From 4cd9b9e756b9abf51e3a8c37f88f976ca67f0544 Mon Sep 17 00:00:00 2001
From: 732642856 <732642856@qq.com>
Date: Sat, 27 Jun 2026 23:53:53 +0800
Subject: [PATCH] Auto-derive finance promise strength

---
 ...adjudicator_third_pass_audit_2026_06_27.md | 163 ++++++++++++++++++
 mcp_server.py                                 |  27 +++
 tests/test_mcp_strict_workflow_finance.py     |  41 ++++-
 3 files changed, 230 insertions(+), 1 deletion(-)
 create mode 100644 docs/research/wealth_adjudicator_third_pass_audit_2026_06_27.md

diff --git a/docs/research/wealth_adjudicator_third_pass_audit_2026_06_27.md b/docs/research/wealth_adjudicator_third_pass_audit_2026_06_27.md
new file mode 100644
index 00000000..02358fe6
--- /dev/null
+++ b/docs/research/wealth_adjudicator_third_pass_audit_2026_06_27.md
@@ -0,0 +1,163 @@
+# Wealth Adjudicator Third-Pass Audit (2026-06-27)
+
+> 目标：把 `wealth_promise_strength` 从“手工喂入的测试字段”推进到“可由仓内真源自动折叠”的真实裁决输入。
+
+---
+
+## 1. 本轮完成了什么
+
+本轮没有新增财富标签，也没有继续扩 promise 来源。  
+只完成了一件关键工作：
+
+**让 `strict_workflow -> _collect_strict_evidence("finance")` 能从**
+
+- `full-reading.modules.yogas_doshas.dhana_yogas`
+
+**自动折叠出**
+
+- `wealth_promise_strength`
+
+当前折叠规则非常克制：
+
+- 至少一条 `strong` Dhana Yoga -> `level = strong`
+- 否则至少一条 `moderate` Dhana Yoga -> `level = moderate`
+- 否则 -> `weak`
+- 没有 `dhana_yogas` -> `None`
+
+并产出结构：
+
+```json
+{
+  "level": "strong",
+  "source": "dhana_yogas",
+  "count": 2
+}
+```
+
+---
+
+## 2. 为什么这一步值钱
+
+前两轮虽然已经把财富裁决器推进到：
+
+- `public_wealth_status` 行为护栏
+- `income_growth` vs `public_wealth_status` 分流
+- `wealth_promise_strength` 显式入分
+
+但那时还有一个明显的工程短板：
+
+> `wealth_promise_strength` 还是手工喂进去的，不是真正来自 skill 自己的知识资产。
+
+本轮之后，这个短板第一次被抹平了一部分。  
+也就是说，财富裁决器已经开始从仓内现成的财富真源里，自己长出 promise 分。
+
+---
+
+## 3. 回归结果
+
+### 3.1 自动 promise 折叠验证
+
+使用含有：
+
+- `D2`
+- `D10`
+- `Shadbala`
+- `Ashtakavarga`
+- `Vimshottari + Narayana`
+- `wealth/gains/career convergence`
+- `dhana_yogas = [strong, moderate]`
+
+的 finance strict evidence 场景，当前返回：
+
+- `present_evidence.wealth_promise_strength = {"level":"strong","source":"dhana_yogas","count":2}`
+- `event_judgement.score = 100`
+- `event_judgement.verdict = moderate_probability_window`
+- `event_judgement.payout_label = public_wealth_status`
+
+### 3.2 这里最重要的不是满分，而是“满分仍被压在 moderate”
+
+这个现象其实是好的，而不是坏的。
+
+它说明当前财富裁决器的边界依旧保持保守：
+
+- 即便完整证据场景把分数顶满
+- 也不会因为 finance score 很高就直接升级到过满承诺
+
+这和项目当前的真实边界是一致的：
+
+> 我们在提升财富裁决器的结构成熟度，  
+> 但还没有宣布“精确财富应期已全球封顶”。
+
+---
+
+## 4. 当前状态相较前两轮的提升
+
+### 第一轮
+
+- 加护栏
+- 堵住弱窗口乱贴 `public_wealth_status`
+
+### 第二轮
+
+- `wealth_promise_strength` 显式入分
+- `income_growth` 与 `public_wealth_status` 开始分家
+
+### 第三轮（本轮）
+
+- `wealth_promise_strength` 不再只是手工输入
+- 它开始从仓里的 `dhana_yogas` 自动生成
+
+所以本轮的本质是：
+
+> 财富 adjudicator 从“能接收 promise 分”进一步走到了“能自己从真源提 promise 分”。
+
+---
+
+## 5. 仍未完成的部分
+
+这轮虽然很值钱，但还没有到终局：
+
+1. `wealth_promise_strength` 目前只吸收了 `dhana_yogas`
+2. `lakshmi_dhana_activation_chain` 还没有自动接入
+3. `yogi_asc_tight_orb_wealth` 还没有自动接入
+4. `asset_accumulation` vs `public_wealth_status` 还没有专门回归样本
+5. `dominant label + secondary context` 还没建立
+
+---
+
+## 6. 下一轮最值钱的动作
+
+如果继续按 ROI 排序，下一轮最值钱的是：
+
+1. 把 `lakshmi_dhana_activation_chain` 接成第二个自动 promise hook
+2. 把 `yogi_asc_tight_orb_wealth` 接成第三个自动 promise hook
+3. 然后再看是否需要从单标签走向：
+   - `dominant label`
+   - `secondary context`
+
+这样我们就能处理你前面提到的下一类样本：
+
+- 收益爆发很强
+- 同时公众事业可见度也不低
+
+而不是只能二选一硬切。
+
+---
+
+## 7. 本轮状态判断
+
+本轮可以诚实宣称：
+
+- 财富 adjudicator 已经开始自动吸收仓内 `Dhana Yoga` 真源
+- `wealth_promise_strength` 已从手工测试输入升级为自动折叠输入
+- 当前保守边界仍在，没有因为自动 promise 折叠就开始乱给高承诺
+
+但还不能宣称：
+
+- 全部财富 promise 资产已自动闭环
+- `Lakshmi / Yogi / Dhana` 三路来源都已折叠完成
+
+更准确的说法是：
+
+> 财富 adjudicator 已进入“自动 promise 折叠”的第三阶段，  
+> 现在不再只是规则聚合器，而开始真正从 skill 自己的财富知识资产中抽取裁决分数。
diff --git a/mcp_server.py b/mcp_server.py
index e2c050d2..7d288d74 100644
--- a/mcp_server.py
+++ b/mcp_server.py
@@ -108,6 +108,32 @@ def _convergence_score(convergence: Any) -> int:
     return mapping.get(level, 0)
 
 
+def _derive_wealth_promise_strength(modules: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+    yogas_doshas = modules.get("yogas_doshas") if isinstance(modules, dict) else {}
+    dhana = yogas_doshas.get("dhana_yogas") if isinstance(yogas_doshas, dict) else {}
+    yogas = dhana.get("yogas") if isinstance(dhana, dict) else None
+    if not isinstance(yogas, list) or not yogas:
+        return None
+
+    strengths = [
+        str(row.get("strength", "")).lower()
+        for row in yogas
+        if isinstance(row, dict)
+    ]
+    if any(level == "strong" for level in strengths):
+        level = "strong"
+    elif any(level == "moderate" for level in strengths):
+        level = "moderate"
+    else:
+        level = "weak"
+
+    return {
+        "level": level,
+        "source": "dhana_yogas",
+        "count": len(yogas),
+    }
+
+
 def _derive_event_judgement(route: str, present: Dict[str, Any], missing: List[str]) -> Dict[str, Any]:
     if route == "relationship":
         score = 0
@@ -290,6 +316,7 @@ def _collect_strict_evidence(route: str, result: Dict[str, Any]) -> Dict[str, An
             "wealth_convergence": domain_activations.get("wealth_family"),
             "gains_convergence": domain_activations.get("gains_wishes"),
             "career_convergence": domain_activations.get("career_status"),
+            "wealth_promise_strength": _derive_wealth_promise_strength(modules),
         }
         missing = [key for key, value in present.items() if key not in {
             "gains_convergence", "career_convergence"
diff --git a/tests/test_mcp_strict_workflow_finance.py b/tests/test_mcp_strict_workflow_finance.py
index 5e0f6e68..f8d9b080 100644
--- a/tests/test_mcp_strict_workflow_finance.py
+++ b/tests/test_mcp_strict_workflow_finance.py
@@ -3,7 +3,7 @@
 
 from __future__ import annotations
 
-from mcp_server import _derive_event_judgement
+from mcp_server import _collect_strict_evidence, _derive_event_judgement
 
 
 def test_finance_public_wealth_label_requires_at_least_moderate_window() -> None:
@@ -80,3 +80,42 @@ def test_finance_strong_wealth_promise_can_unlock_public_wealth_status() -> None
     assert judgement["score"] == 60
     assert judgement["verdict"] == "moderate_probability_window"
     assert judgement["payout_label"] == "public_wealth_status"
+
+
+def test_collect_strict_evidence_finance_derives_wealth_promise_from_dhana_yogas() -> None:
+    result = {
+        "modules": {
+            "varga_full": {"D2_Hora": {"summary": "ok"}, "D10_Dasamsa": {"summary": "ok"}},
+            "shadbala": {"planets": {"Venus": {"total_rupa": 8.2}}},
+            "ashtakavarga": {"house_scores": {"2": 31, "11": 36}},
+            "dasha": {"current_dasha": {"mahadasha": "Venus", "antardasha": "Mercury"}},
+            "narayana_dasha": {"current_dasha": {"sign": "Taurus", "lord": "Venus"}},
+            "dasa_convergence": {
+                "domain_activations": {
+                    "wealth_family": {"convergence_level": "L1", "probability": "+15-20%"},
+                    "gains_wishes": {"convergence_level": "L1", "probability": "+15-20%"},
+                    "career_status": {"convergence_level": "L1", "probability": "+15-20%"},
+                }
+            },
+            "yogas_doshas": {
+                "dhana_yogas": {
+                    "yogas": [
+                        {"type": "Dhana Yoga", "strength": "strong"},
+                        {"type": "Dhana Yoga", "strength": "moderate"},
+                    ],
+                    "summary": "Dhana Yoga检测：共2个格局",
+                }
+            },
+        }
+    }
+
+    strict = _collect_strict_evidence("finance", result)
+    present = strict["present_evidence"]
+    assert present["wealth_promise_strength"] == {
+        "level": "strong",
+        "source": "dhana_yogas",
+        "count": 2,
+    }
+    assert strict["event_judgement"]["score"] == 100
+    assert strict["event_judgement"]["verdict"] == "moderate_probability_window"
+    assert strict["event_judgement"]["payout_label"] == "public_wealth_status"