True)
+SessionLocal = sessionmaker(bind=engine, autoflush=False, expire_on_commit=False, future=True)
+
+
+def _ensure_database():
+    try:
+        conn = psycopg2.connect(
+            host=config.PG_HOST, port=config.PG_PORT, user=config.PG_USER,
+            password=config.PG_PASSWORD, dbname="postgres", connect_timeout=5)
+        conn.autocommit = True
+        cur = conn.cursor()
+        cur.execute("SELECT 1 FROM pg_database WHERE datname = %s", (config.PG_DB,))
+        if not cur.fetchone():
+            cur.execute(f'CREATE DATABASE "{config.PG_DB}"')
+        cur.close()
+        conn.close()
+    except Exception as e:  # 数据库已存在或无权限时不阻断
+        print("[db] ensure_database:", repr(e)[:120])
+
+
+def init_db():
+    _ensure_database()
+    Base.metadata.create_all(engine)
+    print("[db] tables ready")
+
+
+def get_session():
+    return SessionLocal()

backend/ingest.py

@@ -0,0 +1,298 @@
+"""ETL：从数据源抽取并增量入库。
+
+每个 ingest_* 负责一类数据；run_daily_ingest 编排全部并写任务日志。
+"""
+from __future__ import annotations
+
+import datetime as dt
+
+from sqlalchemy.dialects.postgresql import insert as pg_insert
+
+import akshare_service as svc
+import config
+from db import get_session
+from models import (DailyQuote, DragonTiger, FundFlowDaily, IndexDaily, JobRun,
+                    SectorDaily, Security, SentimentDaily, StockMetric)
+
+try:
+    import akshare as ak
+except Exception:
+    ak = None
+
+
+def _today():
+    return dt.date.today()
+
+
+def _upsert(session, model, rows, index_elements, update_cols):
+    if not rows:
+        return 0
+    stmt = pg_insert(model).values(rows)
+    stmt = stmt.on_conflict_do_update(
+        index_elements=index_elements,
+        set_={c: getattr(stmt.excluded, c) for c in update_cols},
+    )
+    session.execute(stmt)
+    return len(rows)
+
+
+# ---------------- 个股日线（保留真实日期） ----------------
+def fetch_daily(code: str, days: int = 400):
+    if ak is None:
+        return []
+    sym = svc._sina_symbol(code)
+    for src in ("sina", "tx"):
+        try:
+            df = ak.stock_zh_a_daily(symbol=sym, adjust="qfq") if src == "sina" else ak.stock_zh_a_hist_tx(symbol=sym)
+            if df is None or df.empty:
+                continue
+            df = df.tail(days)
+            out = []
+            for _, r in df.iterrows():
+                d = r["date"]
+                d = d.date() if hasattr(d, "date") else dt.date.fromisoformat(str(d)[:10])
+                out.append({
+                    "code": code, "date": d,
+                    "open": float(r["open"]), "high": float(r["high"]),
+                    "low": float(r["low"]), "close": float(r["close"]),
+                    "volume": int(r["volume"]) if "volume" in df.columns else 0,
+                    "amount": float(r["amount"]) if "amount" in df.columns else 0.0,
+                })
+            return out
+        except Exception:
+            continue
+    return []
+
+
+# ---------------- 因子计算 ----------------
+def _mean(a):
+    return sum(a) / len(a) if a else 0.0
+
+
+def _ema(arr, n):
+    k = 2 / (n + 1)
+    e = arr[0]
+    out = []
+    for v in arr:
+        e = v * k + e * (1 - k)
+        out.append(e)
+    return out
+
+
+def compute_metrics(code, name, rows):
+    """rows: 按日期升序的日线 dict 列表。返回 StockMetric 字段 dict 或 None。"""
+    if len(rows) < 25:
+        return None
+    closes = [r["close"] for r in rows]
+    highs = [r["high"] for r in rows]
+    lows = [r["low"] for r in rows]
+    vols = [r["volume"] for r in rows]
+    n = len(closes)
+    last = closes[-1]
+    prev = closes[-2]
+
+    def ma(k):
+        return round(_mean(closes[-k:]), 3) if n >= k else 0.0
+
+    def ret(k):
+        return round((last / closes[-k - 1] - 1) * 100, 2) if n > k else 0.0
+
+    ma5, ma10, ma20, ma60 = ma(5), ma(10), ma(20), ma(60)
+    vol_ratio = round(vols[-1] / _mean(vols[-6:-1]), 2) if n >= 6 and _mean(vols[-6:-1]) else 0.0
+    win = min(60, n)
+    high60, low60 = max(highs[-win:]), min(lows[-win:])
+    pos60 = round((last - low60) / (high60 - low60), 3) if high60 > low60 else 0.5
+
+    # RSI14
+    gains, losses = [], []
+    for i in range(max(1, n - 14), n):
+        d = closes[i] - closes[i - 1]
+        gains.append(max(d, 0)); losses.append(max(-d, 0))
+    ag, al = _mean(gains), _mean(losses)
+    rsi14 = round(100 - 100 / (1 + ag / al), 1) if al else (100.0 if ag else 50.0)
+
+    # MACD 金叉
+    e12, e26 = _ema(closes, 12), _ema(closes, 26)
+    dif = [a - b for a, b in zip(e12, e26)]
+    dea = _ema(dif, 9)
+    macd_gold = len(dif) >= 2 and dif[-2] < dea[-2] and dif[-1] >= dea[-1]
+
+    ma_bull = ma5 > ma10 > ma20 > 0
+
+    streak = 0
+    for i in range(n - 1, 0, -1):
+        if closes[i] > closes[i - 1]:
+            streak += 1
+        else:
+            break
+
+    return {
+        "code": code, "name": name, "date": rows[-1]["date"], "close": last,
+        "pct": round((last / prev - 1) * 100, 2) if prev else 0.0,
+        "ma5": ma5, "ma10": ma10, "ma20": ma20, "ma60": ma60,
+        "vol_ratio": vol_ratio, "ret5": ret(5), "ret20": ret(20), "ret60": ret(60),
+        "pos60": pos60, "rsi14": rsi14, "macd_gold": bool(macd_gold),
+        "ma_bull": bool(ma_bull), "up_streak": streak,
+        "amount": round(rows[-1]["amount"] / 1e8, 3),
+    }
+
+
+def ingest_quotes(codes, days=400, with_metrics=True, cmap=None):
+    if with_metrics and cmap is None:
+        cmap = svc._code_name_map()
+    n = 0
+    with get_session() as s:
+        for code in codes:
+            rows = fetch_daily(code, days)
+            if not rows:
+                continue
+            n += _upsert(s, DailyQuote, rows, ["code", "date"],
+                         ["open", "high", "low", "close", "volume", "amount"])
+            if with_metrics:
+                m = compute_metrics(code, (cmap or {}).get(code, code), rows)
+                if m:
+                    _upsert(s, StockMetric, [m], ["code"],
+                            ["name", "date", "close", "pct", "ma5", "ma10", "ma20", "ma60",
+                             "vol_ratio", "ret5", "ret20", "ret60", "pos60", "rsi14",
+                             "macd_gold", "ma_bull", "up_streak", "amount"])
+            s.commit()
+    return n
+
+
+def ingest_quotes_all(days=250, progress_every=300):
+    """全市场回填：对所有证券抓取日线并计算因子。耗时较长。"""
+    cmap = svc._code_name_map()
+    codes = [c for c in cmap.keys() if c[:1] in ("0", "3", "6")]
+    total = len(codes)
+    done = 0
+    with get_session() as s:
+        job = JobRun(job="ingest_all", status="running", message=f"0/{total}")
+        s.add(job); s.commit(); job_id = job.id
+    try:
+        for i in range(0, total, 50):
+            batch = codes[i:i + 50]
+            ingest_quotes(batch, days=days, with_metrics=True, cmap=cmap)
+            done += len(batch)
+            if done % progress_every < 50:
+                with get_session() as s:
+                    j = s.get(JobRun, job_id); j.message = f"{done}/{total}"; s.commit()
+        status, msg = "success", f"{done}/{total}"
+    except Exception as e:
+        status, msg = "error", f"{done}/{total} | {repr(e)[:160]}"
+    with get_session() as s:
+        j = s.get(JobRun, job_id); j.status = status
+        j.finished_at = dt.datetime.now(); j.message = msg; s.commit()
+    return {"status": status, "done": done, "total": total}
+
+
+def ingest_securities():
+    cmap = svc._code_name_map()
+    rows = [{"code": c, "name": n, "market": "A"} for c, n in cmap.items()]
+    with get_session() as s:
+        cnt = _upsert(s, Security, rows, ["code"], ["name", "market"])
+        s.commit()
+    return cnt
+
+
+def ingest_indices():
+    if ak is None:
+        return 0
+    n = 0
+    with get_session() as s:
+        for code, (name, _b) in svc.MAJOR_INDEX.items():
+            try:
+                df = ak.stock_zh_index_daily(symbol=code).tail(400)
+                rows = []
+                for _, r in df.iterrows():
+                    d = r["date"]
+                    d = d.date() if hasattr(d, "date") else dt.date.fromisoformat(str(d)[:10])
+                    rows.append({"code": code, "name": name, "date": d,
+                                 "open": float(r["open"]), "high": float(r["high"]),
+                                 "low": float(r["low"]), "close": float(r["close"]),
+                                 "volume": int(r.get("volume", 0) or 0)})
+                n += _upsert(s, IndexDaily, rows, ["code", "date"],
+                             ["name", "open", "high", "low", "close", "volume"])
+                s.commit()
+            except Exception:
+                continue
+    return n
+
+
+def ingest_sectors():
+    data = svc.get_industry_boards()
+    d = _today()
+    rows = [{"date": d, "name": b["name"], "pct": b["pct"], "amount": b.get("amount", 0),
+             "count": b.get("count", 0), "leader": b.get("leader", "")} for b in data["list"]]
+    with get_session() as s:
+        n = _upsert(s, SectorDaily, rows, ["date", "name"], ["pct", "amount", "count", "leader"])
+        s.commit()
+    return n
+
+
+def ingest_fund_flow():
+    data = svc.get_fund_flow()
+    d = _today()
+    rows = [{"date": d, "name": x["name"], "net": x["net"], "pct": x.get("pct", 0)} for x in data["list"]]
+    with get_session() as s:
+        n = _upsert(s, FundFlowDaily, rows, ["date", "name"], ["net", "pct"])
+        s.commit()
+    return n
+
+
+def ingest_sentiment():
+    x = svc.get_sentiment()
+    d = _today()
+    row = [{"date": d, "up": x["up"], "down": x["down"], "flat": x["flat"],
+            "limit_up": x["limit_up"], "limit_down": x["limit_down"]}]
+    with get_session() as s:
+        n = _upsert(s, SentimentDaily, row, ["date"], ["up", "down", "flat", "limit_up", "limit_down"])
+        s.commit()
+    return n
+
+
+def ingest_dragon():
+    data = svc.get_dragon_tiger()
+    if not data["list"]:
+        return 0
+    try:
+        d = dt.date.fromisoformat(f"{data['date'][:4]}-{data['date'][4:6]}-{data['date'][6:8]}") if data.get("date") else _today()
+    except Exception:
+        d = _today()
+    rows = [{"date": d, "code": x["code"], "name": x["name"], "pct": x["pct"],
+             "net": x["net"], "reason": x["reason"][:120]} for x in data["list"]]
+    with get_session() as s:
+        n = _upsert(s, DragonTiger, rows, ["date", "code", "reason"], ["name", "pct", "net"])
+        s.commit()
+    return n
+
+
+# ---------------- 编排 ----------------
+def run_daily_ingest(universe=None, with_quotes=True):
+    universe = universe or config.DEFAULT_UNIVERSE
+    with get_session() as s:
+        job = JobRun(job="daily_ingest", status="running", message="")
+        s.add(job)
+        s.commit()
+        job_id = job.id
+
+    summary = {}
+    try:
+        summary["securities"] = ingest_securities()
+        summary["indices"] = ingest_indices()
+        summary["sectors"] = ingest_sectors()
+        summary["fund_flow"] = ingest_fund_flow()
+        summary["sentiment"] = ingest_sentiment()
+        summary["dragon"] = ingest_dragon()
+        if with_quotes:
+            summary["quotes"] = ingest_quotes(universe)
+        status, msg = "success", str(summary)
+    except Exception as e:
+        status, msg = "error", f"{summary} | EXC {repr(e)[:200]}"
+
+    with get_session() as s:
+        job = s.get(JobRun, job_id)
+        job.status = status
+        job.finished_at = dt.datetime.now()
+        job.message = msg
+        s.commit()
+    return {"status": status, "summary": summary}

backend/llm.py

@@ -0,0 +1,42 @@
+"""大模型客户端（OpenAI 兼容接口）。
+
+未配置 LLM_API_KEY 时 enabled() 返回 False，调用方应走规则降级。
+"""
+from __future__ import annotations
+
+import requests
+
+import config
+
+
+def enabled() -> bool:
+    return bool(config.LLM_API_KEY)
+
+
+def chat(messages, temperature: float = 0.5, max_tokens: int = 900) -> str:
+    """调用 chat/completions。失败抛异常，由上层捕获降级。"""
+    if not enabled():
+        raise RuntimeError("LLM 未配置")
+    url = config.LLM_BASE_URL.rstrip("/") + "/chat/completions"
+    headers = {"Authorization": f"Bearer {config.LLM_API_KEY}", "Content-Type": "application/json"}
+    payload = {"model": config.LLM_MODEL, "messages": messages,
+               "temperature": temperature, "max_tokens": max_tokens, "stream": False}
+    r = requests.post(url, json=payload, headers=headers, timeout=config.LLM_TIMEOUT)
+    r.raise_for_status()
+    data = r.json()
+    return data["choices"][0]["message"]["content"].strip()
+
+
+SYSTEM_PROMPT = (
+    "你是一名专业的A股市场分析师，擅长复盘、个股诊断与题材分析。"
+    "你的结论必须严格基于用户提供的结构化数据，不臆造未提供的数字。"
+    "语言简洁专业，使用中文。涉及操作建议时必须同时给出风险提示，"
+    "并说明这不构成投资建议。"
+)
+
+
+def ask(user_content: str, temperature: float = 0.5, max_tokens: int = 900) -> str:
+    return chat([
+        {"role": "system", "content": SYSTEM_PROMPT},
+        {"role": "user", "content": user_content},
+    ], temperature=temperature, max_tokens=max_tokens)

backend/main.py

@@ -0,0 +1,624 @@
+"""智策股票终端 — FastAPI 后端入口。
+
+- /api/*  : 数据接口（基于 AkShare，带缓存与降级）
+- /       : 托管前端原型（prototype 目录）
+"""
+import os
+import json
+import datetime as dt
+from contextlib import asynccontextmanager
+
+from fastapi import FastAPI, Query
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.staticfiles import StaticFiles
+from sqlalchemy import select, func, desc
+
+from pydantic import BaseModel
+
+import akshare_service as svc
+import config
+import scheduler
+import backtest as bt
+import ai
+import signals as sig
+import report as rpt
+import portfolio as pf
+import llm
+import alerts as al
+import notifier
+from db import init_db, get_session
+from models import (DailyQuote, IndexDaily, SectorDaily, FundFlowDaily,
+                    SentimentDaily, DragonTiger, Security, JobRun, StockMetric, Trade,
+                    AlertRule, AlertEvent)
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    try:
+        init_db()
+        scheduler.start_scheduler()
+        print("[startup] db + scheduler ready")
+    except Exception as e:
+        print("[startup] WARN:", repr(e)[:160])
+    yield
+
+
+app = FastAPI(title="智策股票终端 API", version="0.2.0", lifespan=lifespan)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# 自选股本地存储
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+WATCH_FILE = os.path.join(BASE_DIR, "watchlist.json")
+DEFAULT_WATCH = ["600519", "300750", "002594", "688981", "300059", "601012"]
+
+
+def load_watch():
+    if os.path.exists(WATCH_FILE):
+        try:
+            with open(WATCH_FILE, "r", encoding="utf-8") as f:
+                return json.load(f)
+        except Exception:
+            pass
+    return DEFAULT_WATCH
+
+
+def save_watch(symbols):
+    with open(WATCH_FILE, "w", encoding="utf-8") as f:
+        json.dump(symbols, f, ensure_ascii=False)
+
+
+# ============ API ============
+@app.get("/api/health")
+def health():
+    return {"ok": True, "akshare": svc.AK_OK}
+
+
+@app.get("/api/indices")
+def indices():
+    return svc.get_indices()
+
+
+@app.get("/api/kline")
+def kline(symbol: str = Query("600519"), days: int = Query(120, ge=20, le=500)):
+    return svc.get_kline(symbol, days)
+
+
+@app.get("/api/sentiment")
+def sentiment():
+    return svc.get_sentiment()
+
+
+@app.get("/api/treemap")
+def treemap(mode: str = Query("sector")):
+    return svc.get_treemap(mode)
+
+
+@app.get("/api/fundflow")
+def fundflow():
+    return svc.get_fund_flow()
+
+
+@app.get("/api/hot/stocks")
+def hot_stocks():
+    return svc.get_hot_stocks()
+
+
+@app.get("/api/hot/sectors")
+def hot_sectors():
+    return svc.get_industry_boards()
+
+
+@app.get("/api/dragon")
+def dragon():
+    return svc.get_dragon_tiger()
+
+
+@app.get("/api/watchlist")
+def watchlist():
+    return svc.get_watchlist(load_watch())
+
+
+@app.post("/api/watchlist/{code}")
+def watch_add(code: str):
+    w = load_watch()
+    if code not in w:
+        w.append(code)
+        save_watch(w)
+    return {"ok": True, "list": w}
+
+
+@app.delete("/api/watchlist/{code}")
+def watch_del(code: str):
+    w = [c for c in load_watch() if c != code]
+    save_watch(w)
+    return {"ok": True, "list": w}
+
+
+# ============ 数据中台 ============
+@app.get("/api/admin/status")
+def admin_status():
+    counts, last_dates = {}, {}
+    with get_session() as s:
+        for label, model in [("securities", Security), ("quotes_daily", DailyQuote),
+                             ("index_daily", IndexDaily), ("sector_daily", SectorDaily),
+                             ("fund_flow_daily", FundFlowDaily), ("sentiment_daily", SentimentDaily),
+                             ("dragon_tiger", DragonTiger)]:
+            counts[label] = s.execute(select(func.count()).select_from(model)).scalar() or 0
+            if hasattr(model, "date"):
+                d = s.execute(select(func.max(model.date))).scalar()
+                last_dates[label] = d.isoformat() if d else None
+        jobs = s.execute(select(JobRun).order_by(desc(JobRun.id)).limit(8)).scalars().all()
+        job_list = [{"id": j.id, "job": j.job, "status": j.status,
+                     "started": j.started_at.strftime("%m-%d %H:%M:%S") if j.started_at else "",
+                     "finished": j.finished_at.strftime("%H:%M:%S") if j.finished_at else "",
+                     "message": j.message[:200]} for j in jobs]
+    return {"counts": counts, "last_dates": last_dates, "jobs": job_list,
+            "running": scheduler.is_running(), "universe": config.DEFAULT_UNIVERSE,
+            "schedule": f"周一至周五 {config.INGEST_HOUR:02d}:{config.INGEST_MINUTE:02d}"}
+
+
+@app.post("/api/admin/ingest")
+def admin_ingest():
+    if scheduler.is_running():
+        return {"started": False, "msg": "已有入库任务在执行"}
+    return scheduler.trigger_async()
+
+
+@app.post("/api/admin/ingest_all")
+def admin_ingest_all():
+    return scheduler.trigger_all_async()
+
+
+@app.get("/api/db/kline")
+def db_kline(symbol: str = Query("600519"), days: int = Query(250, ge=20, le=1000)):
+    with get_session() as s:
+        rows = s.execute(
+            select(DailyQuote).where(DailyQuote.code == symbol)
+            .order_by(DailyQuote.date.desc()).limit(days)
+        ).scalars().all()
+    rows = list(reversed(rows))
+    if not rows:
+        return {"source": "db", "empty": True, "symbol": symbol, "dates": [], "ohlc": [], "vols": []}
+    return {"source": "db", "symbol": symbol,
+            "dates": [r.date.strftime("%m/%d") for r in rows],
+            "ohlc": [[r.open, r.close, r.low, r.high] for r in rows],
+            "vols": [r.volume for r in rows]}
+
+
+@app.get("/api/db/sentiment_history")
+def db_sentiment_history(days: int = Query(60, ge=5, le=365)):
+    with get_session() as s:
+        rows = s.execute(select(SentimentDaily).order_by(SentimentDaily.date.desc()).limit(days)).scalars().all()
+    rows = list(reversed(rows))
+    return {"dates": [r.date.isoformat() for r in rows],
+            "up": [r.up for r in rows], "down": [r.down for r in rows],
+            "limit_up": [r.limit_up for r in rows]}
+
+
+@app.get("/api/review/daily")
+def review_daily(date: str = Query(None)):
+    with get_session() as s:
+        if date:
+            d = dt.date.fromisoformat(date)
+        else:
+            d = s.execute(select(func.max(SectorDaily.date))).scalar()
+        if not d:
+            return {"ok": False, "msg": "暂无入库数据，请先在数据中台执行入库"}
+        sectors = s.execute(select(SectorDaily).where(SectorDaily.date == d).order_by(SectorDaily.pct.desc())).scalars().all()
+        flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d).order_by(FundFlowDaily.net.desc())).scalars().all()
+        senti = s.execute(select(SentimentDaily).where(SentimentDaily.date == d)).scalar_one_or_none()
+        lhb = s.execute(select(DragonTiger).where(DragonTiger.date == d).order_by(DragonTiger.net.desc()).limit(10)).scalars().all()
+
+    top_sec = [{"name": x.name, "pct": x.pct} for x in sectors[:8]]
+    bot_sec = [{"name": x.name, "pct": x.pct} for x in sectors[-5:]]
+    inflow = [{"name": x.name, "net": x.net} for x in flows[:8]]
+    outflow = [{"name": x.name, "net": x.net} for x in flows[-5:][::-1]]
+    senti_d = ({"up": senti.up, "down": senti.down, "limit_up": senti.limit_up,
+                "limit_down": senti.limit_down} if senti else None)
+
+    summary = _gen_review_text(d, senti_d, top_sec, inflow)
+    return {"ok": True, "date": d.isoformat(), "sentiment": senti_d,
+            "top_sectors": top_sec, "weak_sectors": bot_sec,
+            "inflow": inflow, "outflow": outflow,
+            "dragon": [{"name": x.name, "code": x.code, "net": x.net, "pct": x.pct} for x in lhb],
+            "summary": summary}
+
+
+def _gen_review_text(d, senti, top_sec, inflow):
+    parts = [f"【{d.isoformat()} 复盘】"]
+    if senti:
+        tone = "情绪偏暖" if senti["up"] > senti["down"] else "情绪偏弱"
+        parts.append(f"全市场上涨 {senti['up']} 家、下跌 {senti['down']} 家，涨停 {senti['limit_up']} 家、跌停 {senti['limit_down']} 家，{tone}。")
+    if top_sec:
+        names = "、".join(x["name"] for x in top_sec[:3])
+        parts.append(f"领涨板块：{names}。")
+    if inflow:
+        names = "、".join(x["name"] for x in inflow[:3])
+        parts.append(f"主力净流入居前：{names}。")
+    parts.append("注：以上为基于入库数据的自动统计，AI 智能点评将在 AI 分析模块接入大模型后生成。")
+    return " ".join(parts)
+
+
+@app.get("/api/backtest")
+def backtest_api(symbol: str = Query("600519"), fast: int = Query(5, ge=2, le=60),
+                 slow: int = Query(20, ge=5, le=250)):
+    if fast >= slow:
+        return {"ok": False, "msg": "快线周期需小于慢线周期"}
+    return bt.run_backtest(symbol, fast, slow)
+
+
+# ============ 全市场选股 ============
+STRATEGIES = {
+    "surge": "最近暴涨（5日涨幅≥20%）",
+    "plunge": "最近暴跌（5日跌幅≥15%）",
+    "dip": "超跌抄底（60日分位≤20%且当日企稳）",
+    "breakout": "突破走强（逼近60日新高）",
+    "ma_bull": "均线多头（MA5>10>20）",
+    "volume": "放量上攻（量比≥2且上涨）",
+    "macd_gold": "MACD金叉",
+    "strong": "强势连涨（≥3日连阳）",
+}
+
+
+@app.get("/api/screen/strategies")
+def screen_strategies():
+    return {"list": [{"id": k, "name": v} for k, v in STRATEGIES.items()]}
+
+
+@app.get("/api/screen")
+def screen(strategy: str = Query("surge"), limit: int = Query(60, ge=10, le=300),
+           min_amount: float = Query(0.0)):
+    M = StockMetric
+    q = select(M)
+    order = M.ret5.desc()
+    if strategy == "surge":
+        q = q.where(M.ret5 >= 20)
+    elif strategy == "plunge":
+        q = q.where(M.ret5 <= -15); order = M.ret5.asc()
+    elif strategy == "dip":
+        q = q.where(M.pos60 <= 0.2, M.pct > 0); order = M.pos60.asc()
+    elif strategy == "breakout":
+        q = q.where(M.pos60 >= 0.95, M.pct > 0); order = M.ret20.desc()
+    elif strategy == "ma_bull":
+        q = q.where(M.ma_bull.is_(True)); order = M.ret20.desc()
+    elif strategy == "volume":
+        q = q.where(M.vol_ratio >= 2, M.pct > 0); order = M.vol_ratio.desc()
+    elif strategy == "macd_gold":
+        q = q.where(M.macd_gold.is_(True)); order = M.ret5.desc()
+    elif strategy == "strong":
+        q = q.where(M.up_streak >= 3); order = M.up_streak.desc()
+    if min_amount > 0:
+        q = q.where(M.amount >= min_amount)
+    q = q.order_by(order).limit(limit)
+    with get_session() as s:
+        rows = s.execute(q).scalars().all()
+        total = s.execute(select(func.count()).select_from(M)).scalar() or 0
+    return {"strategy": strategy, "name": STRATEGIES.get(strategy, strategy), "pool_size": total,
+            "count": len(rows), "list": [{
+                "code": r.code, "name": r.name, "close": r.close, "pct": r.pct,
+                "ret5": r.ret5, "ret20": r.ret20, "vol_ratio": r.vol_ratio,
+                "rsi14": r.rsi14, "pos60": round(r.pos60 * 100, 1), "amount": r.amount,
+                "up_streak": r.up_streak} for r in rows]}
+
+
+@app.get("/api/securities/search")
+def securities_search(q: str = Query("", min_length=0), limit: int = Query(15, le=50)):
+    with get_session() as s:
+        stmt = select(Security)
+        if q:
+            stmt = stmt.where((Security.code.like(f"{q}%")) | (Security.name.like(f"%{q}%")))
+        rows = s.execute(stmt.limit(limit)).scalars().all()
+    return {"list": [{"code": r.code, "name": r.name} for r in rows]}
+
+
+# ============ 个股复盘（K线 + 买卖点 + 回放） ============
+def _ma_list(close, n):
+    out = [None] * len(close)
+    for i in range(len(close)):
+        if i >= n - 1:
+            out[i] = round(sum(close[i - n + 1:i + 1]) / n, 3)
+    return out
+
+
+@app.get("/api/review/stock")
+def review_stock(symbol: str = Query("600519"), days: int = Query(250, ge=40, le=1000),
+                 fast: int = Query(5), slow: int = Query(20)):
+    with get_session() as s:
+        rows = s.execute(
+            select(DailyQuote).where(DailyQuote.code == symbol)
+            .order_by(DailyQuote.date.desc()).limit(days)
+        ).scalars().all()
+        sec = s.get(Security, symbol)
+    rows = list(reversed(rows))
+    if not rows:
+        return {"ok": False, "msg": "该股票库内无日线，请先在数据中台入库该股或执行全市场回填", "symbol": symbol}
+    dates = [r.date.strftime("%y/%m/%d") for r in rows]
+    ohlc = [[r.open, r.close, r.low, r.high] for r in rows]
+    vols = [r.volume for r in rows]
+    close = [r.close for r in rows]
+    maf, mas = _ma_list(close, fast), _ma_list(close, slow)
+
+    signals = []
+    for i in range(1, len(close)):
+        if maf[i] is None or mas[i] is None or maf[i - 1] is None or mas[i - 1] is None:
+            continue
+        if maf[i - 1] <= mas[i - 1] and maf[i] > mas[i]:
+            signals.append({"idx": i, "date": dates[i], "price": close[i], "type": "buy"})
+        elif maf[i - 1] >= mas[i - 1] and maf[i] < mas[i]:
+            signals.append({"idx": i, "date": dates[i], "price": close[i], "type": "sell"})
+
+    # 区间统计
+    hi = max(r.high for r in rows); lo = min(r.low for r in rows)
+    period_ret = round((close[-1] / close[0] - 1) * 100, 2)
+    return {"ok": True, "symbol": symbol, "name": sec.name if sec else symbol,
+            "dates": dates, "ohlc": ohlc, "vols": vols,
+            "ma_fast": maf, "ma_slow": mas, "fast": fast, "slow": slow,
+            "signals": signals,
+            "stats": {"period_return": period_ret, "high": hi, "low": lo,
+                      "start": dates[0], "end": dates[-1], "bars": len(rows)}}
+
+
+# ============ AI 分析 ============
+@app.get("/api/ai/status")
+def ai_status():
+    return {"enabled": llm.enabled(), "model": config.LLM_MODEL if llm.enabled() else None}
+
+
+@app.get("/api/ai/review_daily")
+def ai_review_daily(date: str = Query(None)):
+    return ai.review_daily_comment(date)
+
+
+@app.get("/api/ai/diagnose")
+def ai_diagnose(symbol: str = Query("600519")):
+    return ai.diagnose(symbol)
+
+
+@app.get("/api/ai/today")
+def ai_today():
+    return ai.today_strategy()
+
+
+# ============ 可回溯：信号历史胜率 + 实测准确率 ============
+@app.get("/api/ai/signal_stats")
+def ai_signal_stats(horizon: int = Query(5, ge=1, le=20)):
+    return {"horizon": horizon, "stats": sig.get_stats(horizon)}
+
+
+@app.post("/api/ai/signal_stats/compute")
+def ai_signal_stats_compute(sample: int = Query(500, ge=50, le=4000), horizon: int = Query(5, ge=1, le=20)):
+    return scheduler.trigger_signal_stats_async(sample, horizon)
+
+
+@app.get("/api/ai/accuracy")
+def ai_accuracy():
+    return sig.accuracy()
+
+
+@app.post("/api/ai/accuracy/verify")
+def ai_accuracy_verify():
+    return sig.verify_predictions()
+
+
+# ============ AI 自动复盘日报 ============
+@app.get("/api/report/daily")
+def report_daily(date: str = Query(None)):
+    return rpt.get_by_date(date) if date else rpt.latest()
+
+
+@app.get("/api/report/history")
+def report_history(limit: int = Query(30, ge=1, le=120)):
+    return rpt.history(limit)
+
+
+@app.post("/api/report/generate")
+def report_generate(date: str = Query(None), push: bool = Query(False)):
+    return rpt.generate(date, push=push)
+
+
+# ============ 交易日志 & 组合 ============
+class TradeIn(BaseModel):
+    code: str
+    name: str = ""
+    side: str = "buy"
+    price: float
+    qty: int
+    fee: float = 0.0
+    date: str = ""
+    reason: str = ""
+    emotion: str = ""
+
+
+@app.get("/api/trades")
+def list_trades():
+    with get_session() as s:
+        rows = s.execute(select(Trade).order_by(Trade.date.desc(), Trade.id.desc())).scalars().all()
+        names = {}
+    return {"list": [{"id": t.id, "date": t.date.isoformat(), "code": t.code, "name": t.name,
+                      "side": t.side, "price": t.price, "qty": t.qty, "fee": t.fee,
+                      "reason": t.reason, "emotion": t.emotion} for t in rows]}
+
+
+@app.post("/api/trades")
+def add_trade(t: TradeIn):
+    d = dt.date.fromisoformat(t.date) if t.date else dt.date.today()
+    name = t.name
+    if not name:
+        with get_session() as s:
+            sec = s.get(Security, t.code)
+            name = sec.name if sec else t.code
+    with get_session() as s:
+        row = Trade(date=d, code=t.code, name=name, side=t.side, price=t.price,
+                    qty=t.qty, fee=t.fee, reason=t.reason, emotion=t.emotion)
+        s.add(row); s.commit()
+        return {"ok": True, "id": row.id}
+
+
+@app.delete("/api/trades/{tid}")
+def del_trade(tid: int):
+    with get_session() as s:
+        row = s.get(Trade, tid)
+        if row:
+            s.delete(row); s.commit()
+    return {"ok": True}
+
+
+@app.get("/api/portfolio")
+def get_portfolio():
+    return pf.compute()
+
+
+@app.get("/api/portfolio/equity")
+def portfolio_equity():
+    return pf.equity_curve()
+
+
+# ============ 推送通知 ============
+@app.get("/api/notify/status")
+def notify_status():
+    return {"channels": notifier.channels_status(), "enabled": notifier.any_enabled()}
+
+
+@app.post("/api/notify/test")
+def notify_test():
+    if not notifier.any_enabled():
+        return {"ok": False, "msg": "未配置任何推送渠道，请在 backend/.env 配置后重启"}
+    res = notifier.notify("【智策】推送测试", "这是一条来自智策股票终端的测试通知，收到即表示推送通道正常。")
+    return {"ok": True, "result": res}
+
+
+# ============ 智能预警 ============
+class AlertIn(BaseModel):
+    code: str
+    kind: str = "price_above"
+    threshold: float
+    note: str = ""
+
+
+@app.get("/api/alerts")
+def list_alerts():
+    with get_session() as s:
+        rows = s.execute(select(AlertRule).order_by(AlertRule.id.desc())).scalars().all()
+    return {"list": [{"id": r.id, "code": r.code, "name": r.name, "kind": r.kind,
+                      "threshold": r.threshold, "status": r.status, "note": r.note,
+                      "last_value": r.last_value,
+                      "triggered_at": r.triggered_at.strftime("%m-%d %H:%M") if r.triggered_at else ""}
+                     for r in rows]}
+
+
+@app.post("/api/alerts")
+def add_alert(a: AlertIn):
+    with get_session() as s:
+        sec = s.get(Security, a.code)
+        name = sec.name if sec else a.code
+        row = AlertRule(code=a.code, name=name, kind=a.kind, threshold=a.threshold, note=a.note)
+        s.add(row); s.commit()
+        return {"ok": True, "id": row.id}
+
+
+@app.delete("/api/alerts/{aid}")
+def del_alert(aid: int):
+    with get_session() as s:
+        row = s.get(AlertRule, aid)
+        if row:
+            s.delete(row); s.commit()
+    return {"ok": True}
+
+
+@app.post("/api/alerts/{aid}/reactivate")
+def reactivate_alert(aid: int):
+    with get_session() as s:
+        row = s.get(AlertRule, aid)
+        if row:
+            row.status = "active"; row.triggered_at = None; s.commit()
+    return {"ok": True}
+
+
+@app.post("/api/alerts/check")
+def manual_check():
+    return al.check_alerts()
+
+
+@app.get("/api/alerts/events")
+def alert_events(unread_only: bool = Query(False), limit: int = Query(30, le=100)):
+    with get_session() as s:
+        stmt = select(AlertEvent).order_by(AlertEvent.id.desc())
+        if unread_only:
+            stmt = stmt.where(AlertEvent.read.is_(False))
+        rows = s.execute(stmt.limit(limit)).scalars().all()
+        unread = s.execute(select(func.count()).select_from(AlertEvent).where(AlertEvent.read.is_(False))).scalar() or 0
+    return {"unread": unread, "list": [{"id": e.id, "code": e.code, "name": e.name,
+            "message": e.message, "time": e.created_at.strftime("%m-%d %H:%M:%S") if e.created_at else ""}
+            for e in rows]}
+
+
+@app.post("/api/alerts/events/read")
+def mark_events_read():
+    with get_session() as s:
+        for e in s.execute(select(AlertEvent).where(AlertEvent.read.is_(False))).scalars():
+            e.read = True
+        s.commit()
+    return {"ok": True}
+
+
+# ============ 资讯中心 ============
+@app.get("/api/news")
+def news(limit: int = Query(40, le=100)):
+    return svc.get_news(limit)
+
+
+@app.get("/api/news/stock")
+def news_stock(code: str = Query(...)):
+    return svc.get_stock_news(code)
+
+
+@app.get("/api/news/watch")
+def news_watch():
+    codes = load_watch()[:6]
+    out = []
+    for c in codes:
+        r = svc.get_stock_news(c, limit=4)
+        for x in r["list"]:
+            x["code"] = c
+            out.append(x)
+    out.sort(key=lambda x: x["time"], reverse=True)
+    return {"list": out[:40]}
+
+
+class NewsAI(BaseModel):
+    title: str
+    content: str = ""
+
+
+@app.post("/api/news/ai")
+def news_ai(n: NewsAI):
+    text_in = (n.title + "。" + n.content).strip()
+    senti, kw = svc.judge_sentiment(text_in)
+    if llm.enabled():
+        try:
+            prompt = ("请分析下面这条财经资讯：\n"
+                      "1) 一句话摘要；2) 利好/利空/中性判断及理由；3) 可能受影响的板块或个股方向。120字内。\n\n"
+                      + text_in[:1200])
+            text = llm.ask(prompt, temperature=0.3, max_tokens=400)
+            return {"ok": True, "source": "llm", "sentiment": senti, "text": text}
+        except Exception:
+            pass
+    return {"ok": True, "source": "rule", "sentiment": senti,
+            "text": f"判断：{senti}（关键词：{'、'.join(kw) or '无'}）。摘要：{text_in[:80]}…\n（配置大模型后可获得更深入的关联分析）"}
+
+
+# ============ 静态前端 ============
+FRONTEND_DIR = os.path.join(os.path.dirname(BASE_DIR), "prototype")
+if os.path.isdir(FRONTEND_DIR):
+    app.mount("/", StaticFiles(directory=FRONTEND_DIR, html=True), name="frontend")
+
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run("main:app", host="0.0.0.0", port=8000, reload=False)

backend/models.py

@@ -0,0 +1,223 @@
+"""数据中台 ORM 模型（SQLAlchemy 2.0）。"""
+from __future__ import annotations
+
+import datetime as dt
+
+from sqlalchemy import (BigInteger, Date, DateTime, Float, Integer, String,
+                        Text, UniqueConstraint, func)
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column
+
+
+class Base(DeclarativeBase):
+    pass
+
+
+class Security(Base):
+    """证券基础信息。"""
+    __tablename__ = "securities"
+    code: Mapped[str] = mapped_column(String(12), primary_key=True)
+    name: Mapped[str] = mapped_column(String(40))
+    market: Mapped[str] = mapped_column(String(8), default="A")
+    updated_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class DailyQuote(Base):
+    """个股日线（前复权）。"""
+    __tablename__ = "quotes_daily"
+    __table_args__ = (UniqueConstraint("code", "date", name="uq_quote_code_date"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    code: Mapped[str] = mapped_column(String(12), index=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    open: Mapped[float] = mapped_column(Float)
+    high: Mapped[float] = mapped_column(Float)
+    low: Mapped[float] = mapped_column(Float)
+    close: Mapped[float] = mapped_column(Float)
+    volume: Mapped[int] = mapped_column(BigInteger, default=0)
+    amount: Mapped[float] = mapped_column(Float, default=0.0)
+
+
+class IndexDaily(Base):
+    """指数日线。"""
+    __tablename__ = "index_daily"
+    __table_args__ = (UniqueConstraint("code", "date", name="uq_index_code_date"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    code: Mapped[str] = mapped_column(String(12), index=True)
+    name: Mapped[str] = mapped_column(String(40), default="")
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    open: Mapped[float] = mapped_column(Float)
+    high: Mapped[float] = mapped_column(Float)
+    low: Mapped[float] = mapped_column(Float)
+    close: Mapped[float] = mapped_column(Float)
+    volume: Mapped[int] = mapped_column(BigInteger, default=0)
+
+
+class SectorDaily(Base):
+    """板块每日快照。"""
+    __tablename__ = "sector_daily"
+    __table_args__ = (UniqueConstraint("date", "name", name="uq_sector_date_name"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    name: Mapped[str] = mapped_column(String(40))
+    pct: Mapped[float] = mapped_column(Float, default=0.0)
+    amount: Mapped[float] = mapped_column(Float, default=0.0)
+    count: Mapped[int] = mapped_column(Integer, default=0)
+    leader: Mapped[str] = mapped_column(String(40), default="")
+
+
+class FundFlowDaily(Base):
+    """行业资金流每日快照。"""
+    __tablename__ = "fund_flow_daily"
+    __table_args__ = (UniqueConstraint("date", "name", name="uq_fund_date_name"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    name: Mapped[str] = mapped_column(String(40))
+    net: Mapped[float] = mapped_column(Float, default=0.0)
+    pct: Mapped[float] = mapped_column(Float, default=0.0)
+
+
+class SentimentDaily(Base):
+    """市场情绪每日快照。"""
+    __tablename__ = "sentiment_daily"
+    date: Mapped[dt.date] = mapped_column(Date, primary_key=True)
+    up: Mapped[int] = mapped_column(Integer, default=0)
+    down: Mapped[int] = mapped_column(Integer, default=0)
+    flat: Mapped[int] = mapped_column(Integer, default=0)
+    limit_up: Mapped[int] = mapped_column(Integer, default=0)
+    limit_down: Mapped[int] = mapped_column(Integer, default=0)
+
+
+class DragonTiger(Base):
+    """龙虎榜明细。"""
+    __tablename__ = "dragon_tiger"
+    __table_args__ = (UniqueConstraint("date", "code", "reason", name="uq_lhb"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    code: Mapped[str] = mapped_column(String(12))
+    name: Mapped[str] = mapped_column(String(40), default="")
+    pct: Mapped[float] = mapped_column(Float, default=0.0)
+    net: Mapped[float] = mapped_column(Float, default=0.0)
+    reason: Mapped[str] = mapped_column(String(120), default="")
+
+
+class StockMetric(Base):
+    """个股最新因子快照（供全市场选股快速查询）。"""
+    __tablename__ = "stock_metrics"
+    code: Mapped[str] = mapped_column(String(12), primary_key=True)
+    name: Mapped[str] = mapped_column(String(40), default="")
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    close: Mapped[float] = mapped_column(Float, default=0.0)
+    pct: Mapped[float] = mapped_column(Float, default=0.0, index=True)
+    ma5: Mapped[float] = mapped_column(Float, default=0.0)
+    ma10: Mapped[float] = mapped_column(Float, default=0.0)
+    ma20: Mapped[float] = mapped_column(Float, default=0.0)
+    ma60: Mapped[float] = mapped_column(Float, default=0.0)
+    vol_ratio: Mapped[float] = mapped_column(Float, default=0.0)
+    ret5: Mapped[float] = mapped_column(Float, default=0.0, index=True)
+    ret20: Mapped[float] = mapped_column(Float, default=0.0)
+    ret60: Mapped[float] = mapped_column(Float, default=0.0)
+    pos60: Mapped[float] = mapped_column(Float, default=0.0)  # 0~1，60日价格分位
+    rsi14: Mapped[float] = mapped_column(Float, default=0.0)
+    macd_gold: Mapped[bool] = mapped_column(default=False)
+    ma_bull: Mapped[bool] = mapped_column(default=False)
+    up_streak: Mapped[int] = mapped_column(Integer, default=0)
+    amount: Mapped[float] = mapped_column(Float, default=0.0)
+
+
+class Trade(Base):
+    """交易记录（用于持仓盈亏与归因）。"""
+    __tablename__ = "trades"
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    code: Mapped[str] = mapped_column(String(12), index=True)
+    name: Mapped[str] = mapped_column(String(40), default="")
+    side: Mapped[str] = mapped_column(String(4))  # buy / sell
+    price: Mapped[float] = mapped_column(Float)
+    qty: Mapped[int] = mapped_column(Integer)
+    fee: Mapped[float] = mapped_column(Float, default=0.0)
+    reason: Mapped[str] = mapped_column(String(60), default="")
+    emotion: Mapped[str] = mapped_column(String(20), default="")
+    created_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class AlertRule(Base):
+    """预警规则。"""
+    __tablename__ = "alert_rules"
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    code: Mapped[str] = mapped_column(String(12), index=True)
+    name: Mapped[str] = mapped_column(String(40), default="")
+    kind: Mapped[str] = mapped_column(String(20))  # price_above/price_below/pct_above/pct_below
+    threshold: Mapped[float] = mapped_column(Float)
+    channel: Mapped[str] = mapped_column(String(20), default="站内")
+    note: Mapped[str] = mapped_column(String(80), default="")
+    status: Mapped[str] = mapped_column(String(12), default="active")  # active/triggered
+    last_value: Mapped[float] = mapped_column(Float, default=0.0)
+    created_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+    triggered_at: Mapped[dt.datetime | None] = mapped_column(DateTime, nullable=True)
+
+
+class AlertEvent(Base):
+    """预警触发事件（站内通知）。"""
+    __tablename__ = "alert_events"
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    rule_id: Mapped[int] = mapped_column(Integer, index=True)
+    code: Mapped[str] = mapped_column(String(12))
+    name: Mapped[str] = mapped_column(String(40), default="")
+    message: Mapped[str] = mapped_column(String(160))
+    value: Mapped[float] = mapped_column(Float, default=0.0)
+    read: Mapped[bool] = mapped_column(default=False)
+    created_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class DailyReport(Base):
+    """AI 自动复盘日报（收盘后生成，可推送）。"""
+    __tablename__ = "daily_reports"
+    date: Mapped[dt.date] = mapped_column(Date, primary_key=True)
+    source: Mapped[str] = mapped_column(String(8), default="rule")  # llm / rule
+    title: Mapped[str] = mapped_column(String(80), default="")
+    content: Mapped[str] = mapped_column(Text, default="")  # markdown 正文
+    pushed: Mapped[bool] = mapped_column(default=False)
+    created_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class SignalStat(Base):
+    """信号历史胜率（基于全市场历史日线回测的统计，支撑 AI 证据链的『历史命中率』）。"""
+    __tablename__ = "signal_stats"
+    __table_args__ = (UniqueConstraint("signal", "horizon", name="uq_signal_horizon"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    signal: Mapped[str] = mapped_column(String(24), index=True)
+    horizon: Mapped[int] = mapped_column(Integer, default=5)  # 向后 N 个交易日
+    samples: Mapped[int] = mapped_column(Integer, default=0)
+    win_rate: Mapped[float] = mapped_column(Float, default=0.0)   # 上涨占比 %
+    avg_ret: Mapped[float] = mapped_column(Float, default=0.0)    # 平均收益 %
+    updated_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class Prediction(Base):
+    """AI 诊断/预测留痕，N 日后核验真实涨跌，形成可回溯的『实测准确率』。"""
+    __tablename__ = "predictions"
+    __table_args__ = (UniqueConstraint("code", "date", "kind", name="uq_pred"),)
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    date: Mapped[dt.date] = mapped_column(Date, index=True)
+    code: Mapped[str] = mapped_column(String(12), index=True)
+    name: Mapped[str] = mapped_column(String(40), default="")
+    kind: Mapped[str] = mapped_column(String(16), default="diagnose")
+    score: Mapped[float] = mapped_column(Float, default=0.0)
+    confidence: Mapped[float] = mapped_column(Float, default=0.0)
+    direction: Mapped[str] = mapped_column(String(6), default="flat")  # up/down/flat
+    horizon: Mapped[int] = mapped_column(Integer, default=5)
+    base_close: Mapped[float] = mapped_column(Float, default=0.0)
+    actual_ret: Mapped[float] = mapped_column(Float, default=0.0)
+    status: Mapped[str] = mapped_column(String(8), default="open")  # open/closed
+    hit: Mapped[bool | None] = mapped_column(nullable=True)
+    created_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+
+
+class JobRun(Base):
+    """定时/手动任务执行日志。"""
+    __tablename__ = "job_runs"
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    job: Mapped[str] = mapped_column(String(40))
+    status: Mapped[str] = mapped_column(String(16))  # running/success/error
+    started_at: Mapped[dt.datetime] = mapped_column(DateTime, server_default=func.now())
+    finished_at: Mapped[dt.datetime | None] = mapped_column(DateTime, nullable=True)
+    message: Mapped[str] = mapped_column(Text, default="")

backend/notifier.py

@@ -0,0 +1,84 @@
+"""推送通知：邮件(SMTP) + Server酱 + 企业微信机器人 + PushPlus。
+
+任意渠道配置了凭证即启用；notify() 会向所有已配置渠道发送，返回各渠道结果。
+"""
+from __future__ import annotations
+
+import smtplib
+import ssl
+from email.mime.text import MIMEText
+from email.header import Header
+
+import requests
+
+import config
+
+
+def channels_status():
+    return {
+        "email": bool(config.SMTP_HOST and config.SMTP_USER and config.SMTP_TO),
+        "serverchan": bool(config.SERVERCHAN_KEY),
+        "wecom": bool(config.WECOM_WEBHOOK),
+        "pushplus": bool(config.PUSHPLUS_TOKEN),
+    }
+
+
+def any_enabled():
+    return any(channels_status().values())
+
+
+def _send_email(title, content):
+    msg = MIMEText(content, "plain", "utf-8")
+    msg["Subject"] = Header(title, "utf-8")
+    msg["From"] = config.SMTP_USER
+    tos = [x.strip() for x in config.SMTP_TO.split(",") if x.strip()]
+    msg["To"] = ",".join(tos)
+    ctx = ssl.create_default_context()
+    if config.SMTP_PORT == 465:
+        with smtplib.SMTP_SSL(config.SMTP_HOST, config.SMTP_PORT, context=ctx, timeout=15) as srv:
+            srv.login(config.SMTP_USER, config.SMTP_PASSWORD)
+            srv.sendmail(config.SMTP_USER, tos, msg.as_string())
+    else:
+        with smtplib.SMTP(config.SMTP_HOST, config.SMTP_PORT, timeout=15) as srv:
+            srv.starttls(context=ctx)
+            srv.login(config.SMTP_USER, config.SMTP_PASSWORD)
+            srv.sendmail(config.SMTP_USER, tos, msg.as_string())
+    return True
+
+
+def _send_serverchan(title, content):
+    url = f"https://sctapi.ftqq.com/{config.SERVERCHAN_KEY}.send"
+    r = requests.post(url, data={"title": title, "desp": content}, timeout=12)
+    r.raise_for_status()
+    return r.json().get("code", 0) == 0
+
+
+def _send_wecom(title, content):
+    r = requests.post(config.WECOM_WEBHOOK,
+                      json={"msgtype": "text", "text": {"content": f"{title}\n{content}"}}, timeout=12)
+    r.raise_for_status()
+    return r.json().get("errcode", -1) == 0
+
+
+def _send_pushplus(title, content):
+    r = requests.post("https://www.pushplus.plus/send",
+                      json={"token": config.PUSHPLUS_TOKEN, "title": title, "content": content}, timeout=12)
+    r.raise_for_status()
+    return r.json().get("code", 0) == 200
+
+
+def notify(title, content):
+    """向所有已配置渠道推送，返回 {渠道: 'ok'|错误信息}。"""
+    st = channels_status()
+    senders = {"email": _send_email, "serverchan": _send_serverchan,
+               "wecom": _send_wecom, "pushplus": _send_pushplus}
+    result = {}
+    for ch, on in st.items():
+        if not on:
+            continue
+        try:
+            senders[ch](title, content)
+            result[ch] = "ok"
+        except Exception as e:
+            result[ch] = repr(e)[:120]
+    return result

backend/portfolio.py

@@ -0,0 +1,176 @@
+"""持仓与盈亏归因（移动加权平均成本法）。"""
+from __future__ import annotations
+
+from collections import defaultdict
+
+from sqlalchemy import select
+
+from db import get_session
+from models import Trade, StockMetric, DailyQuote, Security, IndexDaily
+
+
+def _current_prices(codes):
+    px = {}
+    if not codes:
+        return px
+    with get_session() as s:
+        for m in s.execute(select(StockMetric).where(StockMetric.code.in_(codes))).scalars():
+            px[m.code] = m.close
+        missing = [c for c in codes if c not in px]
+        for c in missing:
+            row = s.execute(select(DailyQuote.close).where(DailyQuote.code == c)
+                            .order_by(DailyQuote.date.desc()).limit(1)).scalar()
+            if row:
+                px[c] = float(row)
+    return px
+
+
+def compute():
+    with get_session() as s:
+        trades = s.execute(select(Trade).order_by(Trade.date, Trade.id)).scalars().all()
+
+    pos = defaultdict(lambda: {"qty": 0, "cost": 0.0, "name": ""})   # 持仓
+    realized = defaultdict(float)        # 各股已实现盈亏
+    by_reason = defaultdict(float)       # 按理由归因（已实现）
+    by_emotion = defaultdict(float)
+    closed_sells = []                    # 每笔卖出的盈亏（胜率用）
+
+    for t in trades:
+        p = pos[t.code]
+        p["name"] = t.name or p["name"]
+        if t.side == "buy":
+            p["cost"] += t.price * t.qty + t.fee
+            p["qty"] += t.qty
+        else:  # sell
+            if p["qty"] <= 0:
+                continue
+            avg = p["cost"] / p["qty"] if p["qty"] else 0.0
+            qty = min(t.qty, p["qty"])
+            pnl = (t.price - avg) * qty - t.fee
+            realized[t.code] += pnl
+            by_reason[t.reason or "未标注"] += pnl
+            by_emotion[t.emotion or "未标注"] += pnl
+            closed_sells.append(pnl)
+            p["cost"] -= avg * qty
+            p["qty"] -= qty
+
+    codes = [c for c, v in pos.items() if v["qty"] > 0]
+    px = _current_prices(codes)
+
+    holdings, mkt_val, hold_cost, unreal = [], 0.0, 0.0, 0.0
+    for c in codes:
+        v = pos[c]
+        avg = v["cost"] / v["qty"] if v["qty"] else 0.0
+        cur = px.get(c, avg)
+        mv = cur * v["qty"]
+        u = (cur - avg) * v["qty"]
+        mkt_val += mv; hold_cost += v["cost"]; unreal += u
+        holdings.append({"code": c, "name": v["name"], "qty": v["qty"],
+                         "avg_cost": round(avg, 3), "cur": round(cur, 3),
+                         "market_value": round(mv, 2), "unrealized": round(u, 2),
+                         "unrealized_pct": round((cur / avg - 1) * 100, 2) if avg else 0.0})
+    holdings.sort(key=lambda x: x["unrealized"], reverse=True)
+
+    total_realized = sum(realized.values())
+    wins = sum(1 for x in closed_sells if x > 0)
+    win_rate = round(wins / len(closed_sells) * 100, 1) if closed_sells else 0.0
+
+    # 盈亏归因：按个股（已实现+浮动）
+    attr_codes = defaultdict(float)
+    for c, r in realized.items():
+        attr_codes[c] += r
+    for h in holdings:
+        attr_codes[h["code"]] += h["unrealized"]
+    name_map = {h["code"]: h["name"] for h in holdings}
+    name_map.update({c: pos[c]["name"] for c in realized})
+    by_stock = sorted([{"code": c, "name": name_map.get(c, c), "pnl": round(v, 2)}
+                       for c, v in attr_codes.items()], key=lambda x: x["pnl"], reverse=True)
+
+    return {
+        "summary": {
+            "market_value": round(mkt_val, 2), "hold_cost": round(hold_cost, 2),
+            "unrealized": round(unreal, 2), "realized": round(total_realized, 2),
+            "total_pnl": round(unreal + total_realized, 2),
+            "positions": len(holdings), "closed_trades": len(closed_sells), "win_rate": win_rate,
+        },
+        "holdings": holdings,
+        "attribution": {
+            "by_stock": by_stock,
+            "by_reason": sorted([{"key": k, "pnl": round(v, 2)} for k, v in by_reason.items()], key=lambda x: x["pnl"], reverse=True),
+            "by_emotion": sorted([{"key": k, "pnl": round(v, 2)} for k, v in by_emotion.items()], key=lambda x: x["pnl"], reverse=True),
+        },
+    }
+
+
+def equity_curve():
+    """重建每日资金曲线：累计盈亏(已实现+浮动) + 投入本金净值，并对比沪深300。"""
+    with get_session() as s:
+        trades = s.execute(select(Trade).order_by(Trade.date, Trade.id)).scalars().all()
+        if not trades:
+            return {"ok": False, "msg": "暂无交易记录"}
+        codes = list({t.code for t in trades})
+        # 各股日线 {code: {date: close}}
+        px = defaultdict(dict)
+        start = min(t.date for t in trades)
+        for c in codes:
+            for d, close in s.execute(select(DailyQuote.date, DailyQuote.close)
+                                      .where(DailyQuote.code == c, DailyQuote.date >= start)
+                                      .order_by(DailyQuote.date)).all():
+                px[c][d] = float(close)
+        # 交易日序列（用沪深300的日期轴）
+        idx = s.execute(select(IndexDaily.date, IndexDaily.close)
+                        .where(IndexDaily.code == "sh000300", IndexDaily.date >= start)
+                        .order_by(IndexDaily.date)).all()
+
+    days = [d for d, _ in idx] or sorted({d for c in px for d in px[c]})
+    idx_map = {d: float(c) for d, c in idx}
+
+    # 按日推进
+    from collections import defaultdict as dd
+    pos = dd(lambda: {"qty": 0, "cost": 0.0})
+    realized = 0.0
+    ti = 0
+    last_px = {}
+    dates, equity, invested_curve, bench = [], [], [], []
+    base_idx = None
+    max_invested = 0.0
+    for d in days:
+        # 应用当天及之前所有交易
+        while ti < len(trades) and trades[ti].date <= d:
+            t = trades[ti]; p = pos[t.code]
+            if t.side == "buy":
+                p["cost"] += t.price * t.qty + t.fee; p["qty"] += t.qty
+            else:
+                if p["qty"] > 0:
+                    avg = p["cost"] / p["qty"]; qty = min(t.qty, p["qty"])
+                    realized += (t.price - avg) * qty - t.fee
+                    p["cost"] -= avg * qty; p["qty"] -= qty
+            ti += 1
+        unreal = 0.0
+        invested = 0.0
+        for c, p in pos.items():
+            if p["qty"] <= 0:
+                continue
+            close = px[c].get(d, last_px.get(c))
+            if close is None:
+                continue
+            last_px[c] = close
+            avg = p["cost"] / p["qty"]
+            unreal += (close - avg) * p["qty"]
+            invested += p["cost"]
+        max_invested = max(max_invested, invested, 1.0)
+        dates.append(d.isoformat())
+        equity.append(round(realized + unreal, 2))
+        invested_curve.append(round(invested, 2))
+        if d in idx_map:
+            base_idx = base_idx or idx_map[d]
+            bench.append(round(idx_map[d] / base_idx, 4))
+        else:
+            bench.append(bench[-1] if bench else 1.0)
+
+    # 净值：以累计最大投入为基准
+    nav = [round(1 + e / max_invested, 4) for e in equity]
+    return {"ok": True, "dates": dates, "equity": equity, "invested": invested_curve,
+            "nav": nav, "bench": bench,
+            "final_pnl": equity[-1] if equity else 0.0,
+            "max_invested": round(max_invested, 2)}

backend/rag.py

@@ -0,0 +1,37 @@
+"""轻量 RAG：检索与个股/大盘相关的资讯，做情绪标注，作为 LLM 上下文降低幻觉。
+
+当前为基于来源接口 + 关键词情绪的检索式上下文；后续可平滑升级为向量检索（embedding + 向量库）。
+"""
+from __future__ import annotations
+
+import akshare_service as svc
+
+
+def stock_news(symbol: str, limit: int = 5):
+    """返回个股相关资讯（已带利好/利空标注）。"""
+    try:
+        data = svc.get_stock_news(symbol, limit=limit)
+        return data.get("list", [])[:limit]
+    except Exception:
+        return []
+
+
+def _senti_score(items):
+    pos = sum(1 for x in items if x.get("sentiment") == "利好")
+    neg = sum(1 for x in items if x.get("sentiment") == "利空")
+    if pos > neg:
+        return "利好", pos, neg
+    if neg > pos:
+        return "利空", pos, neg
+    return "中性", pos, neg
+
+
+def stock_context(symbol: str, limit: int = 5):
+    """供 AI 诊断使用：检索资讯 + 汇总情绪。"""
+    items = stock_news(symbol, limit)
+    tone, pos, neg = _senti_score(items)
+    block = ""
+    if items:
+        block = "近期相关资讯（检索）：\n" + "\n".join(
+            f"- [{x.get('sentiment','中性')}] {x.get('title','')}（{x.get('time','')}）" for x in items)
+    return {"items": items, "tone": tone, "pos": pos, "neg": neg, "block": block}

backend/report.py

@@ -0,0 +1,168 @@
+"""AI 自动复盘日报：汇总当日盘面，生成图文日报（markdown），落库并可推送。
+
+数据全部来自中台（sector_daily / fund_flow_daily / sentiment_daily / dragon_tiger / stock_metrics），
+AI 点评与明日策略复用 ai.py（有 key 走大模型，无 key 规则降级）。
+"""
+from __future__ import annotations
+
+import datetime as dt
+
+from sqlalchemy import select, func
+
+import ai
+import llm
+import notifier
+from db import get_session
+from models import (SectorDaily, FundFlowDaily, SentimentDaily, DragonTiger,
+                    StockMetric, DailyReport)
+
+
+def _gather(date=None):
+    with get_session() as s:
+        d = (dt.date.fromisoformat(date) if date
+             else s.execute(select(func.max(SectorDaily.date))).scalar())
+        if not d:
+            return None
+        secs = s.execute(select(SectorDaily).where(SectorDaily.date == d)
+                         .order_by(SectorDaily.pct.desc())).scalars().all()
+        flows = s.execute(select(FundFlowDaily).where(FundFlowDaily.date == d)
+                          .order_by(FundFlowDaily.net.desc())).scalars().all()
+        senti = s.execute(select(SentimentDaily).where(SentimentDaily.date == d)).scalar_one_or_none()
+        lhb = s.execute(select(DragonTiger).where(DragonTiger.date == d)
+                        .order_by(DragonTiger.net.desc()).limit(6)).scalars().all()
+        mdate = s.execute(select(func.max(StockMetric.date))).scalar()
+        gainers = streak = []
+        if mdate:
+            gainers = s.execute(select(StockMetric).where(StockMetric.date == mdate)
+                                .order_by(StockMetric.pct.desc()).limit(8)).scalars().all()
+            streak = s.execute(select(StockMetric).where(StockMetric.date == mdate, StockMetric.up_streak >= 3)
+                               .order_by(StockMetric.up_streak.desc(), StockMetric.ret5.desc()).limit(8)).scalars().all()
+        return {
+            "date": d, "senti": senti, "secs": secs, "flows": flows, "lhb": lhb,
+            "gainers": [(x.name, x.code, x.pct) for x in gainers],
+            "streak": [(x.name, x.up_streak, x.ret5) for x in streak],
+        }
+
+
+def _build_markdown(g):
+    d = g["date"]
+    s = g["senti"]
+    secs, flows, lhb = g["secs"], g["flows"], g["lhb"]
+    parts = [f"# {d} 收盘复盘日报\n"]
+
+    # 一、市场温度
+    if s:
+        ratio = (s.up / max(1, s.up + s.down)) * 100
+        mood = "赚钱效应较好" if s.up > s.down * 1.5 else ("情绪偏弱" if s.down > s.up else "多空分歧")
+        parts.append("## 一、市场温度")
+        parts.append(f"- 涨跌家数：上涨 **{s.up}** / 下跌 **{s.down}** / 平盘 {s.flat}（上涨占比 {ratio:.0f}%）")
+        parts.append(f"- 涨停 **{s.limit_up}** 家 / 跌停 **{s.limit_down}** 家")
+        parts.append(f"- 综合判断：{mood}\n")
+
+    # 二、板块表现
+    if secs:
+        top = secs[:6]
+        bot = secs[-4:]
+        parts.append("## 二、板块表现")
+        parts.append("- 领涨：" + "，".join(f"{x.name}({x.pct:+.2f}%)" for x in top))
+        parts.append("- 领跌：" + "，".join(f"{x.name}({x.pct:+.2f}%)" for x in bot) + "\n")
+
+    # 三、资金流向
+    if flows:
+        parts.append("## 三、资金流向（主力净额）")
+        parts.append("- 净流入前列：" + "，".join(f"{x.name}({x.net:+.1f}亿)" for x in flows[:6]))
+        parts.append("- 净流出前列：" + "，".join(f"{x.name}({x.net:+.1f}亿)" for x in flows[-4:][::-1]) + "\n")
+
+    # 四、强势梯队
+    if g["streak"]:
+        parts.append("## 四、强势梯队（连涨）")
+        parts.append("、".join(f"{n}（{k}连阳, 5日{r:+.1f}%）" for n, k, r in g["streak"]) + "\n")
+
+    # 五、龙虎榜焦点
+    if lhb:
+        parts.append("## 五、龙虎榜焦点（净买额）")
+        parts.append("、".join(f"{n}({v:+.2f}亿)" for n, v in [(x.name, x.net) for x in lhb]) + "\n")
+
+    return "\n".join(parts)
+
+
+def generate(date=None, push=False):
+    """生成（或重生成）某日复盘日报。默认最新交易日。"""
+    g = _gather(date)
+    if not g:
+        return {"ok": False, "msg": "暂无入库数据，请先到数据中台入库"}
+    d = g["date"]
+    body = _build_markdown(g)
+
+    # AI 综合点评 + 明日策略（复用 ai.py）
+    rv = ai.review_daily_comment(d.isoformat())
+    st = ai.today_strategy()
+    source = "llm" if (rv.get("source") == "llm" or st.get("source") == "llm") else "rule"
+    body += "\n## 六、AI 综合点评\n" + (rv.get("text", "") if rv.get("ok") else "")
+    body += "\n\n## 七、明日策略\n" + (st.get("text", "") if st.get("ok") else "")
+
+    title = f"{d} 收盘复盘日报"
+    with get_session() as s:
+        row = s.get(DailyReport, d)
+        if row:
+            row.source, row.title, row.content = source, title, body
+        else:
+            row = DailyReport(date=d, source=source, title=title, content=body)
+            s.add(row)
+        s.commit()
+
+    result = {"ok": True, "date": d.isoformat(), "source": source, "title": title, "content": body}
+    if push and notifier.any_enabled():
+        try:
+            # 推送精简版（情绪 + 领涨 + AI 点评首段）
+            brief = _push_brief(g, rv)
+            res = notifier.notify("【智策】" + title, brief)
+            with get_session() as s:
+                r2 = s.get(DailyReport, d)
+                if r2:
+                    r2.pushed = True
+                    s.commit()
+            result["push"] = res
+        except Exception as e:
+            result["push"] = {"error": repr(e)[:120]}
+    return result
+
+
+def _push_brief(g, rv):
+    s = g["senti"]
+    lines = [f"{g['date']} 收盘复盘"]
+    if s:
+        lines.append(f"涨{s.up}/跌{s.down}，涨停{s.limit_up}/跌停{s.limit_down}")
+    if g["secs"]:
+        lines.append("领涨：" + "、".join(f"{x.name}({x.pct:+.1f}%)" for x in g["secs"][:4]))
+    if g["flows"]:
+        lines.append("吸金：" + "、".join(x.name for x in g["flows"][:4]))
+    txt = (rv.get("text") or "").strip().split("\n")[0]
+    if txt:
+        lines.append("点评：" + txt[:80])
+    return "\n".join(lines)
+
+
+def latest():
+    with get_session() as s:
+        row = s.execute(select(DailyReport).order_by(DailyReport.date.desc())).scalars().first()
+        if not row:
+            return {"ok": False, "msg": "暂无日报，请先生成"}
+        return {"ok": True, "date": row.date.isoformat(), "source": row.source,
+                "title": row.title, "content": row.content, "pushed": row.pushed}
+
+
+def get_by_date(date):
+    with get_session() as s:
+        row = s.get(DailyReport, dt.date.fromisoformat(date))
+        if not row:
+            return {"ok": False, "msg": "该日无日报"}
+        return {"ok": True, "date": row.date.isoformat(), "source": row.source,
+                "title": row.title, "content": row.content, "pushed": row.pushed}
+
+
+def history(limit=30):
+    with get_session() as s:
+        rows = s.execute(select(DailyReport).order_by(DailyReport.date.desc()).limit(limit)).scalars().all()
+        return {"list": [{"date": r.date.isoformat(), "source": r.source,
+                          "title": r.title, "pushed": r.pushed} for r in rows]}

backend/requirements.txt

@@ -0,0 +1,8 @@
+fastapi==0.115.0
+uvicorn[standard]==0.30.6
+akshare>=1.18.60
+pandas>=2.2.2
+cachetools==5.5.0
+SQLAlchemy>=2.0.30
+APScheduler>=3.10.4
+psycopg2-binary>=2.9.9

backend/scheduler.py

@@ -0,0 +1,144 @@
+"""定时任务：收盘后自动入库。"""
+from __future__ import annotations
+
+import threading
+
+from apscheduler.schedulers.background import BackgroundScheduler
+from apscheduler.triggers.cron import CronTrigger
+from apscheduler.triggers.interval import IntervalTrigger
+
+import config
+import ingest
+import alerts
+import report
+import signals
+
+_scheduler: BackgroundScheduler | None = None
+_lock = threading.Lock()
+_running = {"flag": False}
+
+
+def _job():
+    with _lock:
+        if _running["flag"]:
+            return
+        _running["flag"] = True
+    try:
+        ingest.run_daily_ingest()
+    finally:
+        _running["flag"] = False
+
+
+def trigger_async():
+    """手动触发一次入库（后台线程，不阻塞请求）。"""
+    t = threading.Thread(target=_job, daemon=True)
+    t.start()
+    return {"started": True}
+
+
+def _job_all():
+    with _lock:
+        if _running["flag"]:
+            return
+        _running["flag"] = True
+    try:
+        ingest.ingest_quotes_all()
+    finally:
+        _running["flag"] = False
+
+
+def trigger_all_async():
+    """手动触发全市场回填（后台线程，耗时较长）。"""
+    if _running["flag"]:
+        return {"started": False, "msg": "已有任务在执行"}
+    t = threading.Thread(target=_job_all, daemon=True)
+    t.start()
+    return {"started": True}
+
+
+def is_running():
+    return _running["flag"]
+
+
+def _job_report():
+    try:
+        report.generate(push=True)
+    except Exception as e:
+        print("[report] generate error:", repr(e)[:160])
+
+
+def trigger_report_async(push=True):
+    """手动触发生成日报（后台线程）。"""
+    t = threading.Thread(target=lambda: report.generate(push=push), daemon=True)
+    t.start()
+    return {"started": True}
+
+
+_sig_running = {"flag": False}
+
+
+def _job_signal_stats(sample=500, horizon=5):
+    if _sig_running["flag"]:
+        return
+    _sig_running["flag"] = True
+    try:
+        signals.compute_signal_stats(sample_limit=sample, horizon=horizon)
+    except Exception as e:
+        print("[signals] compute error:", repr(e)[:160])
+    finally:
+        _sig_running["flag"] = False
+
+
+def trigger_signal_stats_async(sample=500, horizon=5):
+    if _sig_running["flag"]:
+        return {"started": False, "msg": "胜率回测进行中"}
+    t = threading.Thread(target=lambda: _job_signal_stats(sample, horizon), daemon=True)
+    t.start()
+    return {"started": True}
+
+
+def _job_verify():
+    try:
+        signals.verify_predictions()
+    except Exception as e:
+        print("[predict] verify error:", repr(e)[:160])
+
+
+def start_scheduler():
+    global _scheduler
+    if _scheduler is not None:
+        return _scheduler
+    _scheduler = BackgroundScheduler(timezone="Asia/Shanghai")
+    _scheduler.add_job(
+        _job, CronTrigger(day_of_week="mon-fri", hour=config.INGEST_HOUR, minute=config.INGEST_MINUTE),
+        id="daily_ingest", replace_existing=True, misfire_grace_time=3600,
+    )
+    _scheduler.add_job(
+        _safe_check_alerts, IntervalTrigger(seconds=60),
+        id="alert_check", replace_existing=True, max_instances=1,
+    )
+    # 收盘入库之后 10 分钟生成 AI 复盘日报并推送
+    _rep_total = config.INGEST_HOUR * 60 + config.INGEST_MINUTE + 10
+    _scheduler.add_job(
+        _job_report, CronTrigger(day_of_week="mon-fri", hour=(_rep_total // 60) % 24, minute=_rep_total % 60),
+        id="daily_report", replace_existing=True, misfire_grace_time=3600,
+    )
+    # 收盘后核验到期预测（实测准确率）
+    _scheduler.add_job(
+        _job_verify, CronTrigger(day_of_week="mon-fri", hour=(_rep_total // 60) % 24, minute=(_rep_total + 5) % 60),
+        id="verify_pred", replace_existing=True, misfire_grace_time=3600,
+    )
+    # 每周六重算信号历史胜率
+    _scheduler.add_job(
+        _job_signal_stats, CronTrigger(day_of_week="sat", hour=9, minute=0),
+        id="signal_stats", replace_existing=True, misfire_grace_time=7200,
+    )
+    _scheduler.start()
+    return _scheduler
+
+
+def _safe_check_alerts():
+    try:
+        alerts.check_alerts()
+    except Exception as e:
+        print("[alert] check error:", repr(e)[:120])

backend/signals.py

@@ -0,0 +1,194 @@
+"""信号历史胜率回测 + AI 预测留痕核验。
+
+- compute_signal_stats: 对全市场历史日线回测各技术信号「N 日后上涨概率/平均收益」，
+  作为 AI 证据链『历史命中率』的客观依据。
+- record_prediction / verify_predictions / accuracy: 记录每次 AI 诊断，N 日后核验真实涨跌，
+  形成可回溯的『实测准确率』。
+"""
+from __future__ import annotations
+
+import datetime as dt
+
+import numpy as np
+import pandas as pd
+from sqlalchemy import select, func, distinct
+
+from db import get_session
+from models import DailyQuote, StockMetric, SignalStat, Prediction
+
+SIGNAL_DEFS = {
+    "ma_bull": "均线多头排列(MA5>MA10>MA20)",
+    "macd_gold": "MACD 金叉",
+    "up_streak3": "三连阳",
+    "vol_breakout": "放量上涨(量比>2)",
+    "rsi_oversold": "RSI 超卖(<30)抄底",
+    "new_high60": "创60日新高",
+}
+
+
+def _indicators(df: pd.DataFrame) -> pd.DataFrame:
+    c = df["close"]
+    df["ma5"] = c.rolling(5).mean()
+    df["ma10"] = c.rolling(10).mean()
+    df["ma20"] = c.rolling(20).mean()
+    e12 = c.ewm(span=12, adjust=False).mean()
+    e26 = c.ewm(span=26, adjust=False).mean()
+    df["dif"] = e12 - e26
+    df["dea"] = df["dif"].ewm(span=9, adjust=False).mean()
+    delta = c.diff()
+    gain = delta.clip(lower=0)
+    loss = -delta.clip(upper=0)
+    ag = gain.rolling(14).mean()
+    al = loss.rolling(14).mean()
+    df["rsi"] = 100 - 100 / (1 + ag / al.replace(0, np.nan))
+    df["vavg5"] = df["volume"].rolling(5).mean().shift(1)
+    df["high60"] = c.rolling(60).max()
+    return df
+
+
+def _masks(df: pd.DataFrame) -> dict:
+    c = df["close"]
+    c1 = c.shift(1)
+    return {
+        "ma_bull": (df.ma5 > df.ma10) & (df.ma10 > df.ma20),
+        "macd_gold": (df.dif.shift(1) < df.dea.shift(1)) & (df.dif >= df.dea),
+        "up_streak3": (c > c1) & (c1 > c.shift(2)) & (c.shift(2) > c.shift(3)),
+        "vol_breakout": (df.volume / df.vavg5 > 2) & (c > c1),
+        "rsi_oversold": df.rsi < 30,
+        "new_high60": c >= df.high60,
+    }
+
+
+def compute_signal_stats(sample_limit: int = 500, horizon: int = 5):
+    """对样本股历史回测各信号 N 日后表现并落库。"""
+    with get_session() as s:
+        codes = [r[0] for r in s.execute(select(StockMetric.code).limit(sample_limit)).all()]
+        if not codes:
+            codes = [r[0] for r in s.execute(select(distinct(DailyQuote.code)).limit(sample_limit)).all()]
+
+    agg = {k: {"n": 0, "win": 0, "sum": 0.0} for k in SIGNAL_DEFS}
+    used = 0
+    with get_session() as s:
+        for code in codes:
+            rows = s.execute(select(DailyQuote.date, DailyQuote.close, DailyQuote.volume)
+                             .where(DailyQuote.code == code).order_by(DailyQuote.date)).all()
+            if len(rows) < 80:
+                continue
+            used += 1
+            df = pd.DataFrame(rows, columns=["date", "close", "volume"])
+            df["close"] = df["close"].astype(float)
+            df["volume"] = df["volume"].astype(float)
+            df = _indicators(df)
+            fwd = df["close"].shift(-horizon) / df["close"] - 1
+            for k, m in _masks(df).items():
+                m = m & fwd.notna()
+                r = fwd[m]
+                if len(r):
+                    agg[k]["n"] += int(len(r))
+                    agg[k]["win"] += int((r > 0).sum())
+                    agg[k]["sum"] += float(r.sum())
+
+    with get_session() as s:
+        for k, a in agg.items():
+            if a["n"] == 0:
+                continue
+            wr = round(a["win"] / a["n"] * 100, 1)
+            ar = round(a["sum"] / a["n"] * 100, 2)
+            row = s.execute(select(SignalStat).where(SignalStat.signal == k, SignalStat.horizon == horizon)).scalar_one_or_none()
+            if row:
+                row.samples, row.win_rate, row.avg_ret, row.updated_at = a["n"], wr, ar, dt.datetime.now()
+            else:
+                s.add(SignalStat(signal=k, horizon=horizon, samples=a["n"], win_rate=wr, avg_ret=ar))
+        s.commit()
+    return {"ok": True, "horizon": horizon, "sampled": used,
+            "result": {k: {"samples": a["n"], "win_rate": round(a["win"] / a["n"] * 100, 1) if a["n"] else None}
+                       for k, a in agg.items()}}
+
+
+def get_stats(horizon: int = 5) -> dict:
+    with get_session() as s:
+        rows = s.execute(select(SignalStat).where(SignalStat.horizon == horizon)).scalars().all()
+    return {r.signal: {"label": SIGNAL_DEFS.get(r.signal, r.signal), "win_rate": r.win_rate,
+                       "avg_ret": r.avg_ret, "samples": r.samples} for r in rows}
+
+
+def active_signals(m: StockMetric) -> list[str]:
+    """根据最新因子快照判断当前激活的信号。"""
+    out = []
+    if m.ma_bull:
+        out.append("ma_bull")
+    if m.macd_gold:
+        out.append("macd_gold")
+    if m.up_streak >= 3:
+        out.append("up_streak3")
+    if m.vol_ratio > 2 and m.pct > 0:
+        out.append("vol_breakout")
+    if m.rsi14 < 30:
+        out.append("rsi_oversold")
+    if m.pos60 >= 0.99:
+        out.append("new_high60")
+    return out
+
+
+# ---------------- 预测留痕与核验 ----------------
+def record_prediction(code, name, date, score, confidence, direction, base_close, horizon=5, kind="diagnose"):
+    with get_session() as s:
+        exist = s.execute(select(Prediction).where(
+            Prediction.code == code, Prediction.date == date, Prediction.kind == kind)).scalar_one_or_none()
+        if exist:
+            return False
+        s.add(Prediction(date=date, code=code, name=name, kind=kind, score=score,
+                         confidence=confidence, direction=direction, horizon=horizon, base_close=base_close))
+        s.commit()
+        return True
+
+
+def verify_predictions():
+    """对到期(已有 horizon 个交易日)的 open 预测核验真实涨跌。"""
+    with get_session() as s:
+        opens = s.execute(select(Prediction).where(Prediction.status == "open")).scalars().all()
+        closed = 0
+        for p in opens:
+            future = s.execute(select(DailyQuote.close).where(
+                DailyQuote.code == p.code, DailyQuote.date > p.date)
+                .order_by(DailyQuote.date).limit(p.horizon)).all()
+            if len(future) < p.horizon:
+                continue
+            end = float(future[-1][0])
+            ret = (end / p.base_close - 1) * 100 if p.base_close else 0.0
+            if p.direction == "up":
+                hit = ret > 0
+            elif p.direction == "down":
+                hit = ret < 0
+            else:
+                hit = abs(ret) < 2
+            p.actual_ret = round(ret, 2)
+            p.hit = bool(hit)
+            p.status = "closed"
+            closed += 1
+        s.commit()
+    return {"ok": True, "closed": closed}
+
+
+def accuracy():
+    with get_session() as s:
+        rows = s.execute(select(Prediction).where(Prediction.status == "closed")).scalars().all()
+        opens = s.execute(select(func.count()).select_from(Prediction).where(Prediction.status == "open")).scalar()
+    n = len(rows)
+    hits = sum(1 for r in rows if r.hit)
+    by_dir = {}
+    for r in rows:
+        d = by_dir.setdefault(r.direction, {"n": 0, "hit": 0})
+        d["n"] += 1
+        d["hit"] += 1 if r.hit else 0
+    recent = sorted(rows, key=lambda r: (r.date, r.id), reverse=True)[:25]
+    return {
+        "closed": n, "open": opens or 0,
+        "hit_rate": round(hits / n * 100, 1) if n else None,
+        "avg_ret": round(sum(r.actual_ret for r in rows) / n, 2) if n else None,
+        "by_direction": {k: {"n": v["n"], "hit_rate": round(v["hit"] / v["n"] * 100, 1)}
+                         for k, v in by_dir.items()},
+        "recent": [{"date": r.date.isoformat(), "code": r.code, "name": r.name,
+                    "direction": r.direction, "score": r.score, "confidence": r.confidence,
+                    "actual_ret": r.actual_ret, "hit": r.hit, "horizon": r.horizon} for r in recent],
+    }

prototype/index.html

@@ -0,0 +1,43 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+  <title>智策 · 股票分析复盘终端</title>
+  <script src="https://cdn.jsdelivr.net/npm/echarts@5.5.0/dist/echarts.min.js"></script>
+  <link rel="stylesheet" href="style.css" />
+</head>
+<body>
+  <!-- 顶部栏 -->
+  <header class="topbar">
+    <div class="brand">智策 <em>StockTerminal</em></div>
+    <div class="search">
+      <input type="text" placeholder="搜索股票 / 代码 / 板块  (Ctrl+K)" />
+    </div>
+    <div class="ticker" id="ticker"></div>
+    <div class="top-right">
+      <span id="alert-bell" title="预警提醒">🔔<span id="alert-count" class="bell-count" style="display:none">0</span></span>
+      <span id="dsource" title="当前数据来源">数据源: -</span>
+      <span class="clock" id="clock"></span>
+      <span class="market-state" id="mstate">●  交易中</span>
+      <span class="user">游客</span>
+    </div>
+  </header>
+
+  <div class="layout">
+    <!-- 左侧菜单 -->
+    <aside class="sidebar" id="sidebar">
+      <nav id="menu"></nav>
+      <div class="collapse" id="collapseBtn" title="折叠菜单">«</div>
+    </aside>
+
+    <!-- 内容区 -->
+    <main class="content">
+      <div class="crumb" id="crumb"></div>
+      <div id="view"></div>
+    </main>
+  </div>
+
+  <script src="app.js"></script>
+</body>
+</html>

prototype/style.css

@@ -0,0 +1,194 @@
+:root {
+  --bg: #0a0e15;
+  --topbar: #0f141d;
+  --sidebar: #0d121a;
+  --sidebar-2: #131a25;
+  --panel: #121822;
+  --panel-head: #161d29;
+  --border: #232c3a;
+  --border-soft: #1b222e;
+  --text: #d7dee8;
+  --text-dim: #7d8796;
+  --text-mute: #586273;
+  --up: #f6465d;     /* 红涨 */
+  --down: #2ebd85;   /* 绿跌 */
+  --accent: #2f6fed;
+  --gold: #e8a13a;
+  --radius: 2px;
+}
+* { box-sizing: border-box; margin: 0; padding: 0; }
+html, body { height: 100%; }
+body {
+  background: var(--bg);
+  color: var(--text);
+  font-family: -apple-system, "Segoe UI", "Microsoft YaHei", sans-serif;
+  font-size: 13px;
+  overflow: hidden;
+}
+.up { color: var(--up); }
+.down { color: var(--down); }
+.num { font-variant-numeric: tabular-nums; }
+
+/* ============ 顶部栏 ============ */
+.topbar {
+  height: 48px; display: flex; align-items: center; gap: 18px;
+  background: var(--topbar); border-bottom: 1px solid var(--border);
+  padding: 0 16px;
+}
+.brand { font-size: 16px; font-weight: 700; letter-spacing: .5px; white-space: nowrap; }
+.brand em { color: var(--gold); font-style: normal; font-weight: 500; font-size: 12px; margin-left: 4px; }
+.search { width: 280px; }
+.search input {
+  width: 100%; height: 30px; background: #0a0e15; border: 1px solid var(--border);
+  border-radius: var(--radius); color: var(--text); padding: 0 10px; font-size: 12px; outline: none;
+}
+.search input:focus { border-color: var(--accent); }
+.ticker { flex: 1; overflow: hidden; display: flex; gap: 22px; white-space: nowrap; color: var(--text-dim); font-size: 12px; }
+.ticker .ti { display: inline-flex; gap: 6px; align-items: center; }
+.ticker .ti b { color: var(--text); font-weight: 500; }
+.top-right { display: flex; align-items: center; gap: 16px; white-space: nowrap; font-size: 12px; color: var(--text-dim); }
+.clock { font-variant-numeric: tabular-nums; }
+.market-state { color: var(--down); }
+.user { padding: 3px 10px; border: 1px solid var(--border); border-radius: var(--radius); }
+
+/* ============ 布局 ============ */
+.layout { display: flex; height: calc(100vh - 48px); }
+
+/* ============ 侧边栏 ============ */
+.sidebar {
+  width: 208px; background: var(--sidebar); border-right: 1px solid var(--border);
+  display: flex; flex-direction: column; transition: width .15s; flex-shrink: 0;
+}
+.sidebar.collapsed { width: 52px; }
+.sidebar nav { flex: 1; overflow-y: auto; padding: 6px 0; }
+.sidebar nav::-webkit-scrollbar { width: 6px; }
+.sidebar nav::-webkit-scrollbar-thumb { background: #222b39; }
+
+.menu-group > .g-head {
+  display: flex; align-items: center; gap: 10px; height: 38px; padding: 0 14px;
+  cursor: pointer; color: var(--text-dim); user-select: none; font-size: 13px;
+  border-left: 2px solid transparent;
+}
+.menu-group > .g-head:hover { background: var(--sidebar-2); color: var(--text); }
+.menu-group.open > .g-head { color: var(--text); }
+.g-head .ico { width: 16px; text-align: center; font-size: 14px; flex-shrink: 0; }
+.g-head .g-name { flex: 1; white-space: nowrap; overflow: hidden; }
+.g-head .arrow { font-size: 10px; color: var(--text-mute); transition: transform .15s; }
+.menu-group.open > .g-head .arrow { transform: rotate(90deg); }
+
+.submenu { display: none; background: #0a0e15; }
+.menu-group.open .submenu { display: block; }
+.submenu a {
+  display: block; height: 32px; line-height: 32px; padding: 0 14px 0 40px;
+  color: var(--text-dim); text-decoration: none; font-size: 12.5px; white-space: nowrap;
+  border-left: 2px solid transparent;
+}
+.submenu a:hover { color: var(--text); background: var(--sidebar-2); }
+.submenu a.active { color: #fff; background: #16243f; border-left-color: var(--accent); }
+
+.sidebar.collapsed .g-name, .sidebar.collapsed .arrow, .sidebar.collapsed .submenu { display: none; }
+.sidebar.collapsed .g-head { justify-content: center; padding: 0; }
+
+.collapse {
+  height: 34px; border-top: 1px solid var(--border); display: flex; align-items: center;
+  justify-content: center; cursor: pointer; color: var(--text-mute); font-size: 14px;
+}
+.collapse:hover { color: var(--text); background: var(--sidebar-2); }
+
+/* ============ 内容区 ============ */
+.content { flex: 1; overflow-y: auto; padding: 12px 14px 24px; }
+.content::-webkit-scrollbar { width: 8px; }
+.content::-webkit-scrollbar-thumb { background: #222b39; }
+.crumb { color: var(--text-mute); font-size: 12px; margin-bottom: 10px; }
+.crumb b { color: var(--text); font-weight: 500; }
+.crumb .sep { margin: 0 6px; }
+
+/* ============ 通用面板 ============ */
+.panel { background: var(--panel); border: 1px solid var(--border-soft); border-radius: var(--radius); }
+.panel-head {
+  height: 36px; display: flex; align-items: center; gap: 10px; padding: 0 12px;
+  border-bottom: 1px solid var(--border-soft); background: var(--panel-head); font-size: 13px; font-weight: 600;
+}
+.panel-head .bar { width: 3px; height: 13px; background: var(--accent); }
+.panel-head .sub { font-weight: 400; color: var(--text-dim); font-size: 11.5px; }
+.panel-head .seg { margin-left: auto; display: flex; }
+.seg button {
+  background: #0d121a; color: var(--text-dim); border: 1px solid var(--border);
+  border-right: none; padding: 3px 11px; cursor: pointer; font-size: 11.5px;
+}
+.seg button:last-child { border-right: 1px solid var(--border); }
+.seg button.active { color: #fff; background: var(--accent); border-color: var(--accent); }
+.panel-body { padding: 10px 12px; }
+
+.row { display: grid; gap: 10px; margin-bottom: 10px; }
+.row.c4 { grid-template-columns: repeat(4, 1fr); }
+.row.c2 { grid-template-columns: 1fr 1fr; }
+.row.c32 { grid-template-columns: 3fr 2fr; }
+@media (max-width: 1200px){ .row.c4 { grid-template-columns: repeat(2,1fr);} .row.c2,.row.c32{grid-template-columns:1fr;} }
+
+/* 指数卡片 */
+.idx-card { background: var(--panel); border: 1px solid var(--border-soft); border-radius: var(--radius); padding: 10px 12px; position: relative; overflow: hidden; }
+.idx-card .nm { font-size: 12px; color: var(--text-dim); }
+.idx-card .pr { font-size: 22px; font-weight: 700; margin: 3px 0; }
+.idx-card .cg { font-size: 12px; }
+.idx-card .sp { position: absolute; right: 6px; bottom: 6px; width: 88px; height: 32px; opacity: .85; }
+
+/* 情绪条 */
+.senti { display: flex; align-items: center; gap: 26px; flex-wrap: wrap; }
+.senti .it { display: flex; flex-direction: column; gap: 2px; }
+.senti .it .l { font-size: 11px; color: var(--text-dim); }
+.senti .it .v { font-size: 17px; font-weight: 600; }
+.rbar { flex: 1; min-width: 220px; height: 20px; display: flex; border-radius: var(--radius); overflow: hidden; }
+.rbar .ru { background: var(--up); } .rbar .rf { background: #4b5563; } .rbar .rd { background: var(--down); }
+
+/* 表格 */
+table.grid-tbl { width: 100%; border-collapse: collapse; font-size: 12.5px; }
+.grid-tbl th { text-align: right; color: var(--text-dim); font-weight: 500; padding: 7px 10px; border-bottom: 1px solid var(--border); background: var(--panel-head); position: sticky; top: 0; }
+.grid-tbl th:first-child, .grid-tbl td:first-child { text-align: left; }
+.grid-tbl td { text-align: right; padding: 7px 10px; border-bottom: 1px solid var(--border-soft); }
+.grid-tbl tr:hover td { background: var(--sidebar-2); }
+.grid-tbl tbody tr:hover td { background: #131a25; }
+.tag { display: inline-block; padding: 1px 7px; border-radius: var(--radius); font-size: 11px; border: 1px solid var(--border); color: var(--text-dim); }
+.tag.hot { color: var(--up); border-color: #5a2630; background: #2a1418; }
+
+/* 占位（建设中） */
+.placeholder { padding: 26px; }
+.placeholder h2 { font-size: 16px; margin-bottom: 6px; }
+.placeholder p { color: var(--text-dim); margin-bottom: 16px; }
+.feat-list { display: grid; grid-template-columns: repeat(auto-fill, minmax(230px,1fr)); gap: 10px; }
+.feat-card { border: 1px solid var(--border-soft); border-radius: var(--radius); padding: 12px 14px; background: var(--panel); }
+.feat-card .ft { font-weight: 600; margin-bottom: 4px; }
+.feat-card .fd { color: var(--text-dim); font-size: 12px; line-height: 1.6; }
+.badge { font-size: 11px; color: var(--gold); border: 1px solid #4a3a18; background: #241c0d; padding: 1px 6px; border-radius: var(--radius); margin-left: 6px; }
+
+.loading { padding: 40px; text-align: center; color: var(--text-mute); }
+.btn-run { background: var(--accent); color: #fff; border: 1px solid var(--accent); height: 26px; padding: 0 14px; cursor: pointer; border-radius: var(--radius); font-size: 12px; }
+.btn-run:hover { filter: brightness(1.1); }
+.btn-run:disabled { background: #2a3140; border-color: #2a3140; color: var(--text-mute); cursor: not-allowed; }
+.bt-form input:focus { outline: none; border-color: var(--accent); }
+.cond { display: flex; align-items: center; gap: 6px; color: var(--text-dim); font-size: 12.5px; }
+.cond input[type=number] { width: 70px; height: 26px; background: #0a0e15; border: 1px solid var(--border); color: var(--text); padding: 0 8px; }
+.cond input:focus { outline: none; border-color: var(--accent); }
+.clickrow { cursor: pointer; }
+#alert-bell { cursor: pointer; position: relative; font-size: 15px; }
+.bell-count { position: absolute; top: -7px; right: -10px; background: var(--up); color: #fff; font-size: 10px; min-width: 15px; height: 15px; line-height: 15px; text-align: center; border-radius: 8px; padding: 0 3px; }
+.news-list { display: flex; flex-direction: column; }
+.news-item { padding: 10px 4px; border-bottom: 1px solid var(--border-soft); }
+.news-meta { font-size: 12px; color: var(--text-dim); margin-bottom: 4px; }
+.news-time { color: var(--text-mute); }
+.news-title { font-size: 14px; font-weight: 600; color: var(--text); line-height: 1.5; }
+.news-sum { font-size: 12.5px; color: var(--text-dim); margin-top: 4px; line-height: 1.6; }
+.news-act { margin-top: 6px; }
+.ai-box { margin-top: 8px; background: #0e1420; border: 1px solid var(--border-soft); border-left: 2px solid var(--accent); padding: 8px 10px; font-size: 12.5px; line-height: 1.7; color: var(--text); }
+.md-doc { line-height: 1.85; color: var(--text); max-width: 920px; }
+.md-doc .md-h1 { font-size: 18px; font-weight: 700; color: var(--text); margin: 2px 0 12px; padding-bottom: 8px; border-bottom: 1px solid var(--border); }
+.md-doc .md-h2 { font-size: 14px; font-weight: 600; color: var(--text); margin: 16px 0 6px; display: flex; align-items: center; gap: 8px; }
+.md-doc .md-h2 .bar { display: inline-block; width: 3px; height: 13px; background: var(--accent); }
+.md-doc .md-ul { margin: 4px 0 4px 2px; padding-left: 16px; }
+.md-doc .md-ul li { margin: 3px 0; color: var(--text-dim); }
+.md-doc .md-p { margin: 6px 0; color: var(--text-dim); }
+.md-doc b { color: var(--text); }
+#dsource { color: var(--text-mute); font-size: 11.5px; }
+#kline { width: 100%; height: 480px; }
+#treemap { width: 100%; height: 480px; }
+#fundflow { width: 100%; height: 360px; }

功能架构.md

@@ -0,0 +1,160 @@
+# 股票分析 · 建议 · 复盘系统 — 功能架构设计
+
+> 版本：v2.0（优化版）
+> 定位：一套覆盖「行情监控 → 选股 → 复盘归因 → AI 辅助决策 → 策略回测」的个人/小团队量化辅助系统
+> 核心理念：**数据先行、AI 结论可回溯、复盘形成闭环**
+
+---
+
+## 一、设计目标与核心壁垒
+
+| 目标 | 说明 |
+|------|------|
+| 数据中台化 | 行情/资金/财务/新闻/龙虎榜统一清洗入库，可历史回溯 |
+| 复盘闭环 | 每日复盘 + 个股复盘 + 交易复盘（盈亏归因），训练盘感 |
+| 可信 AI | 所有 AI 建议附「依据 + 置信度 + 风险点 + 历史命中率」 |
+| 选股→回测→盯盘打通 | 一个策略可一键回测、一键存为预警 |
+| 全局预警 | 价格/量能/技术/资金/新闻触发，多渠道推送 |
+
+---
+
+## 二、整体分层架构
+
+```mermaid
+graph TD
+    A[数据采集层<br/>行情/资金/财务/新闻/龙虎榜] --> B[数据中台<br/>清洗·存储·因子计算·快照]
+    B --> C1[行情中心]
+    B --> C2[选股引擎]
+    B --> C3[复盘中心]
+    B --> C4[策略与回测]
+    B --> C5[组合与交易]
+    B --> C6[资讯中心]
+    B --> D[AI 分析层<br/>RAG+因子+LLM]
+    D --> C1 & C2 & C3
+    C2 & C3 & C4 & C5 --> E[智能预警/通知]
+    C1 & C2 & C3 & C4 & C5 & C6 & D --> F[Web/App 展示层]
+```
+
+---
+
+## 三、功能模块清单
+
+### 1. 行情中心
+- **大盘总览**
+  - 大盘云图（涨跌热力图，可按市值/成交额加权，参考 dapanyuntu.com）
+  - 三大指数 + 北向资金 + 涨跌停家数 + 封板率
+  - A股 / 美股 / 港股 大盘与板块
+- **资金流向**
+  - 主力 / 超大单 / 大单 / 散户 资金分布
+  - 板块资金轮动（桑基图），支持 日 / 周 / 月 / 年
+- **市场情绪温度计**（新增）
+  - 赚钱效应、连板高度、炸板率、量能对比、涨跌比
+
+### 2. 热榜与异动
+- 热股 / 热板块 / 热 ETF / 龙虎榜
+- 强势股：连续增长（三日 >20% / 两周连涨 / 两月连涨）
+- **盘中异动雷达**（新增）：快速拉升、放量突破、涨停打开、大单扫货
+- **龙虎榜深挖**（新增）：游资席位画像、机构净买、一线游资跟踪
+
+### 3. 选股引擎
+- **内置策略**：暴涨 / 暴跌 / 抄底 / 突破 / 底部放量 / MACD金叉 / 筹码集中
+- **自定义条件组合器**（新增）：技术面 + 资金面 + 基本面 多因子拖拽组合
+- **板块选股**：按板块、概念、产业链筛选
+- 选股结果可：一键回测 / 一键存为预警 / 加入自选
+
+### 4. 自选股
+- 自选列表 + 分组
+- 个股详情（行情、技术、资金、财务、消息、AI 诊断聚合页）
+- 上新股（次新股池、打新日历）
+
+### 5. 复盘中心 ⭐（核心模块）
+- **每日复盘**：大盘总结、领涨/领跌板块、涨停梯队、资金流向、明日关注
+- **个股复盘**：走势回放、关键买卖点标注、形态识别
+- **交易复盘**：基于真实交易记录的盈亏归因（赚在哪/亏在哪/是否追高）
+- **AI 自动复盘日报**（新增）：收盘后自动生成图文报告
+- **历史情景回放**（新增）：时间轴重放某天盘面，训练盘感
+
+### 6. 组合与交易日志（新增模块）
+- 多组合 / 模拟盘管理
+- 持仓盈亏、仓位分布、成本分析
+- 交易记录（买卖点 + 理由 + 情绪标签）
+- 风险指标：最大回撤、夏普、胜率、盈亏比
+
+### 7. 策略与回测引擎
+- 向量化回测、参数寻优、滑点 / 手续费建模
+- 资金曲线、回撤曲线、月度收益热力图
+- 技术指标库（MA/MACD/KDJ/BOLL/RSI…）
+- 策略库 + 实盘跟踪对比
+
+### 8. AI 分析层
+- **今天炒什么 / 接下来炒什么**：附资金、题材、政策、相似历史依据
+- **买卖建议**：理由 + 置信度 + 风险点 + 止损位 + 历史命中率
+- **AI 预测**：概率区间 + 模型历史准确率（不给死点位）
+- **个股 AI 诊断**：技术/资金/基本面/消息面 打分卡
+- **题材脉络图**：政策/事件 → 受益板块 → 龙头股 传导链
+- RAG 接入资讯 + 财报，降低幻觉
+
+### 9. 资讯中心
+- 热点 / 要闻 / 快讯 / 自选股相关
+- **AI 摘要 + 利好利空判定 + 关联个股**（新增）
+- 事件日历：财报、解禁、分红、政策
+
+### 10. 智能预警（贯穿全局）
+- 触发类型：价格 / 涨跌幅 / 量能 / 技术信号 / 资金 / 新闻
+- 推送渠道：站内 / 微信 / 邮件 / App
+
+---
+
+## 四、数据模型（核心表草案）
+
+| 表名 | 用途 | 关键字段 |
+|------|------|----------|
+| `securities` | 证券基础信息 | code, name, market, industry, list_date |
+| `quotes_daily` | 日线行情 | code, date, ohlcv, amount, turnover |
+| `quotes_minute` | 分时行情 | code, datetime, price, volume |
+| `fund_flow` | 资金流向 | code/sector, date, main_net, super_big… |
+| `dragon_tiger` | 龙虎榜 | code, date, seat, buy, sell, type |
+| `financials` | 财务数据 | code, report_date, revenue, profit, roe… |
+| `news` | 资讯 | id, time, title, content, sentiment, related_codes |
+| `factors` | 因子库 | code, date, factor_name, value |
+| `watchlist` | 自选 | user_id, code, group |
+| `portfolios` / `trades` | 组合与交易 | user_id, code, side, price, qty, reason, time |
+| `strategies` / `backtests` | 策略与回测 | id, params, metrics, equity_curve |
+| `alerts` | 预警规则 | user_id, code, condition, channel, status |
+| `reviews` | 复盘记录 | date/code, type, content(AI生成), tags |
+
+---
+
+## 五、技术选型
+
+| 层 | 选型 | 说明 |
+|----|------|------|
+| 数据源 | AkShare / Tushare（A股）、yfinance（美股）、东方财富/同花顺接口、新闻爬虫 | 多源互补 |
+| 后端 | Python + FastAPI | 金融生态最佳，pandas/TA-Lib/backtrader/vectorbt |
+| 时序存储 | TimescaleDB 或 ClickHouse | 海量行情读写 |
+| 业务存储 | PostgreSQL | 用户/组合/策略 |
+| 缓存/实时 | Redis | 行情缓存、实时推送 |
+| AI | LLM（DeepSeek/通义/OpenAI）+ 向量库（RAG）+ 因子模型 | 可回溯建议 |
+| 前端 | Next.js / Vue + ECharts | K线、云图、资金桑基图 |
+| 任务调度 | APScheduler / Celery | 收盘自动复盘、盘中预警 |
+
+---
+
+## 六、实施路线（分阶段）
+
+1. **第一阶段（地基）**：数据中台 + 行情中心 + 自选股 + K线/技术指标
+2. **第二阶段（选股+复盘）**：选股引擎 + 每日复盘 + 资讯中心
+3. **第三阶段（闭环）**：组合交易日志 + 回测引擎 + 智能预警
+4. **第四阶段（智能）**：AI 分析层（可回溯）+ AI 自动复盘日报
+
+---
+
+## 七、相比原架构的主要增强
+
+- 新增 **数据中台**：所有上层能力的统一、可回溯数据基础
+- 强化 **复盘中心**：从"系统名"变为真正的核心闭环（每日/个股/交易复盘 + AI 日报 + 情景回放）
+- 新增 **组合与交易日志**：支撑交易复盘与盈亏归因
+- 重构 **AI 分析层**：从"直接给结论"升级为"结论 + 证据链 + 历史胜率"
+- 收敛 **回测/指标/资金曲线**：从散落页面变为服务选股与复盘的引擎
+- 新增 **智能预警**：贯穿全局，减少盯盘成本
+- 新增 **市场情绪指标 / 盘中异动雷达 / 龙虎榜深挖 / 事件日历**

架构总结.md

@@ -0,0 +1,223 @@
+# 智策 StockTerminal — 架构总结
+
+> 版本：当前实现态（2026-06）  
+> 定位：个人/小团队 A 股分析·复盘·智能辅助系统
+
+---
+
+## 一、整体分层
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                      展示层（前端）                          │
+│  HTML + ECharts 5 + 原生 JS                                 │
+│  prototype/index.html · style.css · app.js                  │
+└────────────────────────┬────────────────────────────────────┘
+                         │ HTTP / REST
+┌────────────────────────▼────────────────────────────────────┐
+│                   服务层（后端 FastAPI）                      │
+│  main.py  —— 50+ API 端点，静态文件托管                      │
+│                                                              │
+│  数据服务     akshare_service.py   (行情·情绪·资讯·实时报价)  │
+│  AI 分析      ai.py                (证据链·置信度·LLM/规则)   │
+│  信号统计     signals.py           (胜率回测·预测留痕·核验)   │
+│  RAG 检索     rag.py               (资讯检索·情绪标注)        │
+│  复盘日报     report.py            (结构化日报·推送)          │
+│  组合计算     portfolio.py         (持仓·P&L·资金曲线)        │
+│  回测引擎     backtest.py          (MA交叉·净值曲线)          │
+│  预警引擎     alerts.py            (实时报价判断·触发事件)     │
+│  推送通知     notifier.py          (SMTP·Server酱·企微·PP)   │
+│  大模型客户端 llm.py               (OpenAI 兼容接口)          │
+│  定时调度     scheduler.py         (APScheduler 5 个任务)     │
+│  ETL 入库     ingest.py            (AkShare→PostgreSQL)      │
+│  CLI 工具     cli.py               (init/ingest/ingest_all)  │
+│  配置         config.py + .env     (DB·LLM·推送密钥)         │
+└────────────────────────┬────────────────────────────────────┘
+                         │ SQLAlchemy ORM / psycopg2
+┌────────────────────────▼────────────────────────────────────┐
+│                   数据层（PostgreSQL）                        │
+│  13 张业务表，见下方数据模型                                  │
+└─────────────────────────────────────────────────────────────┘
+```
+
+---
+
+## 二、后端模块职责
+
+| 文件 | 职责 | 关键依赖 |
+|---|---|---|
+| `main.py` | FastAPI 入口，50+ REST 端点，静态托管 | 所有模块 |
+| `config.py` | 环境变量读取，支持 `.env` 文件 | python-dotenv |
+| `db.py` | 引擎/Session，自动建库建表 | SQLAlchemy, psycopg2 |
+| `models.py` | 13 个 ORM 表（见下节） | SQLAlchemy |
+| `akshare_service.py` | AkShare 数据抓取，带 TTL 缓存和 mock 降级 | akshare, cachetools |
+| `ingest.py` | ETL：AkShare → PostgreSQL 增量 upsert | akshare, models |
+| `scheduler.py` | APScheduler 后台任务（5 个定时任务） | apscheduler |
+| `backtest.py` | MA 交叉策略回测，读 DB 日线，输出净值曲线 | sqlalchemy, pandas |
+| `ai.py` | AI 分析：证据链构造、置信度计算、LLM 调用/规则降级 | llm, signals, rag |
+| `signals.py` | 6 类信号历史胜率回测；预测留痕 + 到期核验 | sqlalchemy, pandas, numpy |
+| `rag.py` | 资讯检索 + 利好/利空情绪标注，作为 LLM 上下文 | akshare_service |
+| `report.py` | 七段式 AI 复盘日报生成、落库、推送精简版 | ai, notifier |
+| `portfolio.py` | 持仓计算（移动加权均价）、P&L 归因、逐日资金曲线 | sqlalchemy |
+| `alerts.py` | 实时报价轮询 → 规则命中 → 写事件 + 推送 | akshare_service, notifier |
+| `notifier.py` | 四渠道推送：SMTP 邮件、Server酱、企业微信 Webhook、PushPlus | requests, smtplib |
+| `llm.py` | OpenAI 兼容客户端（DeepSeek/通义/Kimi 均适用） | requests |
+| `cli.py` | 命令行工具：`init` / `ingest` / `ingest_all` | ingest, db |
+
+---
+
+## 三、数据模型（PostgreSQL）
+
+| 表名 | 说明 | 主键 / 唯一约束 |
+|---|---|---|
+| `securities` | 证券基础信息（代码·名称·市场） | `code` |
+| `quotes_daily` | 个股日线（OHLCV，前复权） | `(code, date)` |
+| `index_daily` | 指数日线（上证/深证/沪深300） | `(code, date)` |
+| `sector_daily` | 板块每日快照（涨跌·成交额·龙头） | `(date, name)` |
+| `fund_flow_daily` | 行业主力资金流每日快照 | `(date, name)` |
+| `sentiment_daily` | 全市场情绪（涨跌家数/涨跌停） | `date` |
+| `dragon_tiger` | 龙虎榜明细（代码·席位·净买额） | `(date, code, reason)` |
+| `stock_metrics` | 个股最新因子快照（MA/RSI/MACD/量比/分位/连涨 等 15 个因子） | `code` |
+| `signal_stats` | 6 类技术信号历史胜率（回测样本·N日上涨概率·平均收益） | `(signal, horizon)` |
+| `predictions` | AI 诊断留痕（方向·置信度·N日后核验·命中与否） | `(code, date, kind)` |
+| `daily_reports` | AI 复盘日报（markdown 正文·来源·是否已推送） | `date` |
+| `trades` | 交易记录（买卖·价格·数量·手续费·理由·情绪标签） | `id` |
+| `alert_rules` | 预警规则（价格上穿/下穿/涨跌幅条件·状态） | `id` |
+| `alert_events` | 预警触发事件（站内通知·已读状态） | `id` |
+| `job_runs` | 定时任务执行日志 | `id` |
+
+---
+
+## 四、定时任务（APScheduler，周一至周五）
+
+| 任务 ID | 触发时间 | 功能 |
+|---|---|---|
+| `daily_ingest` | 15:35（收盘后，可配） | 抓取当日板块/资金流/情绪/龙虎榜/自选股日线入库 |
+| `alert_check` | 每 60 秒 | 实时报价核查所有 active 预警规则，触发则写事件并推送 |
+| `daily_report` | 15:45（入库+10分） | 生成 AI 七段式复盘日报并推送微信/邮件 |
+| `verify_pred` | 15:50 | 核验到期 AI 预测，计算命中率 |
+| `signal_stats` | 每周六 09:00 | 对全市场样本股回测 6 类信号历史胜率（可手动触发） |
+
+---
+
+## 五、AI 分析层设计（可回溯）
+
+```
+用户请求 /api/ai/diagnose
+        │
+        ▼
+ _stock_context()           从 stock_metrics + DailyQuote 取最新因子
+        │
+        ├─ signals.get_stats()    读 signal_stats 表，取各信号历史胜率
+        │
+        ├─ rag.stock_context()    拉近期资讯 → 情绪标注（利好/利空/中性）
+        │
+        ▼
+ _build_evidence()          生成 6 维证据链（趋势/技术/动量/资金/位置/消息）
+   每条附：事实描述 · 方向(bull/bear/neutral) · 历史胜率 · 样本数
+        │
+        ▼
+ _confidence_direction()    加权多空净值 → 置信度(%) + 预测方向(up/down/flat)
+        │
+        ├─ signals.record_prediction()   写入 predictions 表留痕
+        │
+        ├─ llm.ask()  (有 key)    将证据链+胜率+RAG资讯构造 prompt → 大模型输出
+        │  或
+        └─ 规则文本降级           格式化证据链为可读文本
+```
+
+实测准确率闭环：
+```
+预测留痕(open) → N 个交易日后 verify_predictions() 拉 DailyQuote 核验
+→ 填写 actual_ret + hit(True/False) → status=closed
+→ /api/ai/accuracy 按方向汇总命中率
+```
+
+---
+
+## 六、推送通知渠道
+
+| 渠道 | 触发条件 | 配置项（`.env`） |
+|---|---|---|
+| SMTP 邮件 | 预警触发 / 复盘日报 / 测试 | `SMTP_HOST/PORT/USER/PASSWORD/TO` |
+| Server酱（微信） | 同上 | `SERVERCHAN_KEY` |
+| 企业微信群机器人 | 同上 | `WECOM_WEBHOOK` |
+| PushPlus（微信） | 同上 | `PUSHPLUS_TOKEN` |
+
+任意渠道配置即自动启用，互不依赖。
+
+---
+
+## 七、前端菜单结构（app.js MENU）
+
+```
+大盘行情
+  ├─ 市场总览       行情/三大指数/情绪温度计/资金流向
+  ├─ 大盘云图       行业/板块涨跌热力图（ECharts treemap）
+  ├─ 热股榜         实时热股+板块
+  └─ 龙虎榜         当日龙虎榜净买额排名
+
+自选股
+  ├─ 自选列表       持仓快照 + K线
+  ├─ 全市场选股     8 个内置策略 + SQL 直查 stock_metrics
+  └─ 条件选股       客户端多因子过滤
+
+复盘中心
+  ├─ 每日复盘       板块/情绪/资金流统计
+  ├─ AI日报         七段式 markdown 日报，历史翻阅，一键推送
+  └─ 个股复盘       K线回放 + MA交叉买卖点标注
+
+策略与中台
+  ├─ 策略回测       MA交叉净值曲线 + 最大回撤
+  └─ 数据中台       入库状态 / 手动触发 / 任务日志
+
+AI 分析
+  ├─ 今日策略       大模型/规则 生成当日操作方向
+  ├─ 个股诊断       6维证据链 + 历史胜率 + 置信度 + RAG资讯
+  ├─ AI复盘点评     当日大模型复盘摘要
+  └─ AI准确率       信号历史胜率表 + 实测命中率 + 近期核验记录
+
+组合交易
+  ├─ 组合持仓       实时市值/浮亏/胜率
+  ├─ 资金曲线       逐日净值 vs 沪深300基准（ECharts折线）
+  ├─ 交易日志       录入买卖 / 理由 / 情绪标签
+  └─ 盈亏归因       按个股/理由/情绪三维归因
+
+智能预警
+  ├─ 预警规则       创建/删除/重激活规则，推送渠道状态检测
+  └─ 触发记录       预警事件列表，标记已读，🔔铃铛角标
+
+资讯中心
+  ├─ 要闻快讯       全球财经资讯 + 情绪标注 + AI摘要
+  └─ 自选相关       自选股关联资讯聚合
+```
+
+---
+
+## 八、技术栈
+
+| 层 | 技术选型 |
+|---|---|
+| 前端 | 纯 HTML + CSS + 原生 JS，ECharts 5（CDN） |
+| 后端 | Python 3.12 + FastAPI 0.115 + uvicorn |
+| 数据库 | PostgreSQL（psycopg2-binary + SQLAlchemy 2.0） |
+| 数据源 | AkShare（新浪/同花顺/乐估备选源），Sina hq 实时报价 |
+| 调度 | APScheduler 3.x BackgroundScheduler |
+| AI | OpenAI 兼容 REST（默认 DeepSeek），规则降级无缝切换 |
+| 缓存 | cachetools TTL（内存），无 Redis 依赖 |
+| 推送 | SMTP(SSL) + Server酱 + 企业微信 Webhook + PushPlus |
+| 部署 | WSL2 / Linux，`nohup python main.py`，可 Docker 化 |
+
+---
+
+## 九、待完善方向
+
+| 优先级 | 模块 | 建议 |
+|---|---|---|
+| 高 | 回测引擎 | 手续费/滑点建模，参数寻优，月度收益热力图 |
+| 高 | 数据稳定性 | 东财源封锁问题，增加 Tushare 备用源，分钟线入库 |
+| 中 | RAG 升级 | Embedding + 向量库（Chroma/Qdrant），支持语义检索财报 |
+| 中 | 多组合 | 目前单一组合，扩展为模拟盘/真实盘多组合管理 |
+| 中 | 盘中异动雷达 | 快速拉升/放量突破/涨停打开实时监测 |
+| 低 | 用户系统 | 登录/鉴权，多用户自选股独立管理 |
+| 低 | 部署文档 | Docker Compose（FastAPI + PostgreSQL），Nginx 反代 |