落地第十五条豁免额度校验;登记三处业务裁定

一、第十五条豁免规则(docs/25 第七节 #2,业务裁定:实现)

政策原文:C3→R4 签署风险揭示书后**可买**,但单只 R4 持仓不超过总资产 20%;
C4→R5 同理,上限 10%。越级购买本身不是违规,超出额度才是 —— 原先扫描侧只看
"留痕是否齐全",于是"签了字但买超额度"这种明确违规没有预警;研判侧也把豁免的
前提条件(留痕齐全)当成了结论,直接判"疑似误报"。

- 扫描侧:新增 EXEMPTION_LIMITS 与 RiskRuleEngine._exemption_state,核算
  "单只持仓 / 总资产"并写进证据快照;触发条件改为
  gap > 0 and (missing_trace or 超出额度)。
- 研判侧:_assess_rw007 先判额度再判留痕。超限 → 证据支持风险;留痕齐全且在额度
  内 → 疑似误报;留痕齐全但快照缺总资产/持仓 → 继续复核。

数据前提(tools/probe_exemption_data.py,证据见 docs/evidence/exemption-data-probe.json):
库内 fin_customer_profile 仅 1 行且 total_asset = 0.00、fin_holding 0 行、无任何申购
交易 —— 这条规则当前不会被触发,与 behavior_score 同源(画像与持仓由本项目之外的
流程写入)。因此刻意不把"算不出来"当成"超限":拿 0 去算会让每一笔 C3→R4 都变成违规,
豁免规则反倒成了误报源。上游把数据写入后无需再改代码即可生效。

二、三处业务裁定(此前挂在"待裁定")

- 模型网关 chat + tools 入口:本轮不补,按基座能力缺口记录。它要贯穿
  ModelGateway → … → BaseAgent 整条链路,属公共契约变更,演示联调期影响面大于收益。
- exclude(关闭误报)是否必须先"调查中":保持现状,不加门禁。
- 政策冲突:以第十四条 C ≥ R 为准;客服侧 check_suitability 复核后确认本来就按
  C ≥ R 实现,无需改动。

三、其他

- 新增 tests/unit/service/test_risk_judgement_rw007.py(6 例)与扫描侧 4 例。
- 风控文档 03/05 同步 RW-007 的豁免额度条件与研判口径。
This commit is contained in:
2026-09-11 14:24:45 +08:00
parent df96275ab5
commit ff681df143
9 changed files with 518 additions and 16 deletions
+101
View File
@@ -0,0 +1,101 @@
"""只读探查:第十五条豁免规则(持仓占比)所需的数据是否齐备。
只做 SELECT。结果写 `docs/evidence/exemption-data-probe.json`:
python tools/probe_exemption_data.py
要回答的问题:
1. `fin_customer_profile.total_asset` 有没有值、是否为正;
2. `fin_holding` 有没有行、`market_value` / `current_value` 是否可用;
3. 库内是否存在 C3→R4 / C4→R5 的申购交易 —— 即这条豁免规则是否真的会被触发。
"""
from __future__ import annotations
import asyncio
import json
from pathlib import Path
from typing import Any
from sqlalchemy import text
from app.infrastructure.db import SessionFactory
OUTPUT = Path("docs/evidence/exemption-data-probe.json")
QUERIES: dict[str, str] = {
"profile_total_asset": """
SELECT COUNT(*) AS rows_count,
SUM(total_asset > 0) AS positive_assets,
SUM(total_asset = 0) AS zero_assets,
MIN(total_asset) AS min_asset,
MAX(total_asset) AS max_asset
FROM fin_customer_profile
""",
"profile_investor_type": """
SELECT investor_type, COUNT(*) AS rows_count
FROM fin_customer_profile GROUP BY investor_type ORDER BY investor_type
""",
"holding_shape": """
SELECT COUNT(*) AS rows_count,
SUM(market_value IS NULL) AS null_market_value,
SUM(current_value > 0) AS positive_current_value,
MIN(current_value) AS min_current_value,
MAX(current_value) AS max_current_value
FROM fin_holding
""",
"subscription_pairs": """
SELECT c.investor_type AS customer_level,
p.risk_level AS product_level,
COUNT(*) AS transactions
FROM fin_transaction t
JOIN fin_customer_profile c ON c.customer_id = t.customer_id
JOIN fin_product p ON p.id = t.product_id
WHERE t.transaction_type = '申购'
GROUP BY c.investor_type, p.risk_level
ORDER BY c.investor_type, p.risk_level
""",
"exemptible_pairs_detail": """
SELECT t.transaction_no,
c.investor_type AS customer_level,
p.risk_level AS product_level,
c.total_asset,
(
SELECT h.current_value FROM fin_holding h
WHERE h.customer_id = t.customer_id AND h.product_id = t.product_id
ORDER BY h.id DESC LIMIT 1
) AS holding_current_value
FROM fin_transaction t
JOIN fin_customer_profile c ON c.customer_id = t.customer_id
JOIN fin_product p ON p.id = t.product_id
WHERE t.transaction_type = '申购'
AND (
(c.investor_type = 'C3' AND p.risk_level = 'R4')
OR (c.investor_type = 'C4' AND p.risk_level = 'R5')
)
ORDER BY t.id
""",
}
async def collect() -> dict[str, Any]:
report: dict[str, Any] = {}
async with SessionFactory() as session:
for name, sql in QUERIES.items():
rows = (await session.execute(text(sql))).mappings().all()
report[name] = [dict(row) for row in rows]
return report
async def main() -> None:
report = await collect()
OUTPUT.parent.mkdir(parents=True, exist_ok=True)
OUTPUT.write_text(
json.dumps(report, ensure_ascii=False, indent=2, default=str),
encoding="utf-8",
)
print(f"wrote {OUTPUT}")
if __name__ == "__main__":
asyncio.run(main())