复盘汇总加相对主榜、相对同市值桶两列超额，最大单日样本占比与样本纪律三档标注，候选单逐日表

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

plan_review.py

@@ -150,14 +150,24 @@ def _md(df: pd.DataFrame) -> str:
     return "\n".join(lines)
 
 
-def summarize(ret: pd.Series, codes: list[str], base_all: float, base_main: float) -> dict | None:
+def summarize(ret: pd.Series, codes: list[str], base_all: float, base_main: float,
+              caps: dict | None = None, cap_base: dict | None = None) -> dict | None:
+    """一份名单在一个计划日、一个期限上的读数。
+    excess_all 相对全池等权；excess_main 相对主榜等权（同为有券商覆盖的篮子，剥掉大票对小票的贝塔）；
+    excess_cap 相对同市值桶均值（每票减当日同桶全池均值再平均，方案 1.8c 的市值中性口径）。"""
     r = ret.reindex([c for c in codes if c in ret.index]).dropna()
     if r.empty:
         return None
-    return {"n": int(len(r)), "ret": round(float(r.mean()), 2),
+    out = {"n": int(len(r)), "ret": round(float(r.mean()), 2),
            "excess_all": round(float(r.mean() - base_all), 2),
            "excess_main": round(float(r.mean() - base_main), 2) if base_main is not None else None,
+           "excess_cap": None,
            "beat": round(float((r > base_all).mean() * 100), 1)}
+    if caps and cap_base:
+        adj = [float(v) - cap_base[caps[c]] for c, v in r.items() if caps.get(c) in cap_base]
+        if adj:
+            out["excess_cap"] = round(sum(adj) / len(adj), 2)
+    return out
 
 
 # ============================================================================
@@ -208,6 +218,12 @@ def run(since: str, until: str | None, horizons: tuple, start: str, out_dir: str
             base_all = float(ret.dropna().mean())
             main_ret = ret.reindex([c for c in main_codes if c in ret.index]).dropna()
             base_main = float(main_ret.mean()) if not main_ret.empty else None
+            cap_base: dict = {}                 # 当日三个市值桶各自的全池均值，供 excess_cap 用
+            if caps:
+                for cap in ("小盘", "中盘", "大盘"):
+                    v = ret.reindex([c for c, b in caps.items() if b == cap and c in ret.index]).dropna()
+                    if not v.empty:
+                        cap_base[cap] = float(v.mean())
             regime_post = "涨周" if base_all > 0 else "跌周"
             regime_pre = (("弱势日" if reg.get("weak_day") else "非弱势日")
                           if reg and reg.get("weak_day") is not None else "无标签")
@@ -217,20 +233,20 @@ def run(since: str, until: str | None, horizons: tuple, start: str, out_dir: str
                 "全池等权": list(ret.dropna().index),
             }
             for name, codes in lists.items():
-                s = summarize(ret, codes, base_all, base_main)
+                s = summarize(ret, codes, base_all, base_main, caps, cap_base)
                 if s:
                     rows.append({"date": day, "h": h, "list": name, "group": "全部",
                                  "regime_post": regime_post, "regime_pre": regime_pre, **s})
             # 分组：档位、判决、吸筹三态、市值
             for tier in ("强传导", "弱传导", "无传导"):
                 codes = [c for c, t in tier_of.items() if t == tier]
-                s = summarize(ret, codes, base_all, base_main)
+                s = summarize(ret, codes, base_all, base_main, caps, cap_base)
                 if s:
                     rows.append({"date": day, "h": h, "list": "主榜", "group": f"档位={tier}",
                                  "regime_post": regime_post, "regime_pre": regime_pre, **s})
             for v in ("候选", "关注", "仅展示"):
                 codes = [c for c, vv in verdict_of.items() if vv == v]
-                s = summarize(ret, codes, base_all, base_main)
+                s = summarize(ret, codes, base_all, base_main, caps, cap_base)
                 if s:
                     rows.append({"date": day, "h": h, "list": "档位表", "group": f"判决={v}",
                                  "regime_post": regime_post, "regime_pre": regime_pre, **s})
@@ -238,14 +254,14 @@ def run(since: str, until: str | None, horizons: tuple, start: str, out_dir: str
                                    ("潜在吸筹", lambda s: str(s).startswith("潜在")),
                                    ("其他", lambda s: not (str(s).startswith("明确") or str(s).startswith("潜在")))):
                 codes = [c for c, s in acc.items() if s and pred(s)]
-                s = summarize(ret, codes, base_all, base_main)
+                s = summarize(ret, codes, base_all, base_main, caps, cap_base)
                 if s:
                     rows.append({"date": day, "h": h, "list": "全池", "group": f"吸筹={st_label}",
                                  "regime_post": regime_post, "regime_pre": regime_pre, **s})
             if caps:
                 for cap in ("小盘", "中盘", "大盘"):
                     codes = [c for c in cands if caps.get(c) == cap]
-                    s = summarize(ret, codes, base_all, base_main)
+                    s = summarize(ret, codes, base_all, base_main, caps, cap_base)
                     if s:
                         rows.append({"date": day, "h": h, "list": "候选单", "group": f"市值={cap}",
                                      "regime_post": regime_post, "regime_pre": regime_pre, **s})
@@ -266,10 +282,17 @@ def run(since: str, until: str | None, horizons: tuple, start: str, out_dir: str
         # 带 _w 的三列按样本加权（大名单的日子权重大），与方案 1.8b/1.8c 手工读数同口径，只作对账。
         n = g["n"].astype(float)
         w = n / n.sum() if n.sum() else n
-        return pd.Series({"days": g["date"].nunique(), "n": int(n.sum()),
+        days, total = g["date"].nunique(), int(n.sum())
+        # 样本纪律（方案 2.4）机器标注：可读=计划日≥20 且样本≥100；方向=样本≥100；其余不作数
+        grade = "可读" if (days >= 20 and total >= 100) else ("方向" if total >= 100 else "不作数")
+        return pd.Series({"days": days, "n": total,
                           "ret": round(g["ret"].mean(), 2),
                           "excess_all": round(g["excess_all"].mean(), 2),
+                          "excess_main": round(g["excess_main"].mean(), 2) if g["excess_main"].notna().any() else None,
+                          "excess_cap": round(g["excess_cap"].mean(), 2) if g["excess_cap"].notna().any() else None,
                           "beat": round(g["beat"].mean(), 1),
+                          "share": round(float(n.max() / n.sum() * 100), 0) if n.sum() else None,
+                          "grade": grade,
                           "ret_w": round(float((g["ret"] * w).sum()), 2),
                           "excess_w": round(float((g["excess_all"] * w).sum()), 2),
                           "beat_w": round(float((g["beat"] * w).sum()), 1)})
@@ -278,14 +301,23 @@ def run(since: str, until: str | None, horizons: tuple, start: str, out_dir: str
     regime_tbl = df[df["group"] == "全部"].groupby(["regime_post", "list", "h"]).apply(agg).reset_index()
     pre_tbl = df[(df["group"] == "全部") & (df["regime_pre"] != "无标签")] \
         .groupby(["regime_pre", "list", "h"]).apply(agg).reset_index()
+    h0 = int(df["h"].min())
+    daily_c = df[(df["list"] == "候选单") & (df["group"] == "全部") & (df["h"] == h0)] \
+        [["date", "n", "ret", "excess_all", "excess_cap", "beat", "regime_post"]].sort_values("date")
 
     md = [f"# 候选单复盘 · {days_plan[0]} 至 {stamp}（起点价 {start}）", "",
           f"计划日 {len(days_plan)} 期；样本纪律：分组样本少于一百只或覆盖计划日少于二十个只看方向；"
           f"口径切换的总样本少于六十个计划日只写方向。", "",
-          "## 一、六份名单 × 期限（超额=相对全池等权，命中=跑赢全池比例；"
-          "不带后缀的列按计划日等权，带 _w 的列按样本加权，后者只用于与方案第一节手工读数对账）", "",
+          "## 一、六份名单 × 期限", "",
+          "列的读法：excess_all 相对全池等权，excess_main 相对主榜等权（剥掉大票对小票的贝塔），"
+          "excess_cap 相对同市值桶（成交额除以换手率三分位，方案 1.8c 口径）；beat 是跑赢全池比例，"
+          "全池自己的 beat 只有四成多（分布右偏），读它时与全池等权那一行相减；share 是最大单日样本占比"
+          "（超过三成说明读数由一天主导，只看逐日表）；grade 是样本纪律三档。"
+          "不带后缀的列按计划日等权（主口径），带 _w 的列按样本加权，只用于与方案第一节手工读数对账。", "",
           _md(lists_tbl), "",
           "## 二、分组读数", "", _md(groups_tbl), "",
+          f"## 二之二、候选单逐日（期限 {h0} 日；第一节按日等权的读数就是这张表的平均，看集中度）", "",
+          _md(daily_c), "",
           "## 三、按事后环境分组（未来 h 日全池涨跌，只作解释，不作交易前置）", "",
           _md(regime_tbl), ""]
     if not pre_tbl.empty: