xiaoguo_scanner: 小果独立扫描线，5分钟一轮，三榜交叉发现

docs/xiaoguo-scanner-design.md

@@ -0,0 +1,78 @@
+# 小果独立扫描线 — 全市场主动发现
+
+## 概述
+
+不依赖趋势信号触发，小果自己盯着各种排行榜，主动发现可能有料的股票，搜新闻判断后喂给知微。
+
+## 时序
+
+```
+每5分钟（独立cron，不碰现有管道）
+小果扫描  → 榜单采样 → 搜索新闻 → LLM判断 → signal_news
+                                           ↓
+知微在下一轮15分钟cron中读到 → 一起分析
+```
+
+## 数据源：三榜交集
+
+每轮同时拉：
+
+| 榜单 | 来源 | 速度 | 内容 |
+|------|------|------|------|
+| 东方财富热榜 | `stock_hot_rank_em()` | <5秒 | 全市场关注度前100，当日 |
+| 同花顺轮流榜 | 以下5个轮流，一轮一个 | ~30秒 | 见下表 |
+
+同花顺轮流拉的榜单（每轮换一个）：
+
+| 榜名 | 函数 | 说明 |
+|------|------|------|
+| 创新高 | `stock_rank_cxg_ths()` | 股价突破N日内新高 |
+| 量价齐升 | `stock_rank_ljqs_ths()` | 成交量+价格同步上涨 |
+| 向上突破 | `stock_rank_xstp_ths()` | 技术形态突破关键位 |
+| 连续上涨 | `stock_rank_cxd_ths()` | 连续N天上涨 |
+| 连续放量 | `stock_rank_cxfl_ths()` | 连续N天放量 |
+
+**为什么这样组合：** 东方财富热榜代表"大家都在看"，同花顺榜单代表"技术面有信号"。一只股票同时上两个榜，比只上一个榜更值得关注。
+
+## 去重策略
+
+每搜完一只股票，记录搜索时间到 `xiaoguo_scan_tracker` 表：
+- 同一股票60分钟内不重复搜索
+- 如果该股票今日已有 signal_news（来源='xiaoguo'），也不再重复
+
+## 有料判断
+
+小果LLM对每篇新闻判断两条：
+
+```
+Q1：这条新闻跟这只股票今天上榜有关吗？
+    → 有关（签大单/出业绩/出政策/被监管/有研报）
+    → 无关（大盘行情、行业普涨、泛泛而谈）
+
+Q2：利好还是利空？
+    → 利好 / 利空 / 中性
+```
+
+两条都有明确结果 → 写入signal_news。
+
+## signal_news 表改动
+
+新增 source 字段：
+
+| source | 含义 |
+|--------|------|
+| trend | 现有管道，由 trend_detector 触发 |
+| xiaoguo | 小果扫描，由榜单发现 |
+
+两类信号在 signal_news 里共存。知微判断时可以看到来源，后续可以用来评估哪个渠道更有效。
+
+## 新增表：xiaoguo_scan_tracker
+
+```sql
+CREATE TABLE IF NOT EXISTS xiaoguo_scan_tracker (
+    code TEXT PRIMARY KEY,
+    name TEXT,
+    last_scanned_at TEXT,
+    found_count INTEGER DEFAULT 0
+);
+```

mofin_db.py

@@ -321,9 +321,24 @@ def init_all_tables(conn: sqlite3.Connection):
             created_at      TEXT DEFAULT (datetime('now','localtime'))
         );
         CREATE INDEX IF NOT EXISTS idx_signal_news_signal ON signal_news(signal_id);
+
+        -- 小果扫描跟踪（去重用）
+        CREATE TABLE IF NOT EXISTS xiaoguo_scan_tracker (
+            code            TEXT PRIMARY KEY,
+            name            TEXT,
+            last_scanned_at TEXT,
+            found_count     INTEGER DEFAULT 0
+        );
     """)
     conn.commit()
 
+    # 迁移：给 signal_news 加 source 字段（幂等）
+    try:
+        conn.execute("ALTER TABLE signal_news ADD COLUMN source TEXT DEFAULT 'trend'")
+    except sqlite3.OperationalError:
+        pass
+    conn.commit()
+
 
 # ═══════════════════════════════════════════════════════════
 # 市场快照写入

xiaoguo_scanner.py

@@ -0,0 +1,236 @@
+#!/usr/bin/env python3
+"""xiaoguo_scanner.py — 小果独立扫描线
+
+每5分钟跑一轮，全市场排行榜主动发现潜在标的。
+不依赖 trend_detector 信号，独立产出到 signal_news。
+"""
+
+import json, os, re, time, urllib.request
+from pathlib import Path
+from datetime import datetime
+
+try:
+    import akshare as ak
+    HAS_AKSHARE = True
+except ImportError:
+    HAS_AKSHARE = False
+
+DATA_DIR = Path(__file__).parent / "data"
+DB_PATH = DATA_DIR / "mofin.db"
+XIAOGUO_API = "http://192.168.1.122:18003/v1/chat/completions"
+XIAOGUO_MODEL = "Qwen3.6-27B-MTPLX-Optimized-Speed"
+SCAN_INTERVAL = 3600  # 同一只股1小时内不重复搜
+MAX_STOCKS_PER_RUN = 15
+ARTICLES_PER_STOCK = 3
+
+# 同花顺轮流榜
+ROTATING_BOARDS = [
+    ("创新高", "stock_rank_cxg_ths"),
+    ("量价齐升", "stock_rank_ljqs_ths"),
+    ("向上突破", "stock_rank_xstp_ths"),
+    ("连续上涨", "stock_rank_cxd_ths"),
+    ("连续放量", "stock_rank_cxfl_ths"),
+]
+
+
+def clean_proxy():
+    for k in ['http_proxy','https_proxy','HTTP_PROXY','HTTPS_PROXY']:
+        os.environ.pop(k, None)
+
+
+def get_conn():
+    import sqlite3
+    conn = sqlite3.connect(str(DB_PATH))
+    conn.row_factory = sqlite3.Row
+    return conn
+
+
+def fetch_hot_board():
+    """东方财富热榜"""
+    if not HAS_AKSHARE:
+        return []
+    try:
+        clean_proxy()
+        df = ak.stock_hot_rank_em()
+        if df is None or len(df) == 0:
+            return []
+        # 东方财富热榜列名变化较大，自动检测
+        cols = list(df.columns)
+        code_candidates = [c for c in cols if any(x in c for x in ['代码', 'code', 'CODE'])]
+        name_candidates = [c for c in cols if any(x in c for x in ['简称', '名称', 'name', 'NAME'])]
+        code_col = code_candidates[0] if code_candidates else cols[1]
+        name_col = name_candidates[0] if name_candidates else cols[2]
+        return [{"code": str(r[code_col]).zfill(6).strip(), "name": str(r[name_col]).strip(),
+                 "rank": i+1, "source": "东方财富热榜"}
+                for i, (_, r) in enumerate(df.head(30).iterrows())]
+    except Exception as e:
+        print(f"  热榜失败: {e}", flush=True)
+    return []
+
+
+def fetch_rotating_board():
+    """同花顺轮流榜（每轮一个）"""
+    if not HAS_AKSHARE:
+        return []
+    conn = get_conn()
+    row = conn.execute("SELECT val FROM state_meta WHERE key='xiaoguo_board_round'").fetchone()
+    round_idx = (int(row[0]) if row else 0) % len(ROTATING_BOARDS)
+    conn.execute("INSERT OR REPLACE INTO state_meta (key, val) VALUES ('xiaoguo_board_round', ?)",
+                 (str((round_idx + 1) % len(ROTATING_BOARDS)),))
+    conn.commit()
+    conn.close()
+
+    board_name, func_name = ROTATING_BOARDS[round_idx]
+    print(f"  同花顺榜: {board_name}", flush=True)
+
+    try:
+        clean_proxy()
+        fn = getattr(ak, func_name)
+        df = fn()
+        cols = list(df.columns)
+        code_col = [c for c in cols if '代码' in c][0]
+        name_col = [c for c in cols if '简称' in c or '名称' in c][0]
+        return [{"code": str(r[code_col]).zfill(6), "name": str(r[name_col]).strip(),
+                 "source": f"同花顺{board_name}"}
+                for _, r in df.head(15).iterrows()]
+    except Exception as e:
+        print(f"  {board_name}失败: {e}", flush=True)
+    return []
+
+
+def get_scanned_codes(conn):
+    """取1小时内已扫描过的代码"""
+    rows = conn.execute(
+        "SELECT code FROM xiaoguo_scan_tracker WHERE datetime(last_scanned_at) > datetime('now', '-1 hour')"
+    ).fetchall()
+    return {r[0] for r in rows}
+
+
+def mark_scanned(conn, code, name, found):
+    conn.execute(
+        "INSERT OR REPLACE INTO xiaoguo_scan_tracker (code, name, last_scanned_at, found_count) "
+        "VALUES (?, ?, datetime('now','localtime'), COALESCE((SELECT found_count FROM xiaoguo_scan_tracker WHERE code=?),0)+?)",
+        (code, name, code, 1 if found else 0)
+    )
+    conn.commit()
+
+
+def search_news(code, max_results=3):
+    """akshare搜个股新闻"""
+    articles = []
+    if not HAS_AKSHARE:
+        return articles
+    try:
+        clean_proxy()
+        df = ak.stock_news_em(symbol=code)
+        for _, r in df.head(max_results).iterrows():
+            title = r.get('新闻标题', '')
+            content = r.get('新闻内容', '')
+            if title and len(title) > 5:
+                articles.append({"title": title, "content": content})
+    except:
+        pass
+    return articles
+
+
+def has_substance(title, content):
+    """小果LLM判断是否有料（返回 True/False + 情感）"""
+    text = title + (content or '')[:100]
+    prompt = f"""新闻：{text}
+问：这条新闻跟该股今日上榜有关吗？
+回答格式：{"有关":"利好|利空|中性"} 或 "无关"
+回答："""
+
+    payload = json.dumps({
+        "model": XIAOGUO_MODEL,
+        "messages": [{"role": "user", "content": prompt}],
+        "temperature": 0.1, "max_tokens": 100,
+    }).encode()
+
+    opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
+    req = urllib.request.Request(XIAOGUO_API, data=payload,
+                                 headers={"Content-Type": "application/json"}, method="POST")
+    try:
+        resp = opener.open(req, timeout=30)
+        reply = json.loads(resp.read())["choices"][0]["message"]["content"]
+        if "有关" in reply:
+            for s in ["利好", "利空", "中性"]:
+                if s in reply:
+                    return True, s
+            return True, "中性"
+    except:
+        pass
+    return False, None
+
+
+def main():
+    start_time = time.time()
+    conn = get_conn()
+
+    # 1. 拉榜
+    hot = fetch_hot_board()
+    rotating = fetch_rotating_board()
+    elapsed = time.time() - start_time
+    print(f"榜单: 东方财富{len(hot)}只, 同花顺{len(rotating)}只 ({elapsed:.0f}s)", flush=True)
+
+    if not hot and not rotating:
+        conn.close()
+        return
+
+    # 2. 合并去重
+    all_stocks = {}
+    for s in hot + rotating:
+        code = s["code"]
+        if code not in all_stocks:
+            all_stocks[code] = {"code": code, "name": s["name"], "sources": []}
+        all_stocks[code]["sources"].append(s["source"])
+
+    # 3. 排除已搜索过的
+    scanned = get_scanned_codes(conn)
+    candidates = [s for code, s in all_stocks.items()
+                  if code not in scanned and len(code) == 6 and code.isdigit()][:MAX_STOCKS_PER_RUN]
+
+    if not candidates:
+        print(f"无新候选（已有 {len(scanned)} 只已扫描）", flush=True)
+        conn.close()
+        return
+
+    print(f"待扫描: {len(candidates)} 只（跳过 {len(all_stocks)-len(candidates)} 只已扫过）", flush=True)
+
+    # 4. 逐只搜新闻+判断
+    found_any = False
+    for stock in candidates:
+        code, name = stock["code"], stock["name"]
+        sources = "|".join(stock["sources"])
+
+        articles = search_news(code, ARTICLES_PER_STOCK)
+        if not articles:
+            mark_scanned(conn, code, name, False)
+            continue
+
+        has_found = False
+        for art in articles:
+            ok, sentiment = has_substance(art["title"], art.get("content", ""))
+            if ok:
+                # 有料 → 写入signal_news
+                conn.execute(
+                    "INSERT INTO signal_news (signal_id, sector, overall_sentiment, summary, key_articles, searched_stocks, source) "
+                    "VALUES (NULL, ?, ?, ?, ?, ?, 'xiaoguo')",
+                    (f"扫描-{name}", sentiment, f"[{sources}] {art['title'][:80]}",
+                     json.dumps([{"title": art["title"], "sentiment": sentiment, "summary": art.get("content","")[:100]}], ensure_ascii=False),
+                     json.dumps([name], ensure_ascii=False))
+                )
+                has_found = True
+                found_any = True
+                print(f"  ✅ {name}({code}) [{sources}] {sentiment}: {art['title'][:50]}", flush=True)
+                break  # 一只股有一条有料就够了
+
+        mark_scanned(conn, code, name, has_found)
+
+    total_time = time.time() - start_time
+    print(f"完成: {len(candidates)}只扫描, {'有发现' if found_any else '无发现'} ({total_time:.0f}s)", flush=True)
+    conn.close()
+
+
+if __name__ == "__main__":
+    main()