diff --git a/.gitea/workflows/deploy-production.yaml b/.gitea/workflows/deploy-production.yaml
index 938b5b8..4e8cad5 100644
--- a/.gitea/workflows/deploy-production.yaml
+++ b/.gitea/workflows/deploy-production.yaml
@@ -8,7 +8,7 @@ on:
 
 jobs:
   deploy:
-    runs-on: ubuntu-latest
+    runs-on: tencent-prod
     timeout-minutes: 30
 
     env:
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/check.jsonl b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/check.jsonl
new file mode 100644
index 0000000..9dd3234
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/check.jsonl
@@ -0,0 +1 @@
+{"_example": "Fill with {\"file\": \"<path>\", \"reason\": \"<why>\"}. Put spec/research files only — no code paths. Run `python3 .trellis/scripts/get_context.py --mode packages` to list available specs. Delete this line once real entries are added."}
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/design.md b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/design.md
new file mode 100644
index 0000000..ab3cc71
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/design.md
@@ -0,0 +1,9 @@
+# 研究设计
+
+这是只读分析任务，不进入产品实现阶段。
+
+公开证据链：页面展示 → 实际引用脚本 → 实际请求的公开数据 → 评分与排名字段。数据库证据链：库表目录 → 字段及单位 → 重叠交易日原始值 → 候选公式复算 → 与公开评分比较。
+
+优先检验可解释的低自由度公式。分开检验资金比率、横截面排序/归一化、时间窗口聚合；用多日和不同板块类型验证，避免单点拟合。识别数据修订、单位换算、成分股聚合与板块原始数据的口径差异。
+
+数据库连接强制 default_transaction_read_only，设置查询超时。仅在本机保留任务所需数据；公开资料可以缓存供复核，凭据不落盘。报告不将无法唯一识别的参数写成确定结论。
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.jsonl b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.jsonl
new file mode 100644
index 0000000..9dd3234
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.jsonl
@@ -0,0 +1 @@
+{"_example": "Fill with {\"file\": \"<path>\", \"reason\": \"<why>\"}. Put spec/research files only — no code paths. Run `python3 .trellis/scripts/get_context.py --mode packages` to list available specs. Delete this line once real entries are added."}
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.md b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.md
new file mode 100644
index 0000000..14cf3a2
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/implement.md
@@ -0,0 +1,9 @@
+# 分析步骤（无产品实现）
+
+- [x] 读取网站公开脚本和数据，记录字段、日期和来源。
+- [x] 只读确认数据库版本、数据表及覆盖范围。
+- [x] 建立日期、板块和单位映射，对比原始输入。
+- [x] 逐层验证单日评分和波段评分候选公式，记录误差。
+- [x] 核验关键结论，完成研究记录与用户答复。
+
+验证使用实际数据计算与证据核对；不运行与分析无关的产品测试。任务不包含代码实施、共享知识推广、提交和发布。
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/prd.md b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/prd.md
new file mode 100644
index 0000000..d7a4d82
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/prd.md
@@ -0,0 +1,25 @@
+# 还原 OneChart 波段与单日资金流评分
+
+## Goal
+
+分析 https://onechartlab.com/ 板块资金雷达的波段流入率、单日流入率及其加权评分，结合用户本机 PostgreSQL 数据给出可复核的公式证据、复算结果和不确定性。
+
+## Requirements
+
+- 区分原始资金比率、评分和排名，明确时间窗口、权重、标准化、排名池和缺失数据规则。
+- 优先读取网站实际公开的 HTML、脚本和数据；不将本项目独立指标策略视为该站点真实公式。
+- 数据库仅使用只读连接及有范围限制的 SELECT，先确认可用库、表、字段与日期。
+- 使用相同日期、板块标识和数据口径进行多样本验证，记录误差及候选公式可识别性。
+- 用户已同意创建任务并记录分析。只修改当前任务记录，不修改产品代码、数据库或共享规格，不提交或发布。
+- 凭据不写入任务记录、研究脚本、结果文件或报告；本机数据不传给外部服务。
+
+## Acceptance Criteria
+
+- [x] 列出评分相关公开字段及来源，说明公式是否直接公开。
+- [x] 给出单日与波段评分的可验证公式，或明确最有依据的候选公式和未解决参数。
+- [x] 用数据库与网站重叠样本核验，报告样本范围、误差和差异原因。
+- [x] 保存必要分析记录与可复算证据，最终回答清楚区分事实、推断和限制。
+
+## 结果
+
+公开原始字段可精确重现最近 12 日 9,492 条记录，两种评分最大绝对误差约 3.41e-13；数据库最新日原始快照试算平均误差为单日 0.9783 分、波段 1.9076 分，个别板块仍有较大输入差异。详见 `research/findings.md` 和 `research/database-validation.md`。本研究已完成，没有产品实施待批准。
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.json
new file mode 100644
index 0000000..63a0dc2
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.json
@@ -0,0 +1,128 @@
+{
+  "normalized_aggregate": {
+    "joined_rows": 12654,
+    "joined_dates": 16,
+    "latest_rows": 791,
+    "Ratio": {
+      "checked_rows": 791,
+      "mean_absolute_error": 4.66428796165067,
+      "max_absolute_error": 129.76729492953035,
+      "same_one_decimal_display": 204,
+      "same_final_rank": 215
+    },
+    "Swing": {
+      "checked_rows": 791,
+      "mean_absolute_error": 4.803044287121097,
+      "max_absolute_error": 179.7496672672861,
+      "same_one_decimal_display": 96,
+      "same_final_rank": 199
+    },
+    "turnover_within_1_01_yuan": 294,
+    "weight_within_1e_10": 280,
+    "examples": [
+      {
+        "ts_code": "BK0581.DC",
+        "index_name": "智能电网 (概念)",
+        "ratio_db": 0.01288923245126,
+        "weight_db": 1.08086957077924,
+        "pred_Ratio_Score": 684.0285689472486,
+        "Ratio_Score": 789.4114202884311,
+        "pred_Swing_Score": 386.3978175732549,
+        "Swing_Score": 433.9148866486079
+      },
+      {
+        "ts_code": "BK0615.DC",
+        "index_name": "中药概念 (概念)",
+        "ratio_db": 0.080175906138025,
+        "weight_db": 1.041522634544339,
+        "pred_Ratio_Score": 1036.4911242325306,
+        "Ratio_Score": 1036.9806099966886,
+        "pred_Swing_Score": 825.1676911365778,
+        "Swing_Score": 823.0404356041679
+      },
+      {
+        "ts_code": "BK0653.DC",
+        "index_name": "养老概念 (概念)",
+        "ratio_db": 0.065820711061371,
+        "weight_db": 1.050692084122948,
+        "pred_Ratio_Score": 1038.0025661987577,
+        "Ratio_Score": 1035.4711369170884,
+        "pred_Swing_Score": 1005.0098195958633,
+        "Swing_Score": 1005.0161034783504
+      },
+      {
+        "ts_code": "BK1657.DC",
+        "index_name": "病原体防治 (概念)",
+        "ratio_db": 0.064807487656449,
+        "weight_db": 1.05614239070725,
+        "pred_Ratio_Score": 1040.8359792477243,
+        "Ratio_Score": 1038.383599887465,
+        "pred_Swing_Score": 829.0972873909569,
+        "Swing_Score": 830.4517488043484
+      }
+    ]
+  },
+  "raw_snapshot_reaggregation": {
+    "joined_rows": 8701,
+    "joined_dates": 11,
+    "latest_rows": 791,
+    "Ratio": {
+      "checked_rows": 791,
+      "mean_absolute_error": 0.9783318452555938,
+      "max_absolute_error": 105.32457821281037,
+      "same_one_decimal_display": 581,
+      "same_final_rank": 576
+    },
+    "Swing": {
+      "checked_rows": 791,
+      "mean_absolute_error": 1.9075528008117222,
+      "max_absolute_error": 150.16516952571226,
+      "same_one_decimal_display": 318,
+      "same_final_rank": 363
+    },
+    "turnover_within_1_01_yuan": 610,
+    "weight_within_1e_10": 496,
+    "examples": [
+      {
+        "ts_code": "BK0581.DC",
+        "index_name": "智能电网 (概念)",
+        "ratio_db": 0.012878069596386,
+        "weight_db": 1.080961651218729,
+        "pred_Ratio_Score": 684.0868420756208,
+        "Ratio_Score": 789.4114202884311,
+        "pred_Swing_Score": 387.73624445889186,
+        "Swing_Score": 433.9148866486079
+      },
+      {
+        "ts_code": "BK0615.DC",
+        "index_name": "中药概念 (概念)",
+        "ratio_db": 0.079794956978515,
+        "weight_db": 1.041976772408121,
+        "pred_Ratio_Score": 1036.9430681935887,
+        "Ratio_Score": 1036.9806099966886,
+        "pred_Swing_Score": 823.0106390759795,
+        "Swing_Score": 823.0404356041679
+      },
+      {
+        "ts_code": "BK0653.DC",
+        "index_name": "养老概念 (概念)",
+        "ratio_db": 0.065820711061371,
+        "weight_db": 1.050692084122948,
+        "pred_Ratio_Score": 1035.4646626139197,
+        "Ratio_Score": 1035.4711369170884,
+        "pred_Swing_Score": 1005.0098195958633,
+        "Swing_Score": 1005.0161034783504
+      },
+      {
+        "ts_code": "BK1657.DC",
+        "index_name": "病原体防治 (概念)",
+        "ratio_db": 0.064737394636734,
+        "weight_db": 1.05622244732493,
+        "pred_Ratio_Score": 1038.3636136745088,
+        "Ratio_Score": 1038.383599887465,
+        "pred_Swing_Score": 829.160133769571,
+        "Swing_Score": 830.4517488043484
+      }
+    ]
+  }
+}
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.md b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.md
new file mode 100644
index 0000000..35daa42
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database-validation.md
@@ -0,0 +1,76 @@
+# PostgreSQL 独立核验
+
+## 查询与范围
+
+数据库：`zhixing-system`，PostgreSQL 18.4。本次通过用户授权的 SSH 隧道访问；连接参数强制 `default_transaction_read_only=on`、`statement_timeout=30000`、`lock_timeout=2000`，实测 `transaction_read_only=on`。只执行目录检查和 SELECT/CTE；没有数据库写入。凭据和私钥内容不保存在任务文件中。
+
+读取了以下数据：
+
+- `sector_radar_publication`：为每个交易日选取最近成功发布。
+- `sector_radar_daily_aggregate`：2026-08-28 至 2026-09-21，16 个有成功发布的交易日、16,000 行。9 月 7 日没有成功发布对应的汇总，因此未用别日汇总冒充。
+- `sector_radar_source_snapshot`：保存的原始 `dc_member`、`daily`、`moneyflow_dc`。对 2026-09-07 至 2026-09-21 的 11 日重新聚合，共 11,000 个板块日；SQL 保存在 `raw-reaggregation.sql`。
+- 小范围检查 `sector_radar_stock_fact` 状态，以及四个板块的当日成员快照。
+
+原始快照重聚合按日期和分区选择最近观测，成员使用逐板块分区，股票字段按日期/代码去重；`daily.amount × 1000` 与 `moneyflow_dc.net_amount × 10000` 统一为元。原始重聚合没有沿用产品的沪深 A 股过滤，目的仅是调查网站口径，未改变产品规则。
+
+股票名单、金额数据仅在本机处理，没有传给外部文档查询或其他服务。
+
+## 验证设计
+
+对齐网站实际的日期、板块代码与类型，使用数据库提供的净额与成交额独立计算 r、3/10 日均值及 5 日成交额权重。预测阶段不使用网站的比率、分数或最终排名。
+
+本次试算将数据库聚合成交额向下取整到元，再使用 `r = 净额/(成交额+100)`、`W = log10(1+MA5(成交额))/10`。这是与公开数值关系一致的候选输入口径，不能把该试算本身当作后端代码证据。
+
+计算百分位前特意限制到网站的板块池。数据库有概念 504、行业 496，共 1000 个板块；网站是概念 414、行业 377，共 791 个。即使原始金额相同，在不同 N 和不同成员的池中计算百分位也不能复刻网站分数。
+
+## 最新日结果
+
+2026-09-21 共 791 条对齐记录：
+
+| 数据输入 | 单日评分 MAE | 单日最大误差 | 单日一位小数一致 | 波段评分 MAE | 波段最大误差 | 波段一位小数一致 |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| 当前产品规范化汇总 | 4.6643 | 129.7673 | 204/791 | 4.8030 | 179.7497 | 96/791 |
+| 库中原始快照重新聚合 | 0.9783 | 105.3246 | 581/791 | 1.9076 | 150.1652 | 318/791 |
+
+原始快照重聚合后，610/791 个最新日成交额与从网站金额/比率反求的成交额相差不超过 1.01 元；496/791 个近五日成交额权重达到 `1e-10` 内一致。这为“近五日成交额取对数作为权重”提供了不依赖网站评分预测输入的数据库佐证。
+
+实际例子：
+
+| 板块 | 数据库单日复算 | 网站单日分数 | 数据库波段复算 | 网站波段分数 |
+| --- | ---: | ---: | ---: | ---: |
+| 中药概念 | 1036.9431 | 1036.9806 | 823.0106 | 823.0404 |
+| 养老概念 | 1035.4647 | 1035.4711 | 1005.0098 | 1005.0161 |
+| 病原体防治 | 1038.3636 | 1038.3836 | 829.1601 | 830.4517 |
+| 智能电网 | 684.0868 | 789.4114 | 387.7362 | 433.9149 |
+
+不能只报告平均误差而忽略个别大偏差；数据库原始快照尚未逐板块精确复刻网站输入。本次请求中的评分机制已完成数值还原，但输入采集、板块池和成员版本的完整复刻属于进一步工作。
+
+## 已确认的输入差异
+
+1. **股票范围不同。** 产品 `normalize.py:242` 起要求当前上市、沪深证券，排除北交所/B 股；`facts.py:72` 起只将 `AVAILABLE` 股票累加到板块金额。最新日事实中有 `lifecycle_invalid=435`、`suspended=12`、`available=5209`。改用原始快照后误差显著减少，但没有完全消失。
+2. **板块池不同。** 1000 与 791 的差异已在上述比较中控制；真正独立生产还需要明确网站选择这 791 个板块的规则。
+3. **公开成员表与数据库当日快照不同。** 按股票代码去除交易所后缀再比较：
+
+| 板块 | 库内成员 | 网站公开成员 | 交集 |
+| --- | ---: | ---: | ---: |
+| 智能电网 | 197 | 195 | 190 |
+| 中药概念 | 146 | 145 | 144 |
+| 碳交易 | 142 | 138 | 137 |
+| 超跌股 | 167 | 22 | 7 |
+
+智能电网库内独有 `002851/003043/301236/301669/605336/688187/920222`；网站公开表独有 `001388/002063/300140/600522/920375`。完整差集见 `member-differences.json`。网站这份 `CONSTITUENT_MAP` 不是按交易日分片的历史成员证据，不能据此认定所有历史评分都使用同一份名单；这里证明的是输入版本确实存在差异，不宣称它解释了每一分残差。
+
+4. **上游净额也不完全一致。** 即使成交额近似一致，个股资金流按万元提供的小数精度、板块级净额来源、成员与观测时点仍可能造成净额差异。网站页脚同时提及东财板块日线资金流；本数据库没有对应 `moneyflow_ind_dc` 快照，不能将两种来源强行视为逐值相同。
+5. **9 月 7 日原始资金流不完整。** 该日重聚合样本的资金流覆盖明显不足，不把它用于声称全部 11 日的评分准确度；最终 9 月 21 日的最近十日窗口从 9 月 8 日开始。
+
+第 1–3 项有直接目录、代码和数值证据；第 4 项中的具体上游精度与发布时间机制尚未取得网站构建端证据，因此保留为差异候选原因。
+
+## 保存与复现
+
+- `db-raw-inputs.json.gz`：原始快照重聚合结果。
+- `db-normalized-inputs.json.gz`：当前产品汇总，作为对照。
+- `db-source-metadata.json`：库版本、只读设置及各数据源覆盖范围。
+- `database_reproduce.py`：只读取固定文件，在对齐的排名池中独立计算。
+- `database-validation.json`：精确误差、显示与排名一致数量。
+
+运行 `database_reproduce.py` 不需要数据库凭据或在线连接。所有产物限于当前 Trellis 任务；未修改产品实现、共享规格或生产数据。
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database_reproduce.py b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database_reproduce.py
new file mode 100644
index 0000000..d5c2fff
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/database_reproduce.py
@@ -0,0 +1,72 @@
+"""使用已读取的 PostgreSQL 金额快照独立算分，无数据库连接和凭据。"""
+
+import gzip
+import json
+
+import numpy as np
+import pandas as pd
+
+from reproduce import ROOT
+
+
+def load(name: str) -> list[dict]:
+    return json.loads(gzip.decompress((ROOT / name).read_bytes()))["rows"]
+
+
+def compare(public: pd.DataFrame, rows: list[dict], normalized: bool) -> dict:
+    """对齐网站实际排名池；评分只使用库内净额和成交额构造。"""
+    source = pd.DataFrame(rows)
+    source["ts_code"] = source["sector_code"]
+    keys = ["trade_date", "ts_code"]
+    if normalized:
+        source["type"] = source["sector_type"].map({"concept": "概念板块", "industry": "行业板块"})
+        keys.append("type")
+    for field in ["net_amount_yuan", "turnover_yuan"]:
+        source[field] = pd.to_numeric(source[field])
+    frame = public.merge(source, on=keys, validate="one_to_one").sort_values(["ts_code", "trade_date"])
+    frame["amount_db"] = np.floor(frame["turnover_yuan"])
+    frame["ratio_db"] = frame["net_amount_yuan"] / (frame["amount_db"] + 100)
+    for field, windows in [("amount_db", [5]), ("ratio_db", [3, 10])]:
+        for window in windows:
+            frame[f"{field}{window}"] = frame.groupby("ts_code")[field].transform(
+                lambda values: values.rolling(window, min_periods=window).mean()
+            )
+    frame["weight_db"] = np.log10(frame["amount_db5"] + 1) / 10
+    groups = frame.groupby(["trade_date", "type"])
+    frame["pred_Ratio_Score"] = 1000 * groups["ratio_db"].rank(pct=True) * frame["weight_db"]
+    frame["pred_Swing_Score"] = (
+        500 * (groups["ratio_db3"].rank(pct=True) + groups["ratio_db10"].rank(pct=True))
+        * frame["weight_db"]
+    )
+    latest = frame[frame["trade_date"].eq("2026-09-21")].copy()
+    result = {"joined_rows": len(frame), "joined_dates": frame["trade_date"].nunique(), "latest_rows": len(latest)}
+    for metric in ["Ratio", "Swing"]:
+        expected, prediction = f"{metric}_Score", f"pred_{metric}_Score"
+        errors = (latest[prediction] - latest[expected]).abs()
+        ranks = latest.groupby("type")[prediction].rank(ascending=False)
+        result[metric] = {
+            "checked_rows": int(errors.notna().sum()), "mean_absolute_error": errors.mean(),
+            "max_absolute_error": errors.max(),
+            "same_one_decimal_display": int(latest[prediction].round(1).eq(latest[expected].round(1)).sum()),
+            "same_final_rank": int(ranks.eq(latest[f"{metric}_RankPos"]).sum()),
+        }
+    # 从网站两个原始字段得到的成交额仅用于末端验证，不参与库内评分预测。
+    target_turnover = latest["Amount_Raw_BN"] * 1e8 / latest["Ratio_Raw_Pct"] - 100
+    result["turnover_within_1_01_yuan"] = int(latest["turnover_yuan"].sub(target_turnover).abs().lt(1.01).sum())
+    website_weight = latest["Ratio_Score"] / (1000 * latest.groupby("type")["Ratio_Raw_Pct"].rank(pct=True))
+    result["weight_within_1e_10"] = int(latest["weight_db"].sub(website_weight).abs().lt(1e-10).sum())
+    examples = latest[latest["ts_code"].isin(["BK0615.DC", "BK0653.DC", "BK1657.DC", "BK0581.DC"])][[
+        "ts_code", "index_name", "ratio_db", "weight_db", "pred_Ratio_Score", "Ratio_Score", "pred_Swing_Score", "Swing_Score"
+    ]]
+    result["examples"] = json.loads(examples.to_json(orient="records", force_ascii=False, double_precision=15))
+    return result
+
+
+if __name__ == "__main__":
+    public = pd.DataFrame(load("public-inputs.json.gz"))
+    result = {
+        "normalized_aggregate": compare(public, load("db-normalized-inputs.json.gz"), True),
+        "raw_snapshot_reaggregation": compare(public, load("db-raw-inputs.json.gz"), False),
+    }
+    (ROOT / "database-validation.json").write_text(json.dumps(result, ensure_ascii=False, indent=2) + "\n")
+    print(json.dumps(result, ensure_ascii=False, indent=2))
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-normalized-inputs.json.gz b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-normalized-inputs.json.gz
new file mode 100644
index 0000000..c2f7ebd
Binary files /dev/null and b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-normalized-inputs.json.gz differ
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-raw-inputs.json.gz b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-raw-inputs.json.gz
new file mode 100644
index 0000000..d29be18
Binary files /dev/null and b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-raw-inputs.json.gz differ
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-source-metadata.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-source-metadata.json
new file mode 100644
index 0000000..6fc472f
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/db-source-metadata.json
@@ -0,0 +1,55 @@
+{
+  "postgres_version": "18.4",
+  "transaction_read_only": "on",
+  "sources": [
+    {
+      "api_name": "daily",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 17,
+      "source_rows": 94336
+    },
+    {
+      "api_name": "dc_index",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 34,
+      "source_rows": 17000
+    },
+    {
+      "api_name": "dc_member",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 17105,
+      "source_rows": 1678901
+    },
+    {
+      "api_name": "moneyflow",
+      "first_date": "2026-09-07",
+      "last_date": "2026-09-21",
+      "snapshot_count": 11,
+      "source_rows": 61054
+    },
+    {
+      "api_name": "moneyflow_dc",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 102,
+      "source_rows": 101285
+    },
+    {
+      "api_name": "suspend_d",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 17,
+      "source_rows": 193
+    },
+    {
+      "api_name": "trade_cal",
+      "first_date": "2026-08-28",
+      "last_date": "2026-09-21",
+      "snapshot_count": 17,
+      "source_rows": 787
+    }
+  ]
+}
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/findings.md b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/findings.md
new file mode 100644
index 0000000..0e8081d
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/findings.md
@@ -0,0 +1,137 @@
+# OneChart 波段与单日加权评分还原
+
+研究日期：2026-09-21。这是一次只读算法研究，不包含产品修改或数据库写入。
+
+## 结论与证据等级
+
+已找到一组低自由度、可直接执行的公式，使用网站公开的原始流入率和净额，精确重现 2026-09-04 至 2026-09-21 共 12 个交易日、9,492 条记录的单日评分、波段评分及最终排名。两个评分最大绝对误差均为 `3.410605131648481e-13`，即浮点运算误差。
+
+这是**对观测数据的数值还原**，不是取得网站后端源码。不能据此保证所有历史版本、未来版本及无观测的边界情况都采用相同实现。旧研究未识别出权重的结论不再适用于本次验证区间，但本任务没有修改旧报告或共享规格。
+
+## 公式
+
+对每个板块及日期，定义：
+
+- `r = Ratio_Raw_Pct`，以小数表示的当日流入率，页面显示时乘 100。
+- `F = Amount_Raw_BN × 100000000`，主力净流入金额，单位元。虽然字段包含 `BN`，网站实际展示单位是亿元。
+- `MA_n(x)`：按板块、日期排序后最近 n 条有效观测的简单均值，包含当日；不擅自补齐公开历史中的缺失记录。稳定验证区间每天都公开了 791 个板块，但部分窗口的前置历史仍有日期缺失。
+- `P_t(x)`：**同日、同板块类型**的升序排名百分位，`rank(x)/N`，取值从 `1/N` 到 1；概念和行业各自排名。该排名由原始指标计算，不使用网站最终 `*_RankPct` 作为输入。
+
+### 单日评分
+
+```text
+Ratio_Score = 1000 × P_t(r) × W
+```
+
+### 波段评分
+
+```text
+R3  = MA_3(r)
+R10 = MA_10(r)
+Swing_Score = 1000 × [0.5 × P_t(R3) + 0.5 × P_t(R10)] × W
+```
+
+关键是**先分别求 3 日、10 日流入率均值的横截面排名，再各乘 50%**。如果改成先把两个均值合成波段流入率，再对合成值排名，就会得到不同评分；2026-09-21 该错误方法平均偏差约 62.54 分。
+
+### 页面显示的波段流入率与波段净额
+
+```text
+Swing_Ratio_Val  = 0.5 × MA_3(r) + 0.5 × MA_10(r)
+Swing_Amount_Val = 0.5 × MA_3(Amount_Raw_BN)
+                 + 0.5 × MA_10(Amount_Raw_BN)
+```
+
+所以页面所称“3–10 日多周期协同加权”，在验证数据中可具体化为 **3 日与 10 日两个窗口，各占 50%**。没有证据表明必须引入 4、5、6、7、8、9 日窗口。该波段流入率展开到每日后，最近 3 日每一天占 `13/60 ≈ 21.6667%`，再往前 7 日每一天占 5%；最近三日合计占 65%。这种每日线性展开仅适用于原始波段流入率，不能直接替代带横截面排名的波段评分。
+
+### 成交额权重 W
+
+从公开字段可以精确验证的表达式是：
+
+```text
+V_proxy = F / r
+W = log10(MA_5(V_proxy) - 99) / 10
+```
+
+若定义与网站计算口径对应的成交额 `A = V_proxy - 100`（元），则等价于：
+
+```text
+W = log10(1 + MA_5(A)) / 10
+r = F / (A + 100)
+```
+
+`-99` 由公开数据中的金额/比率关系定位，在 791 个最新日样本中，反求的 5 日成交额与 `MA_5(F/r)` 的差均为约 99 元；使用该修正后，评分误差降至机器精度。单凭公开字段，不能证明后端源码里真的写了“分母加 100 元”，也不能断言这是防零分母常量而不是单位换算产生的等价结果；原始金额、成员和取整口径还需结合数据库核对。
+
+经济含义是用成交活跃程度调节排名得分，采用对数使规模差异的影响较温和。近 5 日平均成交额为 1 亿、10 亿、100 亿、1000 亿元时，W 约为 0.8、0.9、1.0、1.1。因而评分可以超过 1000；它不是限定在 0–1000 的百分制，也不是收益概率。
+
+作为交叉校验，JSON 中虽然未用于单日净额榜默认排序的 `Amount_Score` 也满足：
+
+```text
+Amount_Score = 1000 × P_t(Amount_Raw_BN) × W
+```
+
+## 最新日计算例子
+
+2026-09-21，概念池 N=414，病原体防治（BK1657.DC）：
+
+| 项目 | 数值 |
+| --- | ---: |
+| 单日流入率 | 6.47373795% |
+| 单日流入率原始百分位 | 407/414 = 0.9830917874 |
+| 3 日均值的百分位 | 376/414 = 0.9082125604 |
+| 10 日均值的百分位 | 275/414 = 0.6642512077 |
+| 近 5 日成交额权重 W | 约 1.056243 |
+| 单日评分 | 1038.383599887465 |
+| 波段评分 | 830.4517488043484 |
+| 网站最终单日/波段排名 | 1 / 47 |
+
+```text
+单日 = 1000 × (407/414) × W = 1038.3836 → 页面 1038.4
+波段 = 1000 × [(376/414 + 275/414)/2] × W = 830.4517 → 页面 830.5
+```
+
+它的单日原始流入率并非全池最高，成交额权重加成后，单日综合评分可以排到第一。完整的三板块样例保存在 `worked-examples.json`。
+
+## 验证范围与限制
+
+公开数据总计 23,613 行、30 个交易日，覆盖 2026-08-11 至 2026-09-21。最新日 791 个板块：概念 414、行业 377。
+
+| 验证对象 | 稳定区间样本 | 最大绝对误差 |
+| --- | ---: | ---: |
+| Ratio_Score | 9,492 | 3.41e-13 |
+| Swing_Score | 9,492 | 3.41e-13 |
+| Swing_Ratio_Val | 9,492 | 2.78e-17 |
+| Swing_Amount_Val（亿元） | 9,492 | 5.68e-14 |
+| Amount_Score（交叉校验） | 9,492 | 4.55e-13 |
+
+最终排名按分数降序完全吻合；`RankPct = 100 × (N - RankPos + 1) / N`。金额榜的最终排名依据是原始净额，不是 Amount_Score。
+
+复算过程中仅使用日期、板块代码、类型、原始单日流入率和净额；评分、最终排名只在最后比较时读取。因此没有用答案反过来构造预测输入。主会话执行了 `reproduce.py`；独立代理 `/root/score_formula_audit` 已完成同样输入边界下的核验，结果一致。公开前端取证由 `/root/onechart_public_evidence` 完成。
+
+不能把上述准确度推广到整份 30 日历史：
+
+- 最早 4/9 个观测缺少足够的 5/10 日前置历史，分别无法计算成交额权重/波段窗口。
+- 单日评分在 2026-08-17 至 08-26 有差异，最大约 6.680462 分；8 月 27 日起可计算样本吻合。
+- 波段评分在 2026-08-24 至 09-03 有差异，最大约 261.703574 分。8 月 27 日公开板块只有 707 条，部分后续窗口的输入不全；早期还存在板块集合或数据修订差异，尚未逐项确认原因。
+- 稳定验证区间没有原始比率或分数并列，不能确定后端的并列排名规则。脚本选用 `average` 仅作为明确的复算约定。
+- 全量没有 `r=0`，不能由本样本识别零净流入、零成交额、极低流动性和无历史板块的所有边界策略。
+
+## 公开实现证据
+
+- [首页](https://onechartlab.com/) 只读取 `${activeTab}_Score`，提示“后端特征算法算出的综合加权值”；本次缓存 `index.html:800–805`。前端没有公开评分构造函数。
+- 首页 `index.html:1032` 说明“3-10 个交易日多周期协同加权”；具体 50%/50% 来自数值检验，而非这句话本身。
+- 首页 `index.html:1253–1257` 按板块类型过滤；`1292–1314` 指定 Swing/Ratio 默认按各自分数排序，并用最终 RankPct 切前后 10%。
+- [radar_manifest.json](https://onechartlab.com/radar_manifest.json) 列出日期分片；已保存本次 manifest。
+- [完整公开 JSON](https://onechartlab.com/radar_data_latest.json) 和 [2026-09-21 分片](https://onechartlab.com/radar_data/dates/2026-09-21.22531a461a8b.json) 给出数值证据。`public-inputs.json.gz` 保存了所需字段和原始文件 SHA-256，避免未来网站更新导致样本改变。
+- [Tushare daily](https://tushare.pro/document/2?doc_id=27) 的 amount 单位为千元；[moneyflow_dc](https://tushare.pro/document/2?doc_id=349) 的 net_amount 单位为万元。数据库重聚合分别乘 1000 和 10000 后统一成元。
+
+## 复现
+
+在仓库根目录运行：
+
+```bash
+zhixing-server/.venv/bin/python .trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/reproduce.py
+```
+
+脚本读取固定样本，不需要网络或数据库凭据，输出 `public-validation.json`。本次使用 Python 3.12.11、pandas 3.0.5、NumPy 2.5.1。
+
+数据库的独立核对另见本目录 `database-validation.md`；研究 SQL 和结果均与生产代码隔离。
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/member-differences.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/member-differences.json
new file mode 100644
index 0000000..31e80a8
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/member-differences.json
@@ -0,0 +1,246 @@
+[
+  {
+    "code": "BK0581.DC",
+    "name": "智能电网",
+    "db_count": 197,
+    "site_count": 195,
+    "intersection_count": 190,
+    "db_only": [
+      "002851",
+      "003043",
+      "301236",
+      "301669",
+      "605336",
+      "688187",
+      "920222"
+    ],
+    "site_only": [
+      "001388",
+      "002063",
+      "300140",
+      "600522",
+      "920375"
+    ],
+    "observed_at": "2026-09-21 10:40:57.972405+00:00"
+  },
+  {
+    "code": "BK0615.DC",
+    "name": "中药概念",
+    "db_count": 146,
+    "site_count": 145,
+    "intersection_count": 144,
+    "db_only": [
+      "000626",
+      "920367"
+    ],
+    "site_only": [
+      "300391"
+    ],
+    "observed_at": "2026-09-21 10:41:08.518051+00:00"
+  },
+  {
+    "code": "BK0966.DC",
+    "name": "碳交易",
+    "db_count": 142,
+    "site_count": 138,
+    "intersection_count": 137,
+    "db_only": [
+      "000875",
+      "002734",
+      "600389",
+      "601678",
+      "603612"
+    ],
+    "site_only": [
+      "600028"
+    ],
+    "observed_at": "2026-09-21 10:43:25.392556+00:00"
+  },
+  {
+    "code": "BK1671.DC",
+    "name": "超跌股",
+    "db_count": 167,
+    "site_count": 22,
+    "intersection_count": 7,
+    "db_only": [
+      "000002",
+      "000010",
+      "000016",
+      "000639",
+      "000677",
+      "002104",
+      "002217",
+      "002227",
+      "002368",
+      "002514",
+      "002542",
+      "002547",
+      "002657",
+      "002691",
+      "002731",
+      "002869",
+      "002891",
+      "300045",
+      "300068",
+      "300100",
+      "300245",
+      "300255",
+      "300290",
+      "300352",
+      "300396",
+      "300430",
+      "300465",
+      "300484",
+      "300492",
+      "300530",
+      "300539",
+      "300584",
+      "300652",
+      "300663",
+      "300682",
+      "300703",
+      "300723",
+      "300779",
+      "300844",
+      "300879",
+      "300896",
+      "300918",
+      "300940",
+      "300995",
+      "301000",
+      "301052",
+      "301076",
+      "301139",
+      "301325",
+      "301498",
+      "301590",
+      "301601",
+      "301622",
+      "301632",
+      "600053",
+      "600180",
+      "600325",
+      "600363",
+      "600418",
+      "600491",
+      "600530",
+      "600702",
+      "600745",
+      "601127",
+      "601865",
+      "601929",
+      "603008",
+      "603189",
+      "603200",
+      "603300",
+      "603359",
+      "603370",
+      "603382",
+      "603392",
+      "603567",
+      "603630",
+      "603718",
+      "603767",
+      "603815",
+      "603848",
+      "605499",
+      "688013",
+      "688066",
+      "688068",
+      "688089",
+      "688121",
+      "688166",
+      "688189",
+      "688201",
+      "688303",
+      "688408",
+      "688496",
+      "688499",
+      "688500",
+      "688567",
+      "688573",
+      "688577",
+      "688588",
+      "688631",
+      "688639",
+      "688648",
+      "688658",
+      "688775",
+      "920001",
+      "920005",
+      "920007",
+      "920056",
+      "920061",
+      "920075",
+      "920090",
+      "920101",
+      "920106",
+      "920108",
+      "920112",
+      "920145",
+      "920146",
+      "920184",
+      "920237",
+      "920239",
+      "920247",
+      "920252",
+      "920263",
+      "920271",
+      "920273",
+      "920274",
+      "920346",
+      "920351",
+      "920375",
+      "920392",
+      "920395",
+      "920414",
+      "920429",
+      "920454",
+      "920469",
+      "920505",
+      "920508",
+      "920522",
+      "920523",
+      "920533",
+      "920578",
+      "920579",
+      "920627",
+      "920634",
+      "920689",
+      "920693",
+      "920719",
+      "920720",
+      "920770",
+      "920781",
+      "920807",
+      "920896",
+      "920906",
+      "920914",
+      "920925",
+      "920926",
+      "920932",
+      "920942",
+      "920982",
+      "920985",
+      "920992"
+    ],
+    "site_only": [
+      "000004",
+      "000056",
+      "000638",
+      "002630",
+      "300081",
+      "300344",
+      "300561",
+      "600355",
+      "600599",
+      "600696",
+      "603369",
+      "605199",
+      "688287",
+      "920130",
+      "920305"
+    ],
+    "observed_at": "2026-09-21 10:50:20.423741+00:00"
+  }
+]
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-inputs.json.gz b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-inputs.json.gz
new file mode 100644
index 0000000..28e3be5
Binary files /dev/null and b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-inputs.json.gz differ
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-validation.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-validation.json
new file mode 100644
index 0000000..098214e
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/public-validation.json
@@ -0,0 +1,116 @@
+{
+  "rows": 23613,
+  "dates": 30,
+  "windows": {
+    "all_available": {
+      "rows": 23613,
+      "dates": 30,
+      "Ratio_Score": {
+        "checked_rows": 20449,
+        "mean_absolute_error": 0.18366711208184938,
+        "max_absolute_error": 6.680461750893414,
+        "errors_above_1e-8": 1651
+      },
+      "Swing_Score": {
+        "checked_rows": 16494,
+        "mean_absolute_error": 0.5097113919412869,
+        "max_absolute_error": 261.7035740929656,
+        "errors_above_1e-8": 3199
+      },
+      "Amount_Score": {
+        "checked_rows": 20449,
+        "mean_absolute_error": 0.1882733317113178,
+        "max_absolute_error": 7.649258428013809,
+        "errors_above_1e-8": 1651
+      },
+      "Swing_Ratio_Val": {
+        "checked_rows": 16494,
+        "mean_absolute_error": 6.7217831480409905e-06,
+        "max_absolute_error": 0.012044989384502386,
+        "errors_above_1e-8": 28
+      },
+      "Swing_Amount_Val": {
+        "checked_rows": 16494,
+        "mean_absolute_error": 0.008108515254276354,
+        "max_absolute_error": 24.526995094,
+        "errors_above_1e-8": 28
+      }
+    },
+    "2026-09-04_to_2026-09-21": {
+      "rows": 9492,
+      "dates": 12,
+      "Ratio_Score": {
+        "checked_rows": 9492,
+        "mean_absolute_error": 3.2767252412841164e-14,
+        "max_absolute_error": 3.410605131648481e-13,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Score": {
+        "checked_rows": 9492,
+        "mean_absolute_error": 3.441106555949961e-14,
+        "max_absolute_error": 3.410605131648481e-13,
+        "errors_above_1e-8": 0
+      },
+      "Amount_Score": {
+        "checked_rows": 9492,
+        "mean_absolute_error": 3.245420975553957e-14,
+        "max_absolute_error": 4.547473508864641e-13,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Ratio_Val": {
+        "checked_rows": 9492,
+        "mean_absolute_error": 4.942276978457954e-18,
+        "max_absolute_error": 2.7755575615628914e-17,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Amount_Val": {
+        "checked_rows": 9492,
+        "mean_absolute_error": 2.053051893003829e-15,
+        "max_absolute_error": 5.684341886080802e-14,
+        "errors_above_1e-8": 0
+      },
+      "Ratio_ranking": {
+        "rank_position_mismatches": 0,
+        "rank_percentile_max_error": 1.4210854715202004e-14
+      },
+      "Swing_ranking": {
+        "rank_position_mismatches": 0,
+        "rank_percentile_max_error": 1.4210854715202004e-14
+      }
+    },
+    "2026-09-14_to_2026-09-21": {
+      "rows": 4746,
+      "dates": 6,
+      "Ratio_Score": {
+        "checked_rows": 4746,
+        "mean_absolute_error": 3.302167267444722e-14,
+        "max_absolute_error": 3.410605131648481e-13,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Score": {
+        "checked_rows": 4746,
+        "mean_absolute_error": 3.4959814225990485e-14,
+        "max_absolute_error": 3.410605131648481e-13,
+        "errors_above_1e-8": 0
+      },
+      "Amount_Score": {
+        "checked_rows": 4746,
+        "mean_absolute_error": 3.260761983972608e-14,
+        "max_absolute_error": 4.547473508864641e-13,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Ratio_Val": {
+        "checked_rows": 4746,
+        "mean_absolute_error": 4.952800461538844e-18,
+        "max_absolute_error": 2.7755575615628914e-17,
+        "errors_above_1e-8": 0
+      },
+      "Swing_Amount_Val": {
+        "checked_rows": 4746,
+        "mean_absolute_error": 2.056700338399156e-15,
+        "max_absolute_error": 5.684341886080802e-14,
+        "errors_above_1e-8": 0
+      }
+    }
+  }
+}
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/radar_manifest.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/radar_manifest.json
new file mode 100644
index 0000000..647af67
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/radar_manifest.json
@@ -0,0 +1 @@
+{"AVAILABLE_DATES":["2026-08-11","2026-08-12","2026-08-13","2026-08-14","2026-08-17","2026-08-18","2026-08-19","2026-08-20","2026-08-21","2026-08-24","2026-08-25","2026-08-26","2026-08-27","2026-08-28","2026-08-31","2026-09-01","2026-09-02","2026-09-03","2026-09-04","2026-09-07","2026-09-08","2026-09-09","2026-09-10","2026-09-11","2026-09-14","2026-09-15","2026-09-16","2026-09-17","2026-09-18","2026-09-21"],"DATA_SOURCE":"","LATEST_DATE":"2026-09-21","files":{"constituents":"radar_data/constituents.25ffae2b8f53.json","dates":{"2026-08-11":"radar_data/dates/2026-08-11.3ed082fdf5aa.json","2026-08-12":"radar_data/dates/2026-08-12.a61e39835037.json","2026-08-13":"radar_data/dates/2026-08-13.db6f36c2dd06.json","2026-08-14":"radar_data/dates/2026-08-14.fb36a77c6ae4.json","2026-08-17":"radar_data/dates/2026-08-17.d69d703faef6.json","2026-08-18":"radar_data/dates/2026-08-18.2c339eebabd6.json","2026-08-19":"radar_data/dates/2026-08-19.fca1cdf24170.json","2026-08-20":"radar_data/dates/2026-08-20.88ff543739cc.json","2026-08-21":"radar_data/dates/2026-08-21.a08c301a8dd7.json","2026-08-24":"radar_data/dates/2026-08-24.9135129011a3.json","2026-08-25":"radar_data/dates/2026-08-25.9393e22d8754.json","2026-08-26":"radar_data/dates/2026-08-26.f49c721846fb.json","2026-08-27":"radar_data/dates/2026-08-27.2dc5f5627eeb.json","2026-08-28":"radar_data/dates/2026-08-28.c92100ea9ae1.json","2026-08-31":"radar_data/dates/2026-08-31.c094ac3098c9.json","2026-09-01":"radar_data/dates/2026-09-01.e6253d4364f6.json","2026-09-02":"radar_data/dates/2026-09-02.1e4a8e6102d5.json","2026-09-03":"radar_data/dates/2026-09-03.ec356ce27a57.json","2026-09-04":"radar_data/dates/2026-09-04.deaf468364a2.json","2026-09-07":"radar_data/dates/2026-09-07.874c2aaa1925.json","2026-09-08":"radar_data/dates/2026-09-08.f93f7b9b2eab.json","2026-09-09":"radar_data/dates/2026-09-09.7a9919c800d9.json","2026-09-10":"radar_data/dates/2026-09-10.29e5d3c66a03.json","2026-09-11":"radar_data/dates/2026-09-11.738e1fec6654.json","2026-09-14":"radar_data/dates/2026-09-14.6b21a73a38f6.json","2026-09-15":"radar_data/dates/2026-09-15.9d5afec9a676.json","2026-09-16":"radar_data/dates/2026-09-16.48d01087bfbd.json","2026-09-17":"radar_data/dates/2026-09-17.04d53b204a8f.json","2026-09-18":"radar_data/dates/2026-09-18.b7a298aedf8d.json","2026-09-21":"radar_data/dates/2026-09-21.22531a461a8b.json"},"rank_history":"radar_data/rank_history.1d389cbfa3e1.json"},"schema_version":1,"version":"7a4fc78a8366"}
\ No newline at end of file
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/raw-reaggregation.sql b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/raw-reaggregation.sql
new file mode 100644
index 0000000..7e7805f
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/raw-reaggregation.sql
@@ -0,0 +1 @@
+WITH s AS MATERIALIZED (SELECT DISTINCT ON (api_name,target_trade_date,partition_key) api_name,target_trade_date,partition_key,payload,observed_at FROM sector_radar_source_snapshot WHERE target_trade_date BETWEEN '2026-09-07' AND '2026-09-21' AND api_name IN ('dc_member','daily','moneyflow_dc') ORDER BY api_name,target_trade_date,partition_key,observed_at DESC), d AS (SELECT DISTINCT ON (target_trade_date,e->>'ts_code') target_trade_date,e->>'ts_code' AS code,(e->>'amount')::numeric*1000 AS amount FROM s CROSS JOIN LATERAL jsonb_array_elements(payload) e WHERE api_name='daily' ORDER BY target_trade_date,e->>'ts_code',observed_at DESC), f AS (SELECT DISTINCT ON (target_trade_date,e->>'ts_code') target_trade_date,e->>'ts_code' AS code,(e->>'net_amount')::numeric*10000 AS net FROM s CROSS JOIN LATERAL jsonb_array_elements(payload) e WHERE api_name='moneyflow_dc' ORDER BY target_trade_date,e->>'ts_code',observed_at DESC), m AS (SELECT DISTINCT target_trade_date,e->>'ts_code' AS sector,e->>'con_code' AS code FROM s CROSS JOIN LATERAL jsonb_array_elements(payload) e WHERE api_name='dc_member' AND partition_key<>'all') SELECT m.target_trade_date AS trade_date,m.sector AS sector_code,count(*) AS member_count,count(d.amount) AS daily_count,count(f.net) AS flow_count,sum(d.amount) AS turnover_yuan,sum(f.net) AS net_amount_yuan,sum(d.amount) FILTER (WHERE m.code LIKE '%.BJ') AS bj_turnover_yuan FROM m LEFT JOIN d ON d.target_trade_date=m.target_trade_date AND d.code=m.code LEFT JOIN f ON f.target_trade_date=m.target_trade_date AND f.code=m.code GROUP BY m.target_trade_date,m.sector ORDER BY m.target_trade_date,m.sector
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/reproduce.py b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/reproduce.py
new file mode 100644
index 0000000..4b99975
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/reproduce.py
@@ -0,0 +1,97 @@
+"""从固定公开输入复算 OneChart 指标；预测阶段不使用网站评分或排名。"""
+
+from __future__ import annotations
+
+import gzip
+import json
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+
+ROOT = Path(__file__).resolve().parent
+
+
+def calculate(rows: list[dict]) -> pd.DataFrame:
+    """按板块有观测的历史记录计算；预热不足保留 NaN，不填补未知输入。"""
+    frame = pd.DataFrame(rows).sort_values(["ts_code", "trade_date"])
+    if frame.duplicated(["trade_date", "type", "ts_code"]).any():
+        raise ValueError("公开输入存在重复主键")
+    if frame["Ratio_Raw_Pct"].eq(0).any():
+        raise ValueError("零流入率无法通过金额/比率反求成交额；需要原始成交额")
+
+    frame["turnover_proxy_yuan"] = (
+        frame["Amount_Raw_BN"] * 1e8 / frame["Ratio_Raw_Pct"]
+    )
+    for field in ["Ratio_Raw_Pct", "Amount_Raw_BN", "turnover_proxy_yuan"]:
+        for window in [3, 5, 10]:
+            frame[f"{field}_ma{window}"] = frame.groupby("ts_code")[field].transform(
+                lambda series: series.rolling(window, min_periods=window).mean()
+            )
+
+    # -99 是公开金额/比率所能直接验证的代数修正，不把它当成原始成交额。
+    frame["liquidity_weight"] = (
+        np.log10(frame["turnover_proxy_yuan_ma5"] - 99) / 10
+    )
+    groups = frame.groupby(["trade_date", "type"])
+    for field in ["Ratio_Raw_Pct", "Ratio_Raw_Pct_ma3", "Ratio_Raw_Pct_ma10", "Amount_Raw_BN"]:
+        frame[f"{field}_percentile"] = groups[field].rank(method="average", pct=True)
+
+    frame["pred_Ratio_Score"] = (
+        1000 * frame["Ratio_Raw_Pct_percentile"] * frame["liquidity_weight"]
+    )
+    frame["pred_Swing_Score"] = (
+        1000
+        * 0.5
+        * (frame["Ratio_Raw_Pct_ma3_percentile"] + frame["Ratio_Raw_Pct_ma10_percentile"])
+        * frame["liquidity_weight"]
+    )
+    frame["pred_Amount_Score"] = (
+        1000 * frame["Amount_Raw_BN_percentile"] * frame["liquidity_weight"]
+    )
+    frame["pred_Swing_Ratio_Val"] = (
+        0.5 * (frame["Ratio_Raw_Pct_ma3"] + frame["Ratio_Raw_Pct_ma10"])
+    )
+    frame["pred_Swing_Amount_Val"] = (
+        0.5 * (frame["Amount_Raw_BN_ma3"] + frame["Amount_Raw_BN_ma10"])
+    )
+    return frame
+
+
+def validate(frame: pd.DataFrame) -> dict:
+    """比较固定公式的预测与网站输出，报告预热、全历史与稳定覆盖窗口。"""
+    metrics = ["Ratio_Score", "Swing_Score", "Amount_Score", "Swing_Ratio_Val", "Swing_Amount_Val"]
+    report = {"rows": len(frame), "dates": frame["trade_date"].nunique(), "windows": {}}
+    for name, subset in [
+        ("all_available", frame),
+        ("2026-09-04_to_2026-09-21", frame[frame["trade_date"].ge("2026-09-04")]),
+        ("2026-09-14_to_2026-09-21", frame[frame["trade_date"].ge("2026-09-14")]),
+    ]:
+        result = {"rows": len(subset), "dates": subset["trade_date"].nunique()}
+        for metric in metrics:
+            errors = (subset[metric] - subset[f"pred_{metric}"]).abs().dropna()
+            result[metric] = {
+                "checked_rows": len(errors), "mean_absolute_error": errors.mean(),
+                "max_absolute_error": errors.max(), "errors_above_1e-8": int(errors.gt(1e-8).sum()),
+            }
+        if name == "2026-09-04_to_2026-09-21":
+            groups = subset.groupby(["trade_date", "type"])
+            count = groups["ts_code"].transform("size")
+            for metric in ["Ratio", "Swing"]:
+                ranks = groups[f"pred_{metric}_Score"].rank(ascending=False)
+                percentiles = 100 * (count - ranks + 1) / count
+                result[f"{metric}_ranking"] = {
+                    "rank_position_mismatches": int(ranks.ne(subset[f"{metric}_RankPos"]).sum()),
+                    "rank_percentile_max_error": percentiles.sub(subset[f"{metric}_RankPct"]).abs().max(),
+                }
+        report["windows"][name] = result
+    return report
+
+
+if __name__ == "__main__":
+    fixture = json.loads(gzip.decompress((ROOT / "public-inputs.json.gz").read_bytes()))
+    calculated = calculate(fixture["rows"])
+    result = validate(calculated)
+    (ROOT / "public-validation.json").write_text(json.dumps(result, ensure_ascii=False, indent=2) + "\n")
+    print(json.dumps(result, ensure_ascii=False, indent=2))
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/worked-examples.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/worked-examples.json
new file mode 100644
index 0000000..3144131
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/research/worked-examples.json
@@ -0,0 +1,59 @@
+[
+  {
+    "index_name":"中药概念 (概念)",
+    "ts_code":"BK0615.DC",
+    "type":"概念板块",
+    "Ratio_Raw_Pct":0.079794960955587,
+    "Ratio_Raw_Pct_ma3":0.033754406576305,
+    "Ratio_Raw_Pct_ma10":-0.005866010935792,
+    "Ratio_Raw_Pct_percentile":0.995169082125604,
+    "Ratio_Raw_Pct_ma3_percentile":0.929951690821256,
+    "Ratio_Raw_Pct_ma10_percentile":0.64975845410628,
+    "turnover_proxy_yuan_ma5":26311461138.200000762939453,
+    "liquidity_weight":1.042014496452983,
+    "pred_Ratio_Score":1036.9806099966886,
+    "Ratio_Score":1036.9806099966886,
+    "pred_Swing_Score":823.040435604167897,
+    "Swing_Score":823.040435604167897,
+    "Ratio_RankPos":2,
+    "Swing_RankPos":49
+  },
+  {
+    "index_name":"养老概念 (概念)",
+    "ts_code":"BK0653.DC",
+    "type":"概念板块",
+    "Ratio_Raw_Pct":0.065820703429991,
+    "Ratio_Raw_Pct_ma3":0.037193483577325,
+    "Ratio_Raw_Pct_ma10":0.006875783666745,
+    "Ratio_Raw_Pct_percentile":0.985507246376812,
+    "Ratio_Raw_Pct_ma3_percentile":0.949275362318841,
+    "Ratio_Raw_Pct_ma10_percentile":0.963768115942029,
+    "turnover_proxy_yuan_ma5":32135609228.599998474121094,
+    "liquidity_weight":1.050698653636457,
+    "pred_Ratio_Score":1035.471136917088188,
+    "Ratio_Score":1035.471136917088415,
+    "pred_Swing_Score":1005.016103478350374,
+    "Swing_Score":1005.016103478350374,
+    "Ratio_RankPos":3,
+    "Swing_RankPos":2
+  },
+  {
+    "index_name":"病原体防治 (概念)",
+    "ts_code":"BK1657.DC",
+    "type":"概念板块",
+    "Ratio_Raw_Pct":0.064737379547795,
+    "Ratio_Raw_Pct_ma3":0.029828519014713,
+    "Ratio_Raw_Pct_ma10":-0.005451455299003,
+    "Ratio_Raw_Pct_percentile":0.983091787439614,
+    "Ratio_Raw_Pct_ma3_percentile":0.908212560386474,
+    "Ratio_Raw_Pct_ma10_percentile":0.664251207729468,
+    "turnover_proxy_yuan_ma5":36511340146.0,
+    "liquidity_weight":1.056242777281107,
+    "pred_Ratio_Score":1038.383599887464925,
+    "Ratio_Score":1038.383599887464925,
+    "pred_Swing_Score":830.451748804348426,
+    "Swing_Score":830.451748804348426,
+    "Ratio_RankPos":1,
+    "Swing_RankPos":47
+  }
+]
diff --git a/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/task.json b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/task.json
new file mode 100644
index 0000000..56fdd4f
--- /dev/null
+++ b/.trellis/tasks/archive/2026-09/09-21-onechart-score-reconstruction/task.json
@@ -0,0 +1,33 @@
+{
+  "id": "onechart-score-reconstruction",
+  "name": "onechart-score-reconstruction",
+  "title": "还原 OneChart 波段与单日资金流评分",
+  "description": "只读数值还原 OneChart 单日与波段资金评分；12 日 9492 条公开样本精确重现，并完成 PostgreSQL 原始快照对照。",
+  "status": "completed",
+  "dev_type": null,
+  "scope": null,
+  "package": null,
+  "priority": "P2",
+  "creator": "yuxuanhui",
+  "assignee": "yuxuanhui",
+  "createdAt": "2026-09-21",
+  "completedAt": "2026-09-21",
+  "branch": null,
+  "base_branch": "main",
+  "worktree_path": null,
+  "commit": null,
+  "pr_url": null,
+  "subtasks": [],
+  "children": [],
+  "parent": null,
+  "relatedFiles": [],
+  "notes": "研究完成。没有实施产品修改、修改数据库、共享规格推广或提交。输入版本与成员口径差异详见 research/database-validation.md。",
+  "meta": {
+    "work_kind": "read_only_research",
+    "public_validation_rows": 9492,
+    "public_score_max_abs_error": 3.410605131648481e-13,
+    "database_latest_rows": 791,
+    "database_ratio_mae": 0.9783318452555938,
+    "database_swing_mae": 1.9075528008117222
+  }
+}
\ No newline at end of file