Harden skill rigor constraints and sync reading quality tools
This commit is contained in:
@@ -0,0 +1,203 @@
|
||||
# 生时校正分盘调用决策树
|
||||
|
||||
> 目标:把生时校正从“所有分盘一起上”改成“按证据价值和过拟合风险分层调用”。
|
||||
> 结论先行:**不是所有分盘一股脑上。**
|
||||
|
||||
## 1. 总原则
|
||||
|
||||
生时校正时,优先级应当是:
|
||||
|
||||
1. `Dasha + dated events`
|
||||
2. `D9`
|
||||
3. `D10`
|
||||
4. `D7 / D12 / D24 / D4 / D2` 按主题选
|
||||
5. `Nakshatra / Pada / house shifts`
|
||||
6. `D30`
|
||||
7. `D60`
|
||||
|
||||
一句话版本:
|
||||
|
||||
**Dasha 定框,D9/D10 定核心,D12/D7/D24 等按题补刀,D30/D60 最后谨慎使用。**
|
||||
|
||||
这和当前仓库里已有的校时权重一致:
|
||||
|
||||
- `Dasha` 40%
|
||||
- `分盘一致性` 35%
|
||||
- `宫位变化` 15%
|
||||
- `Nakshatra Pada` 10%
|
||||
|
||||
对应实现见:
|
||||
|
||||
- [jyotish-app/rectification-engine.js](/Users/wuyongnaren/Documents/印度占星/jyotish-app/rectification-engine.js)
|
||||
- [references/varga-system-quick-reference.md](/Users/wuyongnaren/Documents/印度占星/references/varga-system-quick-reference.md)
|
||||
|
||||
## 2. 第一层:主引擎
|
||||
|
||||
### Dasha + dated events
|
||||
|
||||
这是第一裁判,不是某一张分盘。
|
||||
|
||||
先用真实事件把时间框住:
|
||||
|
||||
- 结婚 / 分手 / 离婚
|
||||
- 入职 / 升职 / 创业 / 换工作
|
||||
- 搬迁 / 出国 / 移民
|
||||
- 生育 / 流产 / 子女关键年份
|
||||
- 健康危机 / 手术 / 事故
|
||||
- 重大财富变化 / 破财
|
||||
|
||||
如果没有 dated events,只靠分盘做校时,风险会显著上升。
|
||||
|
||||
## 3. 第二层:核心分盘
|
||||
|
||||
### D9
|
||||
|
||||
最适合验证:
|
||||
|
||||
- 婚姻 / 关系质量
|
||||
- 亲密关系成熟度
|
||||
- 配偶画像
|
||||
- 内在命运路线是否与 D1 同向
|
||||
|
||||
### D10
|
||||
|
||||
最适合验证:
|
||||
|
||||
- 事业显化
|
||||
- 职业兑现
|
||||
- 升职 / 名声 / 社会角色
|
||||
- 事业方向切换是否成立
|
||||
|
||||
如果用户提供的事件主要是婚恋和事业,**D9 + D10 是最值钱的双核**。
|
||||
|
||||
## 4. 第三层:主题分盘补刀
|
||||
|
||||
当 Dasha 已经定出大框,但两个候选时间都还能解释时,再按主题调专项分盘:
|
||||
|
||||
| 事件主题 | 优先分盘 | 用法 |
|
||||
|---|---|---|
|
||||
| 婚姻/关系 | `D9` | 核心,不是补充 |
|
||||
| 事业/升职/换工作 | `D10` | 核心,不是补充 |
|
||||
| 子女/生育 | `D7` | 用于拉开生育相关年份与子女主题差异 |
|
||||
| 父母/家族 | `D12` | 辅助裁判,不是第一裁判 |
|
||||
| 教育/考试 | `D24` | 学业、考试、研究路线 |
|
||||
| 房产/搬迁 | `D4` | 房产、居所、迁移、家庭落点 |
|
||||
| 财富/收入变化 | `D2` | 收入、资产、财务结构变化 |
|
||||
| 健康/事故/创伤 | `D30` | 只在有明确健康事件时用 |
|
||||
|
||||
其中 `D12` 的角色要说清楚:
|
||||
|
||||
- 它很有价值
|
||||
- 但通常在 `D9/D10` 不能彻底拉开差距时再上
|
||||
- 所以它是**辅助裁判**,不是**第一裁判**
|
||||
|
||||
## 5. 第四层:细分钟收口层
|
||||
|
||||
这一层适合最后 2-10 分钟的缩窄,不适合一开始就拍板:
|
||||
|
||||
- `Nakshatra / Pada`
|
||||
- 特殊 Lagna
|
||||
- 敏感点
|
||||
- 分盘上升切换点
|
||||
- 宫位变动
|
||||
|
||||
它们的作用是:
|
||||
|
||||
- 在大框已经基本收敛后
|
||||
- 继续缩小到更细的分钟段
|
||||
|
||||
不应让这一层反过来主导整体结论。
|
||||
|
||||
## 6. 第五层:超高敏层,慎用
|
||||
|
||||
### D30
|
||||
|
||||
可用,但前提是用户有明确的:
|
||||
|
||||
- 健康危机
|
||||
- 手术
|
||||
- 创伤
|
||||
- 事故
|
||||
- 极端压力事件
|
||||
|
||||
### D60
|
||||
|
||||
只适合在已经高度收敛后,作为最后参考。
|
||||
|
||||
不要让 `D60` 反过来主导校时结论。原因很简单:
|
||||
|
||||
- 对出生时间太敏感
|
||||
- 在事件样本不足时特别容易过拟合
|
||||
- 很容易出现“解释很漂亮,但时间其实没校准好”
|
||||
|
||||
## 7. 实战决策树
|
||||
|
||||
### Step 1
|
||||
|
||||
先问:有没有 5 个以上带年份或日期的真实事件?
|
||||
|
||||
- 没有:先补事件,不急着堆分盘
|
||||
- 有:进入 Dasha 定框
|
||||
|
||||
### Step 2
|
||||
|
||||
事件主要落在哪个主题?
|
||||
|
||||
- 婚姻/关系 -> `D9`
|
||||
- 事业/升职/换工作 -> `D10`
|
||||
- 子女/生育 -> `D7`
|
||||
- 父母/家族 -> `D12`
|
||||
- 教育/考试 -> `D24`
|
||||
- 房产/搬迁 -> `D4`
|
||||
- 财富/收入 -> `D2`
|
||||
- 健康/事故/创伤 -> `D30`
|
||||
|
||||
### Step 3
|
||||
|
||||
如果两个候选时间在核心分盘上仍都说得通:
|
||||
|
||||
- 再看 `Nakshatra / Pada / house shifts`
|
||||
- 再看更窄的分盘上升切换
|
||||
|
||||
### Step 4
|
||||
|
||||
只有当 Dasha + 核心分盘 + 主题分盘都已经明显收敛后,才允许上 `D60` 做最后参考。
|
||||
|
||||
## 8. 时间敏感度提醒
|
||||
|
||||
仓库当前已有的分盘时间精度要求:
|
||||
|
||||
- `D9 / D10`: `±2分钟`
|
||||
- `D2/D3/D4/D7/D12`: `±5分钟`
|
||||
- `D30`: `±1分钟`
|
||||
- `D60`: `±30秒`
|
||||
|
||||
因此:
|
||||
|
||||
- `D9/D10` 足够敏感,适合作为校时主力
|
||||
- `D30/D60` 过于敏感,适合作为后段验证
|
||||
|
||||
## 9. 禁止做法
|
||||
|
||||
以下做法不推荐:
|
||||
|
||||
- 一开始就把所有分盘全开
|
||||
- 用 `D60` 反向主导时间结论
|
||||
- 没有 dated events 时过度依赖高敏分盘
|
||||
- 用单一分盘推翻 `D1 + Dasha + 核心分盘` 的同向证据
|
||||
|
||||
## 10. 落地口径
|
||||
|
||||
校时时,默认解释顺序应是:
|
||||
|
||||
1. `Dasha + dated events`
|
||||
2. `D9`
|
||||
3. `D10`
|
||||
4. 主题专项分盘
|
||||
5. `Nakshatra / Pada / house shifts`
|
||||
6. `D30`
|
||||
7. `D60`
|
||||
|
||||
如果需要一句给用户看的简明口径,就用这句:
|
||||
|
||||
**校时主力不是“所有分盘”,而是“Dasha 定框,D9/D10 定核心,D12/D7/D24 等按题补刀,D30/D60 最后谨慎使用”。**
|
||||
@@ -299,6 +299,50 @@
|
||||
"Upgrade only when D1 promise, D9 support, UL condition and timing all point in the same direction."
|
||||
]
|
||||
},
|
||||
"natural_malefic_functional_role_guardrail": {
|
||||
"title": "Natural Malefic vs Functional Role Guardrail",
|
||||
"status": "frozen_template",
|
||||
"authority_level": "A/B/C layered",
|
||||
"domain": ["interpretation_boundary", "functional_role", "saturn", "mars", "rahu", "ketu"],
|
||||
"source_refs": [
|
||||
"AGENTS.md",
|
||||
"SKILL.md",
|
||||
"references/prediction-boundary-protocol.md",
|
||||
"docs/research/malefic_overread_top10_2026_06_27.md"
|
||||
],
|
||||
"trigger_patterns": [
|
||||
"Saturn negative",
|
||||
"Mars negative",
|
||||
"Rahu negative",
|
||||
"Ketu negative",
|
||||
"功能性吉凶星",
|
||||
"自然凶星",
|
||||
"Rahu/Ketu 负面",
|
||||
"土星是不是坏",
|
||||
"火星是不是坏"
|
||||
],
|
||||
"required_cross_checks": [
|
||||
"Separate natural benefic/malefic status from functional role under the current Lagna",
|
||||
"State the topic-domain role before giving any positive or negative conclusion",
|
||||
"Check D1 promise plus the relevant Varga before calling the planet harmful or helpful",
|
||||
"Check Dasha/Transit activation before upgrading from trait to event claim",
|
||||
"For Rahu/Ketu, declare node mode/dispositor and evaluate both breakthrough and loss patterns"
|
||||
],
|
||||
"confidence_ceiling": "B when natural role, functional role, topic-domain role and timing all agree; C when only natural malefic language is available or the functional role is unresolved.",
|
||||
"forbidden_claims": [
|
||||
"Saturn alone means failure",
|
||||
"Mars alone means relationship destruction",
|
||||
"Rahu or Ketu alone proves damage, scandal or collapse",
|
||||
"A natural malefic label alone overrides functional role, D1 promise or timing"
|
||||
],
|
||||
"safe_output_patterns": [
|
||||
"Always describe the planet in four layers: natural nature, functional role, topic-domain role and current activated outcome.",
|
||||
"Frame Saturn as delay, responsibility, structure or long-cycle payoff depending on promise and activation, not as automatic negativity.",
|
||||
"Frame Mars as conflict, execution, surgery, competition or entrepreneurial drive depending on context, not as automatic damage.",
|
||||
"Frame Rahu and Ketu as dual-edge amplifiers that can produce breakthrough, disruption, detachment or leakage depending on promise, dispositor and activation.",
|
||||
"If natural and functional meanings conflict, explicitly lower confidence and let the promise/timing layer decide the practical reading."
|
||||
]
|
||||
},
|
||||
"pushkara_vargottama_avastha_synthesis": {
|
||||
"title": "Pushkara / Vargottama / Avastha / RTN Synthesis Template",
|
||||
"status": "frozen_template",
|
||||
|
||||
@@ -414,7 +414,8 @@ SCRIPT=~/.workbuddy/skills/jyotish-vedic-astrology/scripts/jyotish_engine.py
|
||||
1. **common-misconceptions.md**:印度占星常见误判与冲突问题集(错题本)⭐⭐⭐⭐⭐
|
||||
2. **modern-language-guide.md**:现代生活措辞指南
|
||||
3. **birth-time-rectification-advanced.md**:出生时间矫正高级方法论
|
||||
4. **birth-time-rectification-cases.md**:出生时间矫正案例集
|
||||
4. **birth-time-rectification-decision-tree.md**:生时校正分盘调用决策树(Dasha 定框,D9/D10 定核心)
|
||||
5. **birth-time-rectification-cases.md**:出生时间矫正案例集
|
||||
5. **pdf-chart-reading-guide.md**:PDF星盘读取指南 v3.0
|
||||
6. **prediction-checklist.md**:预测清单
|
||||
7. **data-bridge-mapping.md**:数据桥接映射
|
||||
|
||||
@@ -0,0 +1,60 @@
|
||||
# Real Reading Quality Checklist
|
||||
|
||||
> 作用:给真实解读、案例分析、代理执行和高严谨输出做统一质检。
|
||||
> 核心原则:**最容易毁掉答案质量的,不是少看一颗星,而是少看一层裁决结构。**
|
||||
|
||||
## A. 参数冻结层
|
||||
|
||||
1. `Ayanamsa` 是否显式声明?
|
||||
2. `Node mode` (`Mean` / `True`) 是否显式声明?
|
||||
3. 相关 `house system / bhava policy` 是否声明?
|
||||
4. 若参数层有冲突或缺失,是否明确降置信度?
|
||||
|
||||
## B. 路由与题目域层
|
||||
|
||||
5. 是否先读取并执行了 `strict-workflow-router.md` 对应 route?
|
||||
6. 是否按题目域强制展开相关分盘,而不是只看 D1?
|
||||
7. 是否把 `covered` 和 `complete` 分开表达,没有偷换成熟度?
|
||||
|
||||
## C. 承诺层
|
||||
|
||||
8. 是否先看了 D1 的 promise,而不是直接进入细节技法?
|
||||
9. 是否检查了题目域核心宫位、宫主、承诺型 Yoga?
|
||||
10. 是否避免让单一 Yoga / 单一文章规则越权?
|
||||
|
||||
## D. 功能性角色层
|
||||
|
||||
11. 是否显式给出 `Functional Benefic/Malefic`?
|
||||
12. 是否区分了自然属性、功能属性、题目域角色、当前激活结果?
|
||||
13. 对 `Saturn / Mars / Rahu / Ketu` 是否避免了“天然负面”式偷懒断法?
|
||||
|
||||
## E. 分盘层
|
||||
|
||||
14. 婚恋是否至少展开 `D9 / UL / DK`?
|
||||
15. 事业是否至少展开 `D10 / A10 / AmK/Karakamsha`?
|
||||
16. 财富是否至少展开 `D2/D11`,并回看 D1/D9/D10 的变现链?
|
||||
17. 若使用 `D30 / D60`,是否明确说明这是高敏后置层而非主裁判?
|
||||
|
||||
## F. Timing 层
|
||||
|
||||
18. 是否至少执行 `Vimshottari + Narayana` 双轨交叉?
|
||||
19. 若做具体事件判断,是否加入 `Transit / Double Transit / annual / KP or Chara support`?
|
||||
20. 如果 timing 只做到窗口级,而不是日期级,是否说清了边界?
|
||||
|
||||
## G. 外部真值与诚信边界
|
||||
|
||||
21. 是否清楚声明 `production_tuning_allowed` 的当前状态?
|
||||
22. 是否把外部 oracle 未闭环的前线(Dasha boundary / Shadbala absolute / Tajika annual)说清楚?
|
||||
23. 是否避免把内部一致性伪装成传统软件级真值冻结?
|
||||
|
||||
## H. 输出层
|
||||
|
||||
24. 是否输出 `Technique Audit Table` 或等价审计表?
|
||||
25. 是否给出 `used / partial / blocked / not used` 级别说明?
|
||||
26. 是否给出置信度,而不是只给断语?
|
||||
27. 是否把 must-not-overclaim 边界说清楚?
|
||||
|
||||
## 一句话用法
|
||||
|
||||
只要这张表里有 3-4 项没过,就不要把输出包装成“高严谨结论”。
|
||||
只要 `参数层 + 承诺层 + 分盘层 + timing 层 + 外部真值边界` 没一起过,就不能把 skill 说成“已经无疏漏封顶”。
|
||||
@@ -12,6 +12,23 @@ The user should not need to ask for Chara Dasha, A10, Argala, Shadbala, Ashtakav
|
||||
|
||||
Every non-trivial reading must end with a **Technique Audit Table**.
|
||||
|
||||
## 0.5 High-Rigor Override
|
||||
|
||||
When the user asks for maximum rigor, no guessing, past-event verification, raw evidence, or full parity with top open-source Jyotish engines, the reading enters **High-Rigor Override**.
|
||||
|
||||
Under this override, the engine must:
|
||||
|
||||
1. Attempt cross-reference against `PyJHora / VedAstro / jyotishganit`, while respecting license boundaries.
|
||||
2. Use native repository implementations instead of lightweight summary scripts whenever a native module already exists.
|
||||
3. Require `Vimshottari + Narayana Dasha` before making major timing claims; if they materially conflict, mark the conclusion `blocked` or downgrade confidence.
|
||||
4. Force relevant divisional microscope layers:
|
||||
- `D10` + `A10` for career
|
||||
- `D2/D11` for wealth
|
||||
- `D9` + `UL` for relationship
|
||||
5. Include raw outputs: exact degrees, dasha boundaries, Shadbala / AV values, Yoga names, ayanamsa / node mode, and external evidence pointers where available.
|
||||
|
||||
If any of the above cannot be completed because of missing birth precision, external oracle gaps, license quarantine, adapter failure, or missing field mapping, the report must explicitly say `blocked` rather than silently continuing with a fake-complete conclusion.
|
||||
|
||||
---
|
||||
|
||||
## 1. Question router
|
||||
|
||||
Reference in New Issue
Block a user